assignment csv python

pandas.read_csv #

Read a comma-separated values (csv) file into DataFrame.

Also supports optionally iterating or breaking of the file into chunks.

Additional help can be found in the online docs for IO Tools .

Any valid string path is acceptable. The string could be a URL. Valid URL schemes include http, ftp, s3, gs, and file. For file URLs, a host is expected. A local file could be: file://localhost/path/to/table.csv .

If you want to pass in a path object, pandas accepts any os.PathLike .

By file-like object, we refer to objects with a read() method, such as a file handle (e.g. via builtin open function) or StringIO .

Character or regex pattern to treat as the delimiter. If sep=None , the C engine cannot automatically detect the separator, but the Python parsing engine can, meaning the latter will be used and automatically detect the separator from only the first valid row of the file by Python’s builtin sniffer tool, csv.Sniffer . In addition, separators longer than 1 character and different from '\s+' will be interpreted as regular expressions and will also force the use of the Python parsing engine. Note that regex delimiters are prone to ignoring quoted data. Regex example: '\r\t' .

Alias for sep .

Row number(s) containing column labels and marking the start of the data (zero-indexed). Default behavior is to infer the column names: if no names are passed the behavior is identical to header=0 and column names are inferred from the first line of the file, if column names are passed explicitly to names then the behavior is identical to header=None . Explicitly pass header=0 to be able to replace existing names. The header can be a list of integers that specify row locations for a MultiIndex on the columns e.g. [0, 1, 3] . Intervening rows that are not specified will be skipped (e.g. 2 in this example is skipped). Note that this parameter ignores commented lines and empty lines if skip_blank_lines=True , so header=0 denotes the first line of data rather than the first line of the file.

Sequence of column labels to apply. If the file contains a header row, then you should explicitly pass header=0 to override the column names. Duplicates in this list are not allowed.

Column(s) to use as row label(s), denoted either by column labels or column indices. If a sequence of labels or indices is given, MultiIndex will be formed for the row labels.

Note: index_col=False can be used to force pandas to not use the first column as the index, e.g., when you have a malformed file with delimiters at the end of each line.

Subset of columns to select, denoted either by column labels or column indices. If list-like, all elements must either be positional (i.e. integer indices into the document columns) or strings that correspond to column names provided either by the user in names or inferred from the document header row(s). If names are given, the document header row(s) are not taken into account. For example, a valid list-like usecols parameter would be [0, 1, 2] or ['foo', 'bar', 'baz'] . Element order is ignored, so usecols=[0, 1] is the same as [1, 0] . To instantiate a DataFrame from data with element order preserved use pd.read_csv(data, usecols=['foo', 'bar'])[['foo', 'bar']] for columns in ['foo', 'bar'] order or pd.read_csv(data, usecols=['foo', 'bar'])[['bar', 'foo']] for ['bar', 'foo'] order.

If callable, the callable function will be evaluated against the column names, returning names where the callable function evaluates to True . An example of a valid callable argument would be lambda x: x.upper() in ['AAA', 'BBB', 'DDD'] . Using this parameter results in much faster parsing time and lower memory usage.

Data type(s) to apply to either the whole dataset or individual columns. E.g., {'a': np.float64, 'b': np.int32, 'c': 'Int64'} Use str or object together with suitable na_values settings to preserve and not interpret dtype . If converters are specified, they will be applied INSTEAD of dtype conversion.

New in version 1.5.0: Support for defaultdict was added. Specify a defaultdict as input where the default determines the dtype of the columns which are not explicitly listed.

Parser engine to use. The C and pyarrow engines are faster, while the python engine is currently more feature-complete. Multithreading is currently only supported by the pyarrow engine.

New in version 1.4.0: The ‘pyarrow’ engine was added as an experimental engine, and some features are unsupported, or may not work correctly, with this engine.

Functions for converting values in specified columns. Keys can either be column labels or column indices.

Values to consider as True in addition to case-insensitive variants of ‘True’.

Values to consider as False in addition to case-insensitive variants of ‘False’.

Skip spaces after delimiter.

Line numbers to skip (0-indexed) or number of lines to skip ( int ) at the start of the file.

If callable, the callable function will be evaluated against the row indices, returning True if the row should be skipped and False otherwise. An example of a valid callable argument would be lambda x: x in [0, 2] .

Number of lines at bottom of file to skip (Unsupported with engine='c' ).

Number of rows of file to read. Useful for reading pieces of large files.

Additional strings to recognize as NA / NaN . If dict passed, specific per-column NA values. By default the following values are interpreted as NaN : “ “, “#N/A”, “#N/A N/A”, “#NA”, “-1.#IND”, “-1.#QNAN”, “-NaN”, “-nan”, “1.#IND”, “1.#QNAN”, “<NA>”, “N/A”, “NA”, “NULL”, “NaN”, “None”, “n/a”, “nan”, “null “.

Whether or not to include the default NaN values when parsing the data. Depending on whether na_values is passed in, the behavior is as follows:

If keep_default_na is True , and na_values are specified, na_values is appended to the default NaN values used for parsing.

If keep_default_na is True , and na_values are not specified, only the default NaN values are used for parsing.

If keep_default_na is False , and na_values are specified, only the NaN values specified na_values are used for parsing.

If keep_default_na is False , and na_values are not specified, no strings will be parsed as NaN .

Note that if na_filter is passed in as False , the keep_default_na and na_values parameters will be ignored.

Detect missing value markers (empty strings and the value of na_values ). In data without any NA values, passing na_filter=False can improve the performance of reading a large file.

Indicate number of NA values placed in non-numeric columns.

Deprecated since version 2.2.0.

If True , skip over blank lines rather than interpreting as NaN values.

The behavior is as follows:

bool . If True -> try parsing the index. Note: Automatically set to True if date_format or date_parser arguments have been passed.

list of int or names. e.g. If [1, 2, 3] -> try parsing columns 1, 2, 3 each as a separate date column.

list of list . e.g. If [[1, 3]] -> combine columns 1 and 3 and parse as a single date column. Values are joined with a space before parsing.

dict , e.g. {'foo' : [1, 3]} -> parse columns 1, 3 as date and call result ‘foo’. Values are joined with a space before parsing.

If a column or index cannot be represented as an array of datetime , say because of an unparsable value or a mixture of timezones, the column or index will be returned unaltered as an object data type. For non-standard datetime parsing, use to_datetime() after read_csv() .

Note: A fast-path exists for iso8601-formatted dates.

If True and parse_dates is enabled, pandas will attempt to infer the format of the datetime strings in the columns, and if it can be inferred, switch to a faster method of parsing them. In some cases this can increase the parsing speed by 5-10x.

Deprecated since version 2.0.0: A strict version of this argument is now the default, passing it has no effect.

If True and parse_dates specifies combining multiple columns then keep the original columns.

Function to use for converting a sequence of string columns to an array of datetime instances. The default uses dateutil.parser.parser to do the conversion. pandas will try to call date_parser in three different ways, advancing to the next if an exception occurs: 1) Pass one or more arrays (as defined by parse_dates ) as arguments; 2) concatenate (row-wise) the string values from the columns defined by parse_dates into a single array and pass that; and 3) call date_parser once for each row using one or more strings (corresponding to the columns defined by parse_dates ) as arguments.

Deprecated since version 2.0.0: Use date_format instead, or read in as object and then apply to_datetime() as-needed.

Format to use for parsing dates when used in conjunction with parse_dates . The strftime to parse time, e.g. "%d/%m/%Y" . See strftime documentation for more information on choices, though note that "%f" will parse all the way up to nanoseconds. You can also pass:

time string (not necessarily in exactly the same format);

and you should probably use it along with dayfirst .

New in version 2.0.0.

DD/MM format dates, international and European format.

If True , use a cache of unique, converted dates to apply the datetime conversion. May produce significant speed-up when parsing duplicate date strings, especially ones with timezone offsets.

Return TextFileReader object for iteration or getting chunks with get_chunk() .

Number of lines to read from the file per chunk. Passing a value will cause the function to return a TextFileReader object for iteration. See the IO Tools docs for more information on iterator and chunksize .

For on-the-fly decompression of on-disk data. If ‘infer’ and ‘filepath_or_buffer’ is path-like, then detect compression from the following extensions: ‘.gz’, ‘.bz2’, ‘.zip’, ‘.xz’, ‘.zst’, ‘.tar’, ‘.tar.gz’, ‘.tar.xz’ or ‘.tar.bz2’ (otherwise no compression). If using ‘zip’ or ‘tar’, the ZIP file must contain only one data file to be read in. Set to None for no decompression. Can also be a dict with key 'method' set to one of { 'zip' , 'gzip' , 'bz2' , 'zstd' , 'xz' , 'tar' } and other key-value pairs are forwarded to zipfile.ZipFile , gzip.GzipFile , bz2.BZ2File , zstandard.ZstdDecompressor , lzma.LZMAFile or tarfile.TarFile , respectively. As an example, the following could be passed for Zstandard decompression using a custom compression dictionary: compression={'method': 'zstd', 'dict_data': my_compression_dict} .

New in version 1.5.0: Added support for .tar files.

Changed in version 1.4.0: Zstandard support.

Character acting as the thousands separator in numerical values.

Character to recognize as decimal point (e.g., use ‘,’ for European data).

Character used to denote a line break. Only valid with C parser.

Character used to denote the start and end of a quoted item. Quoted items can include the delimiter and it will be ignored.

Control field quoting behavior per csv.QUOTE_* constants. Default is csv.QUOTE_MINIMAL (i.e., 0) which implies that only fields containing special characters are quoted (e.g., characters defined in quotechar , delimiter , or lineterminator .

When quotechar is specified and quoting is not QUOTE_NONE , indicate whether or not to interpret two consecutive quotechar elements INSIDE a field as a single quotechar element.

Character used to escape other characters.

Character indicating that the remainder of line should not be parsed. If found at the beginning of a line, the line will be ignored altogether. This parameter must be a single character. Like empty lines (as long as skip_blank_lines=True ), fully commented lines are ignored by the parameter header but not by skiprows . For example, if comment='#' , parsing #empty\na,b,c\n1,2,3 with header=0 will result in 'a,b,c' being treated as the header.

Encoding to use for UTF when reading/writing (ex. 'utf-8' ). List of Python standard encodings .

How encoding errors are treated. List of possible values .

New in version 1.3.0.

If provided, this parameter will override values (default or not) for the following parameters: delimiter , doublequote , escapechar , skipinitialspace , quotechar , and quoting . If it is necessary to override values, a ParserWarning will be issued. See csv.Dialect documentation for more details.

Specifies what to do upon encountering a bad line (a line with too many fields). Allowed values are :

'error' , raise an Exception when a bad line is encountered.

'warn' , raise a warning when a bad line is encountered and skip that line.

'skip' , skip bad lines without raising or warning when they are encountered.

New in version 1.4.0:

Callable, function with signature (bad_line: list[str]) -> list[str] | None that will process a single bad line. bad_line is a list of strings split by the sep . If the function returns None , the bad line will be ignored. If the function returns a new list of strings with more elements than expected, a ParserWarning will be emitted while dropping extra elements. Only supported when engine='python'

Changed in version 2.2.0:

Callable, function with signature as described in pyarrow documentation when engine='pyarrow'

Specifies whether or not whitespace (e.g. ' ' or '\t' ) will be used as the sep delimiter. Equivalent to setting sep='\s+' . If this option is set to True , nothing should be passed in for the delimiter parameter.

Deprecated since version 2.2.0: Use sep="\s+" instead.

Internally process the file in chunks, resulting in lower memory use while parsing, but possibly mixed type inference. To ensure no mixed types either set False , or specify the type with the dtype parameter. Note that the entire file is read into a single DataFrame regardless, use the chunksize or iterator parameter to return the data in chunks. (Only valid with C parser).

If a filepath is provided for filepath_or_buffer , map the file object directly onto memory and access the data directly from there. Using this option can improve performance because there is no longer any I/O overhead.

Specifies which converter the C engine should use for floating-point values. The options are None or 'high' for the ordinary converter, 'legacy' for the original lower precision pandas converter, and 'round_trip' for the round-trip converter.

Extra options that make sense for a particular storage connection, e.g. host, port, username, password, etc. For HTTP(S) URLs the key-value pairs are forwarded to urllib.request.Request as header options. For other URLs (e.g. starting with “s3://”, and “gcs://”) the key-value pairs are forwarded to fsspec.open . Please see fsspec and urllib for more details, and for more examples on storage options refer here .

Back-end data type applied to the resultant DataFrame (still experimental). Behaviour is as follows:

"numpy_nullable" : returns nullable-dtype-backed DataFrame (default).

"pyarrow" : returns pyarrow-backed nullable ArrowDtype DataFrame.

New in version 2.0.

A comma-separated values (csv) file is returned as two-dimensional data structure with labeled axes.

Write DataFrame to a comma-separated values (csv) file.

Read general delimited file into DataFrame.

Read a table of fixed-width formatted lines into DataFrame.

Learn Python practically and Get Certified .

Popular Tutorials

Learn python interactively, popular examples, python introduction.

• Getting Started With Python
• Python 'Hello World!' Program
• Python Comments
• Python Variables
• Python Data Types
• Python Type Conversion
• Python Print and Input
• Python Operators

Python Flow Control

• Python Booleans
• Python if..else Statement
• Python match…case Statement
• Python for Loop
• Python while Loop
• Python break and continue
• Python pass Statement
• Python Functions
• Python Function Arguments: Positional, Keyword & Default
• Python Global and Local Variables
• Python Lambda (Anonymous) Functions
• Python Recursion

Python Data Structures

• Python Numbers
• Python List
• Python Tuple
• Python String
• Python Dictionary
• Python Range
• Python Object-Oriented-Programming
• Python Classes and Objects
• Python Inheritance

Python Modules

• Python Module
• Python Packages

Python Exceptions and Files

• Python try…expect (Exception Handling)
• Python Files: Open, Read, Write & Delete Files
• Python Directory and OS Module

Python Advanced

• Python Iterators
• Python Generators
• Python Decorators

Other Topics

• Python Datetime
• Python RegEx
• Python JSON
• Python CSV: Read & Write CSV files
• Python List Comprehension

Python CSV: Read &; Write CSV Files

In this tutorial, you will learn about reading and writing CSV files in Python with the help of examples.

CSV is the most popular file format for working with tabular data. It has a .csv file extension, and when we load it to a text file (other than spreadsheets or tables), you will see each element will be separated by a single character (generally commas). For example,

Python provides a built-in module named csv that makes working with csv files a lot easier, and to use it, we must first import it.

After importing, we can perform file operations like Reading and Writing on the csv file.

1. Read CSV File

The csv module provides a reader() function to perform the read operation.

Suppose we have a csv file like this

Now, let's use the reader() function to read this file.

In the above example, we have imported the csv module in our program. We then open the info.csv file using open() . This is similar to opening any file in Python. To learn more, visit Python files .

Here, we have used the csv.reader() function to read the csv file. This function returns an iterable object which we then accessed using a for loop.

The above code works fine if each row of the csv file is separated by commas, which might not be the case all the time. Sometimes, in text editors, csv files are separated by the - symbol or a tab.

In such cases, we have to provide the optional delimiter argument to the reader() function. For example, if we have a csv file where data is separated by tabs.

Let's first see what happens with the older code

Here, the code runs fine, but if you look into the output carefully, you can see the individual columns of the csv files are treated as a single element, 'Name-Age-Country' , 'Maria-21-Italy' , etc.

Now, let's use the delimiter parameter.

This time we get the desired output.

2. Write CSV FIle

Similar to the reader() function, the csv module also provides the writer() function to write data to a file.

Let's see an example,

Here, we have used the writer() function to get a writer object, which we then used to call the writerow() function. The writerow() function writes the provided data to the csv file.

After running the above code, our info.csv file will look something like this.

• Learn Python
• Python Lists
• Python Dictionaries
• Python Strings
• Python Functions
• Learn Pandas & NumPy
• Pandas Tutorials
• Numpy Tutorials
• Learn Data Visualization
• Python Seaborn
• Python Matplotlib

How to Write CSV Files in Python (from list, dict)

• December 19, 2022

This guide will teach you how to write CSV files in Python, including from Python lists and dictionaries . The Python csv library gives you significant flexibility in writing CSV files. For example, you can write Python lists to CSV, including writing headers and using custom delimiters. Likewise, you can write Python dictionaries to CSV and handle non-ASCII characters.

By the end of this guide, you’ll have learned the following:

• How to write CSV files in Python using the csv.writer() class and csv.DictWriter() class
• How to write Python lists and dictionaries to CSV files
• How to handle common complexities, such as double quotes, non-ASCII characters, and escape characters

Table of Contents

Quick Answer: How to Write CSV Files from a List in Python

If you’re in a hurry, you can use the following code to write a CSV file from a Python list:

To write a Python list to a CSV file, you can use the csv library and the csv.writer() class. By passing a file into the class, you create an object that allows you to write single rows and multiple rows from a Python list.

Throughout this tutorial, we’ll explore how this method works, allowing you to customize the behavior. Use the table of contents above to skip to a specific section.

How to Write a Python List to a CSV File

In this section, you’ll learn how to write a Python list to a CSV file. Python comes with a library, csv , as part of its standard library. As the name implies, this library is used to handle CSV files, both in terms of reading and writing them.

One of the helpful classes that allow you to write CSV files from Python lists is the csv.writer() class. Let’s dive into how the class works and how you can use it to write CSV files from Python lists.

Understanding the csv.writer() Class in Python

The csv.writer() class can be used to insert data into a CSV file using Python. By passing a file object into the class, a new writer object is returned. This object converts a user’s input data into a delimited string. This string, by default, is comma-separated.

The class provides a number of helpful methods:

• .writerow() which writes a single row at a time. This can be useful to insert a header row into our resulting CSV file.
• .writerows() is used to write multiple rows at one time. This can be useful to insert data into a file.

Let’s see how we can instantiate a csv.writer() class by passing in an opened file:

Let’s break down what is happening in the code block above:

• We import the csv module
• We then use a context manager to open a file in write mode, 'w' . By doing this, if the file doesn’t exist, Python will create it for you. We use the alias file to use the file.
• We then create a writer object by passing this new file into csv.writer()

Note here that we’re using a context manager. While this is not necessary, it allows you to run your code in a safer manner. This is because the context manager will close the file once it’s done being used. This allows your code to be more memory safe.

Now, let’s dive into how we can write a single row to a CSV file from a Python list.

How to Write a Single Row to a CSV File from a Python List

In order to write a single row to a CSV file, we can use the .writerow() method associated with a csvwriter() object. If we pass in a Python list of lists into the method, Python will write a single list at a time. Let’s see what this looks like:

To write a single row at a time to a CSV in Python, you can follow the steps below :

• Import the csv module
• Create a single or multiple list variables
• Open a CSV file in write mode
• Instantiate a csv.writer() object by passing in the file as its argument
• Use the .writerow() method and pass in a single list. This will write the list to a single row in a CSV file.

This generates the following CSV file:

This process can be time consuming if you want to write multiple lines at a single time. Because of this, Python provides the .writerows() method. Let’s explore that in the following section.

How to Write Multiple Rows to a CSV File from a Python List

In order to write multiple rows to a CSV from a Python list of lists, you can use the .writelines() method found in the csv.writer() class.

Similar to our example above, we can create a CSV writer object. However, rather than writing a single line at a time, we can pass in multiple lines at once. Let’s see what this looks like in Python:

To write a multiple rows to a CSV in Python from a list of lists, you can follow the steps below :

• Create a list of lists that contains each row as a separate list
• Use the .writerows() method and pass in the list of lists. This will write multiple rows at a single time in a CSV file.

The code block above returns the following CSV file:

We can see how much simpler this made our code! Now, let’s explore how to add a header to our CSV file from a Python list of lists.

How to Add a Header to a CSV File from a Python List

Python makes it simple to add a header row to our CSV file. This can be done by combining the two methods you have learned so far. In many cases, it doesn’t make sense to store your header row as a list in your data’s list of lists. Instead, the header is often maintained as a separate list or tuple.

Let’s see how we can add a header to a CSV file from a Python list:

In the code block above, we:

• First wrote the header row using the .writerow() method.
• Then, we wrote used the .writerows() method to write the remaining rows of data from our list of lists.

This returns the following CSV file, which includes the header row:

In the following section, you’ll learn how to append a list to a CSV file in Python.

How to Append a List to a CSV File in Python

With each of the previous sections of the tutorial, the file that you wrote to was overwritten. This is something important to be aware of – even if the file exists, all data will be overwritten.

If we want to append a list to a CSV file in Python, we can simply modify the mode in which Python opens the file. So far, we have used a write mode of 'w' . By modifying the mode to append, using 'a' , we can append to our CSV file, rather than overwriting it.

Let’s see what this looks like in Python:

When we use our previously created CSV file, this returns the following CSV file:

In the following sections, you’ll learn how to write a Python dictionary to a CSV file.

How to Write a Python Dictionary to a CSV File with DictWriter

Similar to the csv.writer() object, the csv module provides a class to write dictionaries to CSV files. This class is called the csv.DictWriter() class and provides similar methods for writing data to CSV files.

To write a Python dictionary to a CSV file, you can use the csv library and the csv.DictWriter() class. By passing a file into the class, you create an object that allows you to write single rows and multiple rows from a Python dictionary or list of dictionaries.

What’s different between this approach and the csv.writer() class is that the class assumes that we provide the field names of the CSV file. Let’s take a look at a basic implementation of the class:

It’s important to note that the instantiation will fail if we do not provide the fieldnames= argument. Python will raise a TypeError and instruct you to pass in a fieldnames argument.

Let’s dive into how we can use the csv.DictWriter() class to write single rows and multiple rows to a CSV file.

How to Write a Single Row to a CSV File from a Python Dictionary

In order to write a single dictionary to a CSV file in Python, you can use the csv.DictWriter() class. Unlike the csv.writer() class that we used to write lists to CSV files, the csv.DictWriter() class requires that we pass in the field names.

Why is that? Because Python dictionaries haven’t always been ordered, the order in which fields appear may differ. Similarly, Python dictionaries don’t let you access items by their position (unlike lists). Because of this, we need to find a way to tell Python the order in which to insert values .

Let’s take a look at how we can write a dictionary to a CSV file using the .writerow() method:

In order to write a dictionary to a CSV file in Python, follow the steps below:

• Declare a list of strings that contain the header row
• Create a dictionary containing the headers as keys. The order in the dictionary doesn’t matter.
• Create a CSV file and pass it into the csv.DictWriter() class to instantiate a writer object. Pass in the headers into the fieldnames= parameter.
• Finally, we used the .writerow() method to write a dictionary as a CSV row. We can repeat this multiple times with individual dicationaries.

This returns the following CSV file:

In the following section, you’ll learn how to write multiple dictionaries to a CSV file directly.

How to Write Multiple Rows to a CSV File from a Python Dictionary

In order to write multiple dictionaries to a CSV file with Python, you can use the .writerows() method of the csv.DictWriter() class. This allows you to pass in a list of Python dictionaries.

Being able to do this can be incredibly helpful when working with JSON data. Python provides a number of ways to convert JSON files to CSV .

Let’s take a look at how we can write a list of dictionaries to a CSV file:

In this example, we used the .writerows() method to write a list of dictionaries to a CSV file. This allows you to add as much data as you want, without needing to call the .writerow() method for each dictionary.

In the following section, you’ll learn how to add a header row to a CSV file from a list of dictionaries.

How to Add a Header to a CSV File from a Python Dictionary

It’s very easy to add a header row to a CSV file that’s created using Python dictionaries. Because we already pass in the field names when we create our writer object, we can simply use this information to create a header.

In fact, the csv.DictWriter() class comes with a method for just this purpose: .writeheader() . Let’s take a look at how we can write a header row for our CSV file:

The code block above is unchanged from the previous example except for the addition of the second last row. By calling the .writeheader() method, Python adds a header row. It uses the values passed into the fieldnames= parameter to create our header.

We can see that Python was able to successfully create a CSV file from our list of dictionaries, including inserting a header row.

How to Append a Dictionary to a CSV File in Python

In the following sections of this tutorial, you’ll learn how to customize the resulting CSV files using special attributes, such as custom delimeters and quoted characters.

Additional Options in the CSV DictWriter Class

The csv.DictWriter() class provides a number of different options to customize how dictionaries are written to CSV files. In particular, the class provides two additional parameters:

• restval= specifies a value to write when a key is missing from a dictionary
• extrasaction= indicates what action to take when a key is included that doesn’t exist in the field names parameter. By default, the argument is set to 'raise' which will raise a TypeError if an extra field is included. You can also set this to 'ignore' .

How to Use Custom Delimiters when Writing to CSV in Python

We can customize the delimiter that is used to split columns in a CSV file. The delimeter= parameter is available for both the csv.writer() class and the csv.DictWriter() class. We can pass in a string that represents the delimiter we want to use.

Let’s see how we can customize the delimiter when using Python to write a list or a dictionary to a CSV file:

In the code above, we passed a pipe operator into the delimiter= argument. Keep in mind that this also works for the csv.writer() class. This returns the following CSV.

In the following section, you’ll learn how to create and use dialects that allow you to easily customize the way in which CSV files are written.

How to Use Dialects when Writing to CSV in Python

Dialects in the CSV module allow you to set customization criteria to easily write and read files in the same style. While we have only covered custom delimiters so far, the CSV library provides significant flexibility to customize CSV files. Because of this, dialects allow you to set preset styles that can be used to read and write CSV files in Python .

Let’s see how we can create a dialect and use it when writing a CSV file:

In the example above we created a dialect using the csv.register_dialect() function. The function allows you pass in a name using a string. Then, it can subclass either existing dialects or add custom formatting as we have done.

This allows you to easily re-use the dialect when you are writing other files or even reading CSV files. Because the function isn’t tied to a particular writer class, it works with both the csv.writer() class and the csv.DictWriter() class.

How to Deal with Quotes when Writing to CSV in Python

By default, Python will wrap individual entries in double quotes in order to ensure that items are properly delimited. Imagine the case where you list items have commas in them, Python needs a way to ensure that items are delimited appropriately.

Similarly, when we write items with quotes in it themselves, the csv module will escape quotes by adding double quotes. Let’s see what this looks like:

We can see that the character that contains a " in itself is escaped using the double quotes.

There are two ways in which this behavior can be customized:

• quotechar= customizes whether the quote character should be displayed. By default, it’s set to True . When set to False , the escape character is used. When we set it to False , we need to pass in an escapechar= .
• escapechar= allows you to customize the escape character. This is necessary when the quotechar= character is set to False .

Let’s see how we can disable the functionality of double quoting and use a custom escape character:

In the following section, you’ll learn how to write non-ASCII characters to a CSV file in Python.

How to Write UTF-8 Encoding to a CSV in Python

In order to write CSV files that are encoded by UTF-8, you can simply declare the encoding when you open the file. Because this operation is done before creating a writer object, this process is the same for creating CSV files based on lists or dictionaries.

Let’s see how we can do this in Python:

In the example above, we used the same code as in previous examples. However, when we opened the file, we passed in encoding='utf-8' , which ensures that the file is opened (or in this case, created) with the correct encoding.

In this guide, you learned how to use Python to write to CSV files using both lists and dictionaries. The Python csv module provides significant flexibility in terms of writing to CSV files. By providing writer objects for both lists and dictionaries, the module lets you customize the output of your CSV file.

You first learned how to write lists to CSV files, including writing a single row, multiple rows, and adding headers. Then, you covered the same topics with Python dictionaries, as well as working with the additional options that the DictWriter class provides. Finally, you learned how to customize behavior by exploring different options such as modifying encoding, changing delimiters and double quoting, as well as creating dialects for easier reusability.

Additional Resources

To learn more about related topics, check out the resources below:

• File Handling in Python: A Complete Guide
• Python: Check if a File or Directory Exists
• How to Read a Text File in Python (Python open)
• How to Use Python to Write a Text File (.txt)
• Pandas Dataframe to CSV File – Export Using .to_csv()
• Python csv Module: Official Documentation

Nik Piepenbreier

Nik is the author of datagy.io and has over a decade of experience working with data analytics, data science, and Python. He specializes in teaching developers how to use Python for data science using hands-on tutorials. View Author posts

Leave a Reply Cancel reply

Your email address will not be published. Required fields are marked *

Save my name, email, and website in this browser for the next time I comment.

pandas read csv() Tutorial: Importing Data

pandas is a widely-used Python library for data science, analysis, and machine learning that offers a flexible and intuitive way to handle data sets of all sizes. One of the most important functionalities of pandas is the tools it provides for reading and writing data. For data available in a tabular format and stored as a CSV file, you can use pandas to read it into memory using the read_csv() function, which returns a pandas dataframe. But there are other functionalities too. For example, you can use pandas to perform merging, reshaping, joining, and concatenation operations. 

In this article, you will learn about the read_csv() function and how you can alter the parameters to customize the output you receive once the function is executed. We will also cover the different methods available to a pandas dataframe object, including how to write pandas dataframe to a CSV file and how to quickly learn more about your data through various methods and attributes.  

Practice pandas functions with hands-on exercises from our Intermediate Python course.

Note : Check out this DataLab workbook  to follow along with the code. 

Importing a CSV file using the read_csv() function

Before reading a CSV file into a pandas dataframe , you should have some insight into what the data contains. Thus, it’s recommended you skim the file before attempting to load it into memory: this will give you more insight into what columns are required and which ones can be discarded.

Let’s write some code to import a file using read_csv() . Then we can talk about what’s going on and how we can customize the output we receive while reading the data into memory.

All that has gone on in the code above is we have:

• Imported the pandas library into our environment
• Passed the filepath to read_csv to read the data into memory as a pandas dataframe.
• Printed the first five rows of the dataframe.

But there’s a lot more to the read_csv() function.

Learn Python From Scratch

Setting a column as the index.

The default behavior of pandas is to add an initial index to the dataframe returned from the CSV file it has loaded into memory. However, you can explicitly specify what column to make as the index to the read_csv function by setting the index_co l parameter.

Note the value you assign to index_col may be given as either a string name, column index or a sequence of string names or column indexes. Assigning the parameter a sequence will result in a multiIndex (a grouping of data by multiple levels).

Let’s read in the data again and set the id column as the index.

Selecting specific columns to read into memory

What if you only want to read specific columns into memory because not all of them are important? This is a common scenario that occurs in the real world. Using the read_csv function, you can select only the columns you need after loading the file, but this means you must know what columns you need prior to loading in the data if you wish to perform this operation from within the read_csv function.

If you do know the columns you need, you’re in luck; you can save time and memory by passing a list-like object to the usecols parameter of the read_csv function.

We have barely scratched the surface of different ways to customize the output of the read_csv function, but going into more depth would certainly be an information overload.

We recommend you bookmark the importing data in Python cheat sheet and check out Introduction to importing data in Python to learn more. If that’s a little too easy, there is also the intermediate importing data in Python interactive course.

Reading Data from a URL

Once you know how to read a CSV file from local storage into memory, reading data from other sources is a breeze. It’s ultimately the same process, except that you’re no longer passing a file path.

Let’s say there’s data you want from a specific webpage; how would you read it into memory?

We will use the Iris dataset from the UCI repository as an example:

You may have noticed we assigned a list of strings to the names parameter in the read_csv function. This is just so we can rename the column headers while reading the data into memory.

Methods and Attributes of the DataFrame Structure

The most common object in the pandas library is, by far, the dataframe object. It’s a 2-dimensional labeled data structure consisting of rows and columns that may be of different data types (i.e., float, numeric, categorical, etc.).

Conceptually, you can think of a pandas dataframe like a spreadsheet, SQL table, or a dictionary of series objects – whichever you’re more familiar with. The cool thing about the pandas dataframe is that it comes with many methods that make it easy for you to become acquainted with your data as quickly as possible.

You have already seen one of those methods: iris_data.head() , which shows the first n (the default is 5) rows. The “opposite” method of head() is tail(), which shows the last n (5 by default) rows of the dataframe object. For example:

You can quickly discover the column names by using the columns attribute on your dataframe object:

Another important method you can use on your dataframe object is info() . This method prints out a concise summary of the dataframe, including information about the index, data types, columns, non-null values, and memory usage.

DataFrame.describe() generates descriptive statistics, including those that summarize the central tendency, dispersion, and shape of the dataset’s distribution. If your data has missing values, don’t worry; they are not included in the descriptive statistics.

Let’s call the describe method on the Iris dataset:

Exporting the DataFrame to a CSV File

Another method available to pandas dataframe objects is to_csv() . When you have cleaned and preprocessed your data, the next step may be to export the dataframe to a file – this is pretty straightforward:

Executing this code will create a CSV in the current working directory called cleaned_iris_data.csv .

But what if you want to use a different delimiter to mark the beginning and end of a unit of data or you wanted to specify how your missing values should be represented? Maybe you don’t want the headers to be exported to the file.

Well, you can adjust the parameters of the to_csv() method to suit your requirements for the data you want to export.

Let’s take a look at a few examples of how you can adjust the output of to_csv() :

• Export data to the current working directory but using a tab delimiter.
• Exporting data without the index
• Change the name of missing values (the default is ““)
• Export dataframe to file without headers (column names)

Final thoughts

Let’s recap what we covered in this tutorial; you learned how to:

• Import a CSV file using the read_csv() function from the pandas library.
• Set a column index while reading your data into memory.
• Specify the columns in your data that you want the read_csv() function to return.
• Read data from a URL with the pandas.read_csv()
• Quickly gather insights about your data using methods and attributes on your dataframe object.
• Export a dataframe object to a CSV file
• Customize the output of the export file from the to_csv() method.

In this tutorial, we focused solely on importing and exporting data from the perspective of a CSV file; you now have a good sense of how useful pandas is when importing and exporting CSV files. CSV is one of the most common data storage formats, but it’s not the only one. There are various other file formats used in data science, such as parquet, JSON, and excel.

Plenty of useful, high-quality datasets are hosted on the web, which you can access through APIs, for example. If you want to understand how to handle loading data into Python in more detail, DataCamp's Introduction to Importing Data in Python course will teach you all the best practices.

There are also tutorials on how to import JSON and HTML data into pandas and a beginner-friendly ultimate guide to pandas tutorial . Be sure to check those out to dive deeper into the pandas framework.

Learn more about Python and pandas

.css-1531qan{-webkit-text-decoration:none;text-decoration:none;color:inherit;} Pandas Joins for Spreadsheet Users

Data manipulation with pandas, joining data with pandas.

cheat-sheet

Pandas Cheat Sheet for Data Science in Python

Karlijn Willems

Importing Data in Python Cheat Sheet

Data preparation with pandas.

Sejal Jaiswal

Pandas Tutorial: DataFrames in Python

How to import json and html data into pandas.

Aditya Sharma

Python pandas tutorial: The ultimate guide for beginners

Vidhi Chugh

How To Read ‘CSV’ File In Python

Our day-to-day encounter with the computer system requires a lot of data and information exchange. To ensure that this exchange is smooth and easy, users prefer text files, and one of the most common formats for these text files is '.csv .'

But what is a CSV file? How can we use it to extract and alter data? What is a  python library?  Are there other available libraries to process this format? What are the commands and programs in each library?

Don't worry! All these questions are answered here in this article. You'll learn how to use the csv files and process them properly using code to extract the most out of your data. Let's get started.

What Is A CSV File?

The image shows how a CSV file looks. Each line of the file represents a row of the tabular data. You can observe that commas separate the fields.

Why are CSV files created?

The programs which handle considerable amounts of data create CSV files to export data from spreadsheets or databases conveniently and import them into different programs. From a programmer's point of view, CSV files are straightforward to work with. Any programming language that accepts text file input and supports string data manipulation can work with CSV files directly.

Reading CSV Files With Python's Built-in CSV Library

The built-in CSV library of Python contains functions to read from as well as write to CSV files. This library supports various csv formats and includes objects and code functions for processing CSV files.

Reading Files With csv

Object used to read from a CSV file : reader object  

Function used to open the CSV file  : open()

The built-in open() function of Python opens the CSV file as a text file. This function provides a file object that is then passed to the reader object, which further processes the file.

Let us take an example of a CSV file, storing the details of students as its data.

  Filename: students.txt

The output is as follows:

Each row returned by the reader object is an array of string elements containing data after the delimiters have been removed.

Reading Files Into a Dictionary With csv  

We use the same file as above.

The code to read it as a dictionary is as follows:

Notice the difference in the code here. When we use the dictionary, the row name is written in the row{} statement rather than the row number(as done in the previous method).

The result is the same output as before:

The first line of the csv file contains the keys to build the dictionary. If, for instance, the keys are not present on your csv file, then you can specify your own keys.

Optional Parameters of reader object:

The reader object can work with different styles of CSV files with the help of additional parameters. Some of these parameters are discussed here:

• delimiter parameter specifies the character used to separate each field in a record. The default is the comma (',').
• quotechar defines the character used to represent fields containing the delimiter character. The default value of this parameter is a double quote (' " ').
• escapechar describes the character used to escape the delimiter character in case the user doesn't use quotes. The default is no escape character.

For example, in the ‘students.txt’ file discussed above, we want to add another field as 'date.'

The format of date itself contains 'commas,' and the use of 'commas' as the delimiter in this situation will create a lot of confusion and make the processing cumbersome. To make it more convenient to use, we can specify another character as the delimiter. Alternatively, you can put your data in quotes by using the quotechar parameter, as any data placed in quoted strings ignore the function of delimiter. If you want to nullify the interpretation of delimiter completely, then you can use the escapechar parameter.

Writing CSV Files With csv

The object used: writer object.

Method used :  .write_row()

The quotechar optional parameter is used to define which character is being used to quote fields when writing. The quoting parameter can take the following values:

• If the value of the quoting parameter equals to csv.QUOTE_MINIMAL , then the method  .writerow() will quote fields only if they contain the delimiter or the quotechar. This setting is the default case.
• If the value of the quoting parameter equals to csv.QUOTE_ALL, then the method .writerow()   quotes all the fields present in the data. 
• If the value of the quoting parameter equals to csv.QUOTE_NONNUMERIC , then the method .writerow() will quote all fields containing text data, and all numeric fields are converted to the float data type.
• If the value of quoting parameter equals to csv.QUOTE_NONE , then the method .writerow() will escape delimiters instead of quoting them.

Writing Files From A Dictionary With csv

Notice that in the writerow() field, we have to mention fieldnames as well.

Processing CSV Files With Pandas Library

pandas is another library available for processing CSV files. This library is recommended if the user has a considerable amount of data to analyze. It is an open-source Python library available for all python installations. One of the most preferred python installations for the usage of pandas library is the Anaconda distribution equipped with Jupyter notebook. It can process, share and analyze data with the help of many available tools and data structures.

Installing pandas and its dependencies in Anaconda is done as shown below:

If you are using pip/pipenv for python installations, then the command is as follows:

With the installation of the pandas library complete, let us now learn how to read and write csv files in this library.

Reading CSV Files with pandas.

Reading a csv file in the pandas library is very easy compared to other libraries because of its compact code structure.

df here is short for 'DataFrame.' The pandas.read_csv() method opens and analyses the CSV file provided and saves the data in a DataFrame. Printing the DataFrame gives the desired output.

We can see the difference in output when we use the methods as mentioned earlier. Also, the result obtained using the pandas library has the index starting from '0' rather than '1'.

Writing CSV Files with pandas

The code for writing data into a CSV file is as shown below:

Filename: students.csv

Here the ‘Name’ column of the file has been made the index column. Moreover, the print(df) command changes to df.to_csv() in the writing mode.

CSV files have tremendous use in the current data and information exchange scenario. Hence, if you understand their structure, composition, and usage, then you are ready to rule the programming world. 

The application of CSV files to assess data makes the operation increasingly efficient. When incorporated with programming languages like Python, the efficiency of these files increases manifolds in data manipulation, assessment and application. There are other libraries available as well for the processing of CSV files in Python. Still, the most efficient and compact ones have been discussed in this article for the programmer to have the most efficient program.

• Python Tips and Tricks
• Python Library Tutorials
• Python Tools
• Python How To's
• Python Tutorials

Related Articles

• Introduction to SQLite in Python
• Cutting and Slicing Strings in Python
• How to Open A File in Python
• Advanced SQLite Usage in Python
• How To Install Pandas In Python? An Easy Step By Step Multimedia Guide

Signup for new content

Thank you for joining our mailing list!

Latest Articles

• Discover How to Get FL Studio for Free and Start Creating Music Today
• Discover the Ultimate Features of Free FL Studio and How to Maximize Its Potential
• Looks Matter: How UI Design and Python Code Quality Impact User Satisfaction
• Free PHP, HTML, Python, CSS, JavaScript editor - Codelobster IDE
• Everything You Need to Know About Software Development Consulting
• Python salaries
• pyenv global
• NumPy arrays
• insertion sort
• in place reversal
• learn python
• python packages
• zeros() function
• Scikit Learn
• HTML Parser
• circular queue
• effiiciency
• python maps
• Num Py Zeros
• Python Lists
• HTML Extraction
• selection sort
• Programming
• install python on windows
• reverse string
• python Code Editors
• pandas.reset_index
• Infinite Numbers in Python
• Python Readlines()
• Programming language
• remove python
• concatenate string
• Code Editors
• reset_index()
• Train Test Split
• Local Testing Server
• Python Input
• priority queue
• web development
• uninstall python
• python string
• code interface
• round numbers
• train_test_split()
• Flask module
• Linked List
• machine learning
• compare string
• pandas dataframes
• arange() method
• Singly Linked List
• python scripts
• learning python
• python bugs
• ZipFunction
• plus equals
• np.linspace
• SQLAlchemy advance
• Data Structure
• csv in python
• logging in python
• Python Counter
• python subprocess
• numpy module
• Python code generators
• python tutorial
• csv file python
• python logging
• Counter class
• Python assert
• numbers_list
• binary search
• Insert Node
• Python tips
• python dictionary
• Python's Built-in CSV Library
• logging APIs
• Constructing Counters
• Matplotlib Plotting
• any() Function
• linear search
• Python tools
• python update
• logging module
• Concatenate Data Frames
• python comments
• Recursion Limit
• Data structures
• installation
• python function
• pandas installation
• Zen of Python
• concatenation
• Echo Client
• NumPy Pad()
• install python
• how to install pandas
• Philosophy of Programming
• concat() function
• Socket State
• Python YAML
• remove a node
• function scope
• Tuple in Python
• pandas groupby
• socket programming
• Python Modulo
• Dictionary Update()
• datastructure
• bubble sort
• find a node
• calling function
• GroupBy method
• Np.Arange()
• Modulo Operator
• Python Or Operator

CSE 163, Spring 2020: Homework 2: Processing CSV Data

In this assignment, you will use the foundational Python skills you've been developing and apply them to analyze a small dataset. Many datasets you’ll be working with are structured as CSV or tabular representation - this assignment will be an introduction to reading, processing, and grouping rows and columns to calculate some interesting statistics. These skills will be very useful to have a strong foundation in when we work with much larger (and less complete) real-world datasets!

This assignment is broken in to two main parts, where each part mostly does the same computations in different ways. This is to give you the opportunity to compare/contrast different approaches to solving problems.

Learning Objectives

• Writing a Python script from scratch and turning in the assignment.
• Use the course infrastructure ( flake8 , test suites, course resources).
• Use/manipulation of various data types including numbers and strings.
• Control structures (conditional statements, loops, parameters, returns) to implement basic functions provided by a specification.
• Basic text file processing.
• Documenting code.
• Write unit tests to test each function written including their edge cases.
• Work with data structures (lists, sets, dictionaries) in Python
• Handle edge cases appropriately, including addressing missing values/data
• Practice user-friendly error-handling
• Apply programming to identify and investigate a question on a dataset using basic statistical concepts (e.g. mean, max)

Expectations

Here are some baseline expectations we expect you to meet:

Follow the course collaboration policies

• Do not round any outputs or return values for this assignment!

If you are developing on Ed, all the files are there. If you are developing locally, you should download the starter code hw2.zip and open it as the project in Visual Studio Code. The files included are:

• hw2_manual.py : The file for you to put solutions to Part 0 .
• hw2_pandas.py : The file for you to put solutions to Part 1 .
• hw2_test.py : The file for you to put your tests for Part 0 and Part 1 .
• cse163_utils.py : A file where we will store utility functions for helping you write tests.
• run_hw2.py : A client program provided to call your functions. This is just for your convenience.
• pokemon_box.csv : A CSV file that stores information about Pokemon. This columns of this file are explained below.
• pokemon_test.csv : A very small CSV file that stores information about Pokemon. This columns of this file are explained below.

For this assignment, you will be working with a dataset of Pokemon that you have caught on your Pokemon journey so far. The file pokemon_box.csv stores all the data about the captured Pokemon and has a format that looks like:

Note that because this is a CSV file, the file contents have these cells separated by commas.

Column Descriptions

• id: Unique identification number corresponding to the species of a Pokemon. Note that if there are multiple Pokemon of the same species in the dataset, they all share the id.
• name: Name of the species of Pokemon. For example Pikachu.
• level: The level of this Pokemon (an integer)
• personality: A one-word string describing the personality of this Pokemon
• type: A one-word string describing the type of the Pokemon (e.g. "Grass" for Bulbasaur)
• weakness: What type this Pokemon is weak to. For example, Bulbasaur is considered weak to the fire type.
• atk, def, hp: Pokemon stats that indicate how many hits a Pokemon can take (hp), how strong its attacks are (atk), and how much hits affect it (def)

In this part of the homework, you will write code to perform various analytical operations on data parsed from a file into the list of dictionaries represenation.

For this step of the assignment, you will be implementing various functions to answer questions about the dataset.

Each function should take the list returned by the cse163_utils.parse function (provided for you) as the first argument, along with any other arguments specified in each problem. For example, for the third function, we would call filter_range(data, 1, 10) where data was the list returned by cse163_utils.parse .

This data structure should not be modified by any function you write. Every problem that deals with strings should be case-sensitive (this means "chArIzard" is a different species than "Charizard"). You may make the following assumptions about the inputs:

• You may assume the given list is non-empty for all functions you implement.
• For each problem, you may assume we pass parameters of the expected types described for that problem and that those parameters are not None .
• You should make no other assumptions about the parameters or the data.

For each of the problems, we will use the file pokemon_test.csv to show what should be returned.

Part 0 Expectations

• All functions for this part of the assignment should be written in hw2_manual.py
• For this part of the assignment, you may import the math module, but you may not use any other imports to solve these problems.

Problem 1: species_count

Write a function species_count that returns the number of unique Pokemon species (determined by the name attribute) found in the dataset. You may assume that the data is well formatted in the sense that you don't have to transform any values in the name column.

For example, assuming we have parsed pokemon_test.csv and stored it in a variable called data :

Problem 2: max_level

Write a function max_level that finds the Pokemon with the max level and returns a tuple of length 2, where the first element is the name of the Pokemon and the second is its level . If there is a tie, the Pokemon that appears earlier in the file should be returned.

Problem 3: filter_range

Write a function called filter_range that takes as arguments a smallest (inclusive) and largest (exclusive) level value and returns a list of Pokemon names having a level within that range. The list should return the species names in the same order that they appear in the provided list of dictionaries.

Problem 4: mean_attack_for_type

Write a function called mean_attack_for_type that takes a Pokemon type (string) as an argument and that returns the average attack stat for all the Pokemon in the dataset with that type.

If there are no Pokemon of the given type, this function should return None .

Problem 5: count_types

Write a function called count_types that returns a dictionary with keys that are Pokemon types and values that are the number of times that type appears in the dataset.

The order of the keys in the returned dictionary does not matter. In terms of efficiency, your solution should NOT iterate over the whole dataset once for each type of Pokemon since that would be overly inefficient.

Problem 6: highest_stage_per_type

Write a function called highest_stage_per_type that calculates the largest stage reached for each type of Pokemon in the dataset. This function should return a dictionary that has keys that are the Pokemon types and values that are the highest value of stage column for that type of Pokemon.

Problem 7: mean_attack_per_type

Write a function called mean_attack_per_type that calculates the average attack for every type of Pokemon in the dataset. This function should return a dictionary that has keys that are the Pokemon types and values that are the average attack for that Pokemon type.

In this part of the homework, you will write code to do various analytical operations on CSV data. Sounds a bit like Déjà vu! The functions you write for this assignment are exactly the same, but instead you will be using the pandas library to solve the functions.

For this step of the assignment, you will be implementing various functions to answer questions about the dataset. All of the problems are the same as Part 0 save for the fact that the input is now a DataFrame ; you can check back to that page to see any examples you might need for this part.

Each function should take the DataFrame returned by the pd.read_csv function as the first argument, along with any other arguments specified in each problem. For example, for the third function, we would call filter_range(data, 1, 10) where data was the DataFrame returned by pd.read_csv .

• You may assume the DataFrame is non-empty for all functions you implement.

Hint: pandas wordbank

Because this is the first assignment where you are using pandas , we will provide a "wordbank" of pandas functions/features that you might want to use on these problems. A brief version this list is shown below and you can find more background information about what these items mean by reading this Jupyter Notebook . At the top-right of this page in Ed is a "Fork" button (looks like a fork in the road). This will make your own copy of this Notebook so you can run the code and experiment with anything there!

Not every entry in the wordbank will be necessarily used and you will probably use certain functions/features listed multiple times for this assignment.

• Get a column of a DataFrame
• Get a row of a DataFrame ( loc )
• Loop over Series

Development Strategy

When using data science libraries like pandas , it's extremely helpful to actually interact with the tools your using so you can have a better idea about the shape of your data. The preferred practice by people in industry is to use a Jupyter Notebook, like we have been in lecture, to play around with the dataset to help figure out how to answer the questions you want to answer. This is incredibly helpful when you're first learning a tool as you can actually experiment and get real-time feedback if the code you wrote does what you want.

We recommend that you try figuring out how to solve these problems in a Jupyter Notebook so you can actually interact with the data. We have made a playground Jupyter Notebook for you to use that already has the data loaded. You will want to press the "Fork" button at the top-right (looks like a fork in the road) so you can get your own copy that you may edit.

About Returns

Some of the functions below ask you to return a Python list or dict to keep it symmetric with Part 0. This will be difficult to do if you are working with pandas objects and are asked to not use any loops! You will want to use the following fact: you can use Python's casting to convert a Series into either a list or a dict .

For example, suppose I had the following CSV represented in a pandas DataFrame named data :

This is not any sort of magic! For list , it just uses the values in the Series . For dict , it uses the index as keys and the values as values.

Part 1 Expectations

• All functions for this part of the assignment should be written in hw2_pandas.py
• For this part of the assignment, you may import the math and pandas modules, but you may not use any other imports to solve these problems.
• For all of the problems below, you should not use ANY loops or list/dictionary comprehensions. You should be able to solve all of the problems by only calling functions on pandas objects. The goal of this part of the assignment is to use pandas as a tool to help answer questions about your dataset.

Note that you will want to return a Python list for this function so you will have to convert from a pandas object to a list.

The order of the keys in the dictionary does not matter.

Note that you will want to return a Python dictionary for this function so you will have to convert from a pandas object to a dictionary.

The order of the keys in the returned dictionary does not matter.

In this part of the assignment, you will write tests for your solutions in Part 0 and Part 1 .

Like in Homework 1, we have provided a function called assert_equals that takes an expected value and the value returned by your function, and compares them: if they don't match, the function will crash the program and tell you what was wrong. You can see more instructions an example for tests from the Homework 1 - Part 1 to see examples of how to call the tests.

About paths on Ed

Recall that on HW1, you had to use this absolute path (e.g. /home/poem.txt ) on Ed in your testing program. This has to do with how Ed runs your program when marking it (it only copies your Python files in to its testing directory). You will need to continue to do this for future assignments.

What this means is in your hw2_test.py , any place you specify a file name (e.g. poem.txt ), you should use the absolute path on Ed instead (e.g. /home/poem.txt )

Part 2 Expectations

For full credit, your hw2_test.py must satisfy all of the following conditions:

• You should not use the file pokemon_box.csv to count for your own test cases . The file is too large to be able to meaningfully come up with the correct answer on your own (e.g., it's not valid to run your code and then paste the output as the "correct output"). You should submit your own testing CSV files.
• Uses the main method pattern shown in class.
• Your tests should be efficient in the sense that you don't have to re-read the datasets from the CSV for every tests function. If you want to use the same dataset in multiple functions, read it in once in main and pass them around as parameters to your test functions.
• Has a function to test each of the functions in the hw2_manual.py and hw2_pandas.py that you were asked to write. It's okay to merge the test functions for the same problem in the different parts of the homework since their outputs are the same. You should not organize your tests such that there are only two test functions (e.g., test_hw2_manual and test_hw2_pandas ) as we want to separate the test functions by each of the problems in the spec. There might be some redundancy in your tests which is expected since it's hard to factor out the different function calls and input types.
• For each function you need to test, you should include a test that tests the example from the spec and one additional test that is different from what is shown in the spec. A single test is considered a call on assert_equals .
• Each of these test functions should have a descriptive name that indicates which function is being tested (e.g. test_funky_sum )
• Each of the test functions must be called from main.
• Turn in any test CSV files you generate.

Your submission will be evaluated on the following dimensions:

• Your solution correctly implements the described behaviors. You will have access to some tests when you turn in your assignment, but we will withhold other tests to test your solution when grading. All behavior we test is completely described by the problem specification or shown in an example.
• When we run your code, it should produce no errors or warnings.
• You should remove any debug print statements. Some students report Ed crashing when they try to print the whole dataset; remove those extra print statements to prevent excessive output.
• All files submitted pass flake8
• Your program should be written with good programming style. This means you should use the proper naming convention for methods ( snake_case ), your code should not be overly redundant and should avoid unnecessary computations.
• Every function written is commented using a doc-string format that describes its behavior, parameters, returns, and highlights any special cases.
• There is a comment at the top of each file you write with your name, section, and a brief description of what that program does.
• Any expectations on this page or the sub-pages for the assignment are met as well as all requirements for each of the problems are met.

A note on advanced material

A lot of students have been asking questions like "Can I use this method or can I use this language feature in this class?". The general answer to this question is it depends on what you want to use, what the problem is asking you to do and if there are any restrictions that problem places on your solution.

There is no automatic deduction for using some advanced feature or using material that we have not covered in class yet, but if it violates the restrictions of the assignment, it is possible you will lose points. It's not possible for us to list out every possible thing you can't use on the assignment, but we can say for sure that you are safe to use anything we have covered in class so far as long as it meets what the specification asks and you are appropriately using it as we showed in class.

For example, some things that are probably okay to use even though we didn't cover them:

• Using the update method on the set class even though I didn't show it in lecture. It was clear we talked about sets and that you are allowed to use them on future assignments and if you found a method on them that does what you need, it's probably fine as long as it isn't violating some explicit restriction on that assignment.
• Using something like a ternary operator in Python. This doesn't make a problem any easier, it's just syntax.

For example, some things that are probably not okay to use:

• Importing some random library that can solve the problem we ask you to solve in one line.
• If the problem says "don't use a loop" to solve it, it would not be appropriate to use some advanced programming concept like recursion to "get around" that restriction.

These are not allowed because they might make the problem trivially easy or violate what the learning objective of the problem is.

You should think about what the spec is asking you to do and as long as you are meeting those requirements, we will award credit. If you are concerned that an advanced feature you want to use falls in that second category above and might cost you points, then you should just not use it! These problems are designed to be solvable with the material we have learned so far so it's entirely not necessary to go look up a bunch of advanced material to solve them.

tl;dr; We will not be answering every question of "Can I use X" or "Will I lose points if I use Y" because the general answer is "You are not forbidden from using anything as long as it meets the spec requirements. If you're unsure if it violates a spec restriction, don't use it and just stick to what we learned before the assignment was released."

This assignment is due by Thursday, April 23 at 23:59 (PDT) . Please refer to the late policy for the policy on how late you may turn an assignment in.

You should submit your finished hw2_manual.py , hw2_pandas.py , and hw2_test.py on Ed .

You may submit your assignment as many times as you want before the late cutoff (remember submitting after the due date will cost late days). Recall on Ed, you submit by pressing the "Mark" button. You are welcome to develop the assignment on Ed or develop locally and then upload to Ed before marking.

Join us and get access to thousands of tutorials and a community of expert Pythonistas.

This lesson is for members only. Join us and get access to thousands of tutorials and a community of expert Pythonistas.

Writing CSVs With Python's "csv" Module

• Discussion (14)

In this lesson, you’ll learn how to write data to CSV files using the two csv writer methods csv.writer and csv.DictWriter . The lesson also covers how to quote specific fields using constants such as csv.QUOTE_MINIMAL , csv.QUOTE_NONNUMERIC , csv.QUOTE_NONE , and csv.QUOTE_ALL .

00:00 Now it’s time to start writing CSV files. Just like reading, you can write CSV files line by line, or by using a dictionary. Open up a new Python script, and let’s get started by importing csv .

00:14 Also like before, you’ll need to use a with open() statement to open that file, which you can call something like 'employee_file.csv' .

00:25 Make sure you set the mode equal to write, and just call that something like employee_file . Next, create an employee_writer object, which you can set equal to a csv.writer() , pass in the employee_file , set the delimiter equal to a standard comma, a quotechar ,

00:50 which can just be a single double quotes. And unlike reading the file, you’ll need to set this parameter called quoting , which—for now—just put QUOTE_MINIMAL .

01:04 I’ll move this out of the way so that we can see. And now with that employee_writer , you can just call something like .writerow() and then pass in a list of the items that you’d like to add.

01:18 So, 'John Smith' from 'Accounting' ,

01:24 and he was born in 'November' . And then employee_writer.writerow() , and 'Erica Meyers' from 'IT' who was born in 'March' .

01:41 Okay. Save this, open up a terminal, and run it. All right. Didn’t see any errors. Let’s go to this directory. And here you can see a CSV was created using the delimiter you specified and the quoting , which in this case was minimal— so, QUOTE_MINIMAL just means that it’ll add quotes if the field contains the delimiter or the quotechar .

02:09 Let’s take a look at some of the other options. You can also do QUOTE_ALL instead of QUOTE_MINIMAL , which will add quotes to everything, no matter what.

02:18 So, let’s actually save this as a new file, run that, and here you can see every field now is wrapped in those quotes. The third option is to QUOTE_NONNUMERIC , which—just like it sounds—will go and quote any fields that don’t contain numeric data.

02:38 And then finally, you can use something called QUOTE_NONE , which won’t add any quotes and will just escape any characters that it needs to. And if you’re going to use this, you need to add in an escapechar , which we’ll set equal—just like we had before—to the pipe.

02:54 So, let’s save this as a new file, and just to make sure it works, let’s put some commas into the names here. Go ahead, run that, open up the new file, and you can see that there are no quotes and it used the escape character before the commas that we added to the data fields.

03:13 So, let’s go close this terminal out, go back to this, and let’s see how to write a CSV file using a dictionary. I’m just going to change the filename here to something like that, and now, you need to add in a fieldnames variable, so this’ll just be something like ['name', 'department'] — actually, just to keep things short—and 'birth_month' . And kind of like before, instead of just the regular writer , this will now be a DictWriter (dictionary writer).

03:50 You can get rid of all of this information here and just pass in fieldnames=fieldnames — which I can tell from my autofill, I forgot an s up here, so let’s fix that right now.

04:04 And what’s a little different now is before you actually start writing, you need to call the .writeheader() method, and that’ll add that first row that’ll contain the header information.

04:16 And for these, instead of passing in lists, they’ll now be dictionaries, so 'name' equals 'John Smith' , 'dept' is 'Accounting' , and 'birth_month' is 'November' .

04:41 Let’s just go ahead and copy this row,

04:46 and we’ll fill out Erica’s data down here. So, 'Erica Meyers' ,

04:55 'IT' , and 'March' . Alrighty! Let’s try to run this.

05:03 Cool! Open up the file, and now you can see that we have the same information down here, but we also have that header information up top. And that’s it! Now you can write CSV files using the csv library in Python!

05:16 The last thing we’ll cover is reading and writing CSV files, but using the pandas library instead. You could have entire courses on pandas , but we’re just going to focus on using it to get CSVs into it and out of it.

05:28 Thanks for watching.

Robert on July 1, 2019

What is the IDE or editor that you are using?

Joe Tatusko RP Team on July 5, 2019

Hi Robert! I’m using Atom and platformio-ide-terminal for the integrated terminal at the bottom in these videos.

Robert on July 9, 2019

Thank you, I’ll check those out

Dave Wilson on Feb. 23, 2020

Such a great tutorial, Joe - as always!

I’m not sure why, but my csv’s printed rows on every other line. I was able to fix it by adding a lineterminator to the employee_writer definition: employee_writer = csv.DictWriter(employee_file, fieldnames=fieldnames, lineterminator = '\n')

hayatmor on March 15, 2020

In python3 in order to work without double space you shall add:

Source: Stackoverflow

sweir12525 on Sept. 24, 2020

Hi, I got a reply to a prior question regarding why I am getting a syntax error when I put in pip install pandas. The RP Team member, Bartosz, answered telling me what a csv file was and that any text editor can open them. Problem is that I know exactly what a CSV file is, I am using PYcharm and Thorny and on both applications, when I enter the script in this lesson, I get a syntax error. I have installed Anaconda and when I put in the entire script I get attribution error. When I debug, it says I should try calling something other than pandas. So, please explain to me what I am doing wrong since all I am doing is following the lesson plan.

Ricky White RP Team on Sept. 25, 2020

Hi @sweir12525. It might be useful in this case if you copy the bottom part of the Traceback error. Just above where it says Syntax error, it’ll show you which line of code is causing the problem. It would be helpful if we could see that so we can help you further.

sweir12525 on Sept. 29, 2020

Thanks, Ricky. I did manage to finally figure this out after a ton of reading in Anaconda.

karmajna on Sept. 29, 2020

Added newline variable to open function to not print on every other line as other mentioned.

Kim on Nov. 4, 2021

I received an error when trying to write a csv file with dictionary. Here is my code

Here is the error I received:

What do I need to do to fix? Thanks!

Bartosz Zaczyński RP Team on Nov. 5, 2021

@Kim Unlike the regular CSV writer, the dict writer’s .writerow() method expects a dictionary and not a list:

Kim on Nov. 7, 2021

OK. Thank you! I’ll use what you have above.

Realblogger on Sept. 13, 2023

Hello, I receive error : AttributeError: ‘module’ object has no attribute ‘writer’

Any suggestions.

walterrieppi on Feb. 5, 2024

When i try to repeat the example where i write a csv file with the quoting parameter set to QUOTE_ALL i have this situation

employee_writer.writerow([‘John Smith’,’Accounting’,’November’]) employee_writer.writerow([‘Erica Meyers’,’IT’,’March’])

In my new file created by the program i have these row

John Smith,”Accounting”,”November” Erica Meyers,”IT”,”March”

Why the first element is written without the ” ” ????? thanks in advance to all

Become a Member to join the conversation.

Learn Python practically and Get Certified .

Popular Tutorials

Popular examples, reference materials, learn python interactively, python introduction.

• Get Started With Python
• Your First Python Program
• Python Comments

Python Fundamentals

• Python Variables and Literals
• Python Type Conversion
• Python Basic Input and Output
• Python Operators

Python Flow Control

• Python if...else Statement
• Python for Loop
• Python while Loop
• Python break and continue
• Python pass Statement

Python Data types

• Python Numbers and Mathematics
• Python List
• Python Tuple
• Python String
• Python Dictionary
• Python Functions
• Python Function Arguments
• Python Variable Scope
• Python Global Keyword
• Python Recursion
• Python Modules
• Python Package
• Python Main function

Python Files

• Python Directory and Files Management
• Python CSV: Read and Write CSV files

Reading CSV files in Python

Writing CSV files in Python

• Python Exception Handling
• Python Exceptions
• Python Custom Exceptions

Python Object & Class

• Python Objects and Classes
• Python Inheritance
• Python Multiple Inheritance
• Polymorphism in Python
• Python Operator Overloading

Python Advanced Topics

• List comprehension
• Python Lambda/Anonymous Function
• Python Iterators
• Python Generators
• Python Namespace and Scope
• Python Closures
• Python Decorators
• Python @property decorator
• Python RegEx

Python Date and Time

• Python datetime
• Python strftime()
• Python strptime()
• How to get current date and time in Python?
• Python Get Current Time
• Python timestamp to datetime and vice-versa
• Python time Module
• Python sleep()

Additional Topic

• Precedence and Associativity of Operators in Python
• Python Keywords and Identifiers
• Python Asserts
• Python Json
• Python *args and **kwargs

Python Tutorials

Working with CSV files in Python

Python CSV: Read and Write CSV Files

Python open()

• Python JSON
• Python File Operation

We are going to exclusively use the csv module built into Python for this task. But first, we will have to import the module as :

We have already covered the basics of how to use the csv module to read and write into CSV files. If you don't have any idea on using the csv module, check out our tutorial on Python CSV: Read and Write CSV files

• Basic Usage of csv.reader()

Let's look at a basic example of using csv.reader() to refresh your existing knowledge.

Example 1: Read CSV files with csv.reader()

Suppose we have a CSV file with the following entries:

We can read the contents of the file with the following program:

Here, we have opened the innovators.csv file in reading mode using open() function.

To learn more about opening files in Python, visit: Python File Input/Output

Then, the csv.reader() is used to read the file, which returns an iterable reader object.

The reader object is then iterated using a for loop to print the contents of each row.

Now, we will look at CSV files with different formats. We will then learn how to customize the csv.reader() function to read them.

• CSV files with Custom Delimiters

By default, a comma is used as a delimiter in a CSV file. However, some CSV files can use delimiters other than a comma. Few popular ones are | and \t .

Suppose the innovators.csv file in Example 1 was using tab as a delimiter. To read the file, we can pass an additional delimiter parameter to the csv.reader() function.

Let's take an example.

Example 2: Read CSV file Having Tab Delimiter

As we can see, the optional parameter delimiter = '\t' helps specify the reader object that the CSV file we are reading from, has tabs as a delimiter.

• CSV files with initial spaces

Some CSV files can have a space character after a delimiter. When we use the default csv.reader() function to read these CSV files, we will get spaces in the output as well.

To remove these initial spaces, we need to pass an additional parameter called skipinitialspace . Let us look at an example:

Example 3: Read CSV files with initial spaces

Suppose we have a CSV file called people.csv with the following content:

We can read the CSV file as follows:

The program is similar to other examples but has an additional skipinitialspace parameter which is set to True.

This allows the reader object to know that the entries have initial whitespace. As a result, the initial spaces that were present after a delimiter is removed.

• CSV files with quotes

Some CSV files can have quotes around each or some of the entries.

Let's take quotes.csv as an example, with the following entries:

Using csv.reader() in minimal mode will result in output with the quotation marks.

In order to remove them, we will have to use another optional parameter called quoting .

Let's look at an example of how to read the above program.

Example 4: Read CSV files with quotes

As you can see, we have passed csv.QUOTE_ALL to the quoting parameter. It is a constant defined by the csv module.

csv.QUOTE_ALL specifies the reader object that all the values in the CSV file are present inside quotation marks.

There are 3 other predefined constants you can pass to the quoting parameter:

• csv.QUOTE_MINIMAL - Specifies reader object that CSV file has quotes around those entries which contain special characters such as delimiter , quotechar or any of the characters in lineterminator .
• csv.QUOTE_NONNUMERIC - Specifies the reader object that the CSV file has quotes around the non-numeric entries.
• csv.QUOTE_NONE - Specifies the reader object that none of the entries have quotes around them.
• Dialects in CSV module

Notice in Example 4 that we have passed multiple parameters ( quoting and skipinitialspace ) to the csv.reader() function.

This practice is acceptable when dealing with one or two files. But it will make the code more redundant and ugly once we start working with multiple CSV files with similar formats.

As a solution to this, the csv module offers dialect as an optional parameter.

Dialect helps in grouping together many specific formatting patterns like delimiter , skipinitialspace , quoting , escapechar into a single dialect name.

It can then be passed as a parameter to multiple writer or reader instances.

Example 5: Read CSV files using dialect

Suppose we have a CSV file ( office.csv ) with the following content:

The CSV file has initial spaces, quotes around each entry, and uses a | delimiter.

Instead of passing three individual formatting patterns, let's look at how to use dialects to read this file.

From this example, we can see that the csv.register_dialect() function is used to define a custom dialect. It has the following syntax:

The custom dialect requires a name in the form of a string. Other specifications can be done either by passing a sub-class of Dialect class, or by individual formatting patterns as shown in the example.

While creating the reader object, we pass dialect='myDialect' to specify that the reader instance must use that particular dialect.

The advantage of using dialect is that it makes the program more modular. Notice that we can reuse 'myDialect' to open other files without having to re-specify the CSV format.

• Read CSV files with csv.DictReader()

The objects of a csv.DictReader() class can be used to read a CSV file as a dictionary.

Example 6: Python csv.DictReader()

Suppose we have a CSV file ( people.csv ) with the following entries:

Let's see how csv.DictReader() can be used.

As we can see, the entries of the first row are the dictionary keys. And, the entries in the other rows are the dictionary values.

Here, csv_file is a csv.DictReader() object. The object can be iterated over using a for loop. The csv.DictReader() returned an OrderedDict type for each row. That's why we used dict() to convert each row to a dictionary.

Notice that we have explicitly used the dict() method to create dictionaries inside the for loop.

Note : Starting from Python 3.8, csv.DictReader() returns a dictionary for each row, and we do not need to use dict() explicitly.

The full syntax of the csv.DictReader() class is:

To learn more about it in detail, visit: Python csv.DictReader() class

• Using csv.Sniffer class

The Sniffer class is used to deduce the format of a CSV file.

The Sniffer class offers two methods:

• sniff(sample, delimiters=None) - This function analyses a given sample of the CSV text and returns a Dialect subclass that contains all the parameters deduced.

An optional delimiters parameter can be passed as a string containing possible valid delimiter characters.

• has_header(sample) - This function returns True or False based on analyzing whether the sample CSV has the first row as column headers.

Let's look at an example of using these functions:

Example 7: Using csv.Sniffer() to deduce the dialect of CSV files

Let's look at how we can deduce the format of this file using csv.Sniffer() class:

As you can see, we read only 64 characters of office.csv and stored it in the sample variable.

This sample was then passed as a parameter to the Sniffer().has_header() function. It deduced that the first row must have column headers. Thus, it returned True which was then printed out.

Similarly, sample was also passed to the Sniffer().sniff() function. It returned all the deduced parameters as a Dialect subclass which was then stored in the deduced_dialect variable.

Later, we re-opened the CSV file and passed the deduced_dialect variable as a parameter to csv.reader() .

It was correctly able to predict delimiter , quoting and skipinitialspace parameters in the office.csv file without us explicitly mentioning them.

Note: The csv module can also be used for other file extensions (like: .txt ) as long as their contents are in proper structure.

Recommended Reading: Write to CSV Files in Python

Table of Contents

Sorry about that.

Our premium learning platform, created with over a decade of experience and thousands of feedbacks .

Learn and improve your coding skills like never before.

• Interactive Courses
• Certificates
• 2000+ Challenges

Related Tutorials

Python Tutorial

Python Library

• Stack Overflow for Teams Where developers & technologists share private knowledge with coworkers
• Advertising & Talent Reach devs & technologists worldwide about your product, service or employer brand
• OverflowAI GenAI features for Teams
• OverflowAPI Train & fine-tune LLMs
• Labs The future of collective knowledge sharing
• About the company Visit the blog

Collectives™ on Stack Overflow

Find centralized, trusted content and collaborate around the technologies you use most.

Q&A for work

Connect and share knowledge within a single location that is structured and easy to search.

Get early access and see previews of new features.

Python Assigning CSV File to a Variable

My task was to import a CSV File and make it into an array, which I have done. Now my task now is to assign the data to a variable- is this possible? For instance, how would I go about accessing the 2nd column to the array by typing in myVariableName[1] ?

• you might want to take a look at np.loadtxt or pandas.read_csv –  maxymoo Commented Jul 21, 2015 at 3:54
• 1 Hint, you've already assigned other things to variable names. e.g. f = open("Test.csv") assigns the file object that is returned by open to the variable f . You did it again with csv_f . From here, it shouldn't be too hard to figure out how to assign a variable name to any particular object in the code ... –  mgilson Commented Jul 21, 2015 at 3:54
• I've tried to assign a variable however is it possible to include the "for" statement as well? –  aaa905 Commented Jul 21, 2015 at 3:55

2 Answers 2

You can use numpy.loadtxt() .

Create a list to which items from column 2 can be appended as you process each row from the CSV file. The following code will contain items from column 2 in the variable column_2 :

You could replace all of that code with a list comprehension:

Your Answer

Reminder: Answers generated by artificial intelligence tools are not allowed on Stack Overflow. Learn more

Sign up or log in

Post as a guest.

Required, but never shown

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy .

Not the answer you're looking for? Browse other questions tagged python arrays csv or ask your own question .

• The Overflow Blog
• The hidden cost of speed
• The creator of Jenkins discusses CI/CD and balancing business with open source
• Featured on Meta
• Announcing a change to the data-dump process
• Bringing clarity to status tag usage on meta sites
• What does a new user need in a homepage experience on Stack Overflow?
• Feedback requested: How do you use tag hover descriptions for curating and do...
• Staging Ground Reviewer Motivation

Hot Network Questions

• Could a lawyer agree not to take any further cases against a company?
• How long should a wooden construct burn (and continue to take damage) until it burns out (and stops doing damage)
• What do these expressions mean in NASA's Steve Stitch's brief Starliner undocking statement?
• Why does the guardian who admits guilt pay more than when his guilt is established by witnesses?
• Pólya trees counted efficiently
• Fundamental Question on Noise Figure
• How cheap would rocket fuel have to be to make Mars colonization feasible (according to Musk)?
• Do US universities invite faculty applicants from outside the US for an interview?
• What is this phenomenon?
• How high does the ocean tide rise every 90 minutes due to the gravitational pull of the space station?
• How does a miner create multiple outputs in a coinbase transaction?
• Information theoretical interpretation of Free Energy
• Multiplicity of the smallest non-zero Laplacian eigenvalue for tree graphs
• Topos notions coming from topology and uniqueness of generalizations
• What is the Kronecker product of two 1D vectors?
• Best approach to make lasagna fill pan
• Setting the desired kernel in GRUB menu
• Size of the functor category
• What is the missing fifth of the missing fifths?
• Beatles reference in parody story from the 1980s
• Is there a way to read lawyers arguments in various trials?
• How to raise and lower indices as a physicist would handle it?
• Where is this railroad track as seen in Rocky II during the training montage?
• What are the estimate dates for destruction of planets by the Sun in the distant future?

• modülleri |
• Python »
• 3.14.0a0 Documentation »
• What’s New in Python »
• Theme Auto Light Dark |

Changelog ¶

Python next ¶.

Release date: XXXX-XX-XX

gh-123418 : Updated macOS installer build to use OpenSSL 3.0.15.

gh-123418 : Updated Windows build to use OpenSSL 3.0.15.

gh-123476 : Add support for socket.TCP_QUICKACK on Windows platforms.

gh-122573 : The Windows build of CPython now requires 3.10 or newer.

gh-100256 : mimetypes no longer fails when it encounters an inaccessible registry key.

gh-119679 : Ensures correct import libraries are included in Windows installs.

gh-119690 : Adds Unicode support and fixes audit events for _winapi.CreateNamedPipe .

gh-111201 : Add support for new pyrepl on Windows

gh-119070 : Fixes py.exe handling of shebangs like /usr/bin/env python3.12 , which were previously interpreted as python3.exe instead of python3.12.exe .

gh-117505 : Fixes an issue with the Windows installer not running ensurepip in a fully isolated environment. This could cause unexpected interactions with the user site-packages.

gh-118209 : Avoid crashing in mmap on Windows when the mapped memory is inaccessible due to file system errors or access violations.

gh-116145 : Updated bundled Tcl/Tk to 8.6.14.

gh-79846 : Makes ssl.create_default_context() ignore invalid certificates in the Windows certificate store

Tools/Demos ¶

gh-123418 : Update GitHub CI workflows to use OpenSSL 3.0.15 and multissltests to use 3.0.15, 3.1.7, and 3.2.3.

gh-121921 : Update Lib/test/crashers/bogus_code_obj.py so that it crashes properly again.

gh-112301 : Add tooling to check for changes in compiler warnings. Patch by Nate Ohlson.

gh-59022 : Add tests for pkgutil.extend_path() . Patch by Andreas Stocker.

gh-99242 : os.getloadavg() may throw OSError when running regression tests under certain conditions (e.g. chroot). This error is now caught and ignored, since reporting load average is optional.

gh-121084 : Fix test_typing random leaks. Clear typing ABC caches when running tests for refleaks ( -R option): call _abc_caches_clear() on typing abstract classes and their subclasses. Patch by Victor Stinner.

gh-121160 : Add a test for readline.set_history_length() . Note that this test may fail on readline libraries.

gh-121200 : Fix test_expanduser_pwd2() of test_posixpath . Call getpwnam() to get pw_dir , since it can be different than getpwall() pw_dir . Patch by Victor Stinner.

gh-121188 : When creating the JUnit XML file, regrtest now escapes characters which are invalid in XML, such as the chr(27) control character used in ANSI escape sequences. Patch by Victor Stinner.

gh-120801 : Cleaned up fixtures for importlib.metadata tests and consolidated behavior with ‘test.support.os_helper’.

gh-119727 : Add --single-process command line option to Python test runner (regrtest). Patch by Victor Stinner.

gh-119273 : Python test runner no longer runs tests using TTY (ex: test_ioctl) in a process group (using setsid() ). Previously, tests using TTY were skipped. Patch by Victor Stinner.

gh-119050 : regrtest test runner: Add XML support to the refleak checker (-R option). Patch by Victor Stinner.

gh-101525 : Skip test_gdb if the binary is relocated by BOLT. Patch by Donghee Na.

gh-123678 : Upgrade libexpat to 2.6.3

gh-112301 : Add ability to ignore warnings per file with warning count in warning checking tooling. Patch by Nate Ohlson.

gh-112301 : Add macOS warning tracking to warning check tooling. Patch by Nate Ohlson.

gh-122133 : Authenticate the socket connection for the socket.socketpair() fallback on platforms where AF_UNIX is not available like Windows.

Patch by Gregory P. Smith < greg @ krypto . org > and Seth Larson < seth @ python . org >. Reported by Ellie < el @ horse64 . org >

gh-121957 : Fixed missing audit events around interactive use of Python, now also properly firing for python -i , as well as for python -m asyncio . The events in question are cpython.run_stdin and cpython.run_startup .

gh-112301 : Enable runtime protections for glibc to abort execution when unsafe behavior is encountered, for all platforms except Windows.

gh-121285 : Remove backtracking from tarfile header parsing for hdrcharset , PAX, and GNU sparse headers.

gh-112301 : Add default compiler options to improve security. Enable -Wimplicit-fallthrough, -fstack-protector-strong, -Wtrampolines.

gh-118773 : Fixes creation of ACLs in os.mkdir() on Windows to work correctly on non-English machines.

gh-118486 : os.mkdir() on Windows now accepts mode of 0o700 to restrict the new directory to the current user. This fixes CVE-2024-4030 affecting tempfile.mkdtemp() in scenarios where the base temporary directory is more permissive than the default.

gh-123374 : Remove check for redefined memo entry in pickletools.dis() .

gh-123504 : Fixed reference leak in the finalization of tkinter .

gh-123430 : Pages generated by the http.server module allow the browser to apply its default dark mode.

gh-123446 : Fix empty function name in TypeError when csv.reader() , csv.writer() , or csv.register_dialect() are used without the required args.

gh-123448 : Fixed memory leak of typing.NoDefault by moving it to the static types array.

gh-123409 : Fix ipaddress.IPv6Address.reverse_pointer output according to RFC 3596, §2.5 . Patch by Bénédikt Tran.

gh-123089 : Make weakref.WeakSet safe against concurrent mutations while it is being iterated. Patch by Kumar Aditya.

gh-123363 : Show string value of CONTAINS_OP oparg in dis output. Patch by Alexandr153.

gh-123341 : Add __class_getitem__() to tkinter.Event for type subscript support at runtime. Patch by Adonis Rakateli.

gh-123340 : Show string value of IS_OP oparg in dis output.

gh-119518 : Speed up normalization of pathlib.PurePath and Path objects by not interning string parts.

gh-123270 : Applied a more surgical fix for malformed payloads in zipfile.Path causing infinite loops ( gh-122905 ) without breaking contents using legitimate characters.

gh-73991 : Add pathlib.Path.copy_into() and move_into() , which copy and move files and directories into existing directories.

gh-123228 : Fix return type for _pyrepl.readline._ReadlineWrapper.get_line_buffer() to be str() . Patch by Sergey B Kirpichev.

gh-123240 : Raise audit events for the input() in the new REPL.

gh-76960 : Fix urllib.parse.urljoin() and urllib.parse.urldefrag() for URIs containing empty components. For example, urljoin() with relative reference “?” now sets empty query and removes fragment. Preserve empty components (authority, params, query, fragment) in urljoin() . Preserve empty components (authority, params, query) in urldefrag() .

gh-123243 : Fix memory leak in _decimal .

gh-122546 : Consistently use same file name for different exceptions in the new repl. Patch by Sergey B Kirpichev.

gh-123213 : xml.etree.ElementTree.Element.extend() and Element assignment no longer hide the internal exception if an erronous generator is passed. Patch by Bar Harel.

gh-85110 : Preserve relative path in URL without netloc in urllib.parse.urlunsplit() and urllib.parse.urlunparse() .

gh-123165 : Add support for rendering Positions in dis .

gh-122909 : In urllib.request when URLError is raised opening an ftp URL, the exception argument is now consistently a string. Earlier versions passed either a string or an ftplib exception instance as the argument to URLError.

gh-123084 : Deprecate shutil.ExecError , which hasn’t been raised by any shutil function since Python 3.4. It’s now an alias for RuntimeError .

gh-123067 : Fix quadratic complexity in parsing " -quoted cookie values with backslashes by http.cookies .

gh-123049 : Add support for UNNAMED_SECTION in configparser.ConfigParser.add_section() .

gh-122981 : Fix inspect.getsource() for generated classes with Python base classes (e.g. enums).

gh-122903 : zipfile.Path.glob now correctly matches directories instead of silently omitting them.

gh-122905 : zipfile.Path objects now sanitize names from the zipfile.

gh-122873 : Enable json module to work as a script using the -m switch: python -m json . See the JSON command-line interface documentation. Patch by Trey Hunner.

gh-122858 : Deprecate asyncio.iscoroutinefunction() in favor of inspect.iscoroutinefunction() .

gh-116263 : logging.handlers.RotatingFileHandler no longer rolls over empty log files.

gh-105376 : Restore the deprecated logging warn() method. It was removed in Python 3.13 alpha 1. Keep the deprecated warn() method in Python 3.13. Patch by Victor Stinner.

gh-122744 : Bump the version of pip bundled in ensurepip to version 24.2.

gh-118761 : Improve import time of pprint by around seven times. Patch by Hugo van Kemenade.

gh-118814 : Fix the typing.TypeVar constructor when name is passed by keyword.

gh-122637 : Adjust cmath.tanh(nanj) and cmath.tanh(infj) for recent C standards.

gh-122478 : Remove internal frames from tracebacks shown in code.InteractiveInterpreter with non-default sys.excepthook() . Save correct tracebacks in sys.last_traceback and update __traceback__ attribute of sys.last_value and sys.last_exc .

gh-116622 : On Android, the FICLONE and FICLONERANGE constants are no longer exposed by fcntl , as these ioctls are blocked by SELinux.

gh-82378 : Make sure that the new REPL interprets sys.tracebacklimit in the same way that the classic REPL did.

gh-122334 : Fix crash when importing ssl after the main interpreter restarts.

gh-122459 : Optimize pickling by name objects without the __module__ attribute.

gh-87320 : In code.InteractiveInterpreter , handle exceptions caused by calling a non-default sys.excepthook() . Before, the exception bubbled up to the caller, ending the REPL .

gh-122272 : On some platforms such as Linux, year with century was not 0-padded when formatted by strftime() with C99-specific specifiers '%C' or '%F' . The 0-padding behavior is now guaranteed when the format specifiers '%C' and '%F' are supported by the C library. Patch by Ben Hsing

gh-122400 : Handle ValueError s raised by os.stat() in filecmp.dircmp and filecmp.cmpfiles() . Patch by Bénédikt Tran.

gh-122311 : Fix some error messages in pickle .

gh-121650 : email headers with embedded newlines are now quoted on output. The generator will now refuse to serialize (write) headers that are unsafely folded or delimited; see verify_generated_headers . (Contributed by Bas Bloemsaat and Petr Viktorin in gh-121650 .)

gh-122332 : Fixed segfault with asyncio.Task.get_coro() when using an eager task factory.

gh-105733 : ctypes.ARRAY() is now soft deprecated : it no longer emits deprecation warnings and is not scheduled for removal.

gh-119180 : As part of PEP 749 , add the following attributes for customizing evaluation of annotation scopes:

evaluate_value on typing.TypeAliasType

evaluate_bound , evaluate_constraints , and evaluate_default on typing.TypeVar

evaluate_default on typing.ParamSpec

evaluate_default on typing.TypeVarTuple

gh-119180 : Fix handling of classes with custom metaclasses in annotationlib.get_annotations .

gh-122170 : Handle ValueError s raised by os.stat() in linecache . Patch by Bénédikt Tran.

gh-122163 : Add notes for JSON serialization errors that allow to identify the source of the error.

gh-122129 : Improve support of method descriptors and wrappers in the help title.

gh-121723 : Make logging.config.dictConfig() accept any object implementing the Queue public API. See the queue configuration section for details. Patch by Bénédikt Tran.

gh-82951 : Serializing objects with complex __qualname__ (such as unbound methods and nested classes) by name no longer involves serializing parent objects by value in pickle protocols < 4.

gh-120754 : Pathlib.read_bytes no longer opens the file in Python’s buffered I/O mode. This reduces overheads as the code reads a file in whole leading to a modest speedup.

gh-113785 : csv now correctly parses numeric fields (when used with csv.QUOTE_NONNUMERIC or csv.QUOTE_STRINGS ) which start with an escape character.

gh-122088 : @warnings.deprecated now copies the coroutine status of functions and methods so that inspect.iscoroutinefunction() returns the correct result.

gh-122081 : Fix a crash in the decimal.IEEEContext() optional function available via the EXTRA_FUNCTIONALITY configuration flag.

gh-73991 : Add pathlib.Path.move() , which moves a file or directory tree.

gh-121268 : Remove workarounds for non-IEEE 754 systems in cmath .

gh-119698 : Due to the lack of interest for symtable.Class.get_methods() , the method is marked as deprecated and will be removed in Python 3.16. Patch by Bénédikt Tran.

gh-121889 : Adjusts cmath.acosh(complex('0+nanj')) for recent C standards.

gh-121804 : Correctly show error locations, when SyntaxError raised in new repl. Patch by Sergey B Kirpichev.

gh-120930 : Fixed a bug introduced by gh-92081 that added an incorrect extra blank to encoded words occurring in wrapped headers.

gh-57141 : The shallow argument to filecmp.dircmp (new in Python 3.13) is now keyword-only.

gh-121245 : Simplify handling of the history file in site.register_readline() helper. The CAN_USE_PYREPL variable now will be initialized, when imported. Patch by Sergey B Kirpichev.

gh-121486 : math functions isqrt() , log() , log2() and log10() now support integers larger than 2**2**32 on 32-bit platforms.

gh-121474 : Fix missing sanity check for parties arg in threading.Barrier constructor. Patch by Clinton Christian (pygeek).

gh-121450 : Hard-coded breakpoints ( breakpoint() and pdb.set_trace() ) now reuse the most recent Pdb instance that calls Pdb.set_trace() , instead of creating a new one each time. As a result, all the instance specific data like display and commands are preserved across Hard-coded breakpoints.

gh-119169 : Slightly speed up os.walk() by simplifying exception handling.

gh-121423 : Improve import time of socket by lazy importing modules and writing socket.errorTab as a constant.

gh-59110 : zipimport supports now namespace packages when no directory entry exists.

gh-121313 : Limit the reading size in the multiprocessing.connection.Connection class to 64 KiB to prevent memory overallocation and unnecessary memory management system calls.

gh-121332 : Fix constructor of ast nodes with custom _attributes . Previously, passing custom attributes would raise a DeprecationWarning . Passing arguments to the constructor that are not in _fields or _attributes remains deprecated. Patch by Jelle Zijlstra.

gh-121245 : Fix a bug in the handling of the command history of the new REPL that caused the history file to be wiped at REPL exit.

gh-121210 : Handle AST nodes with missing runtime fields or attributes in ast.compare() . Patch by Bénédikt Tran.

gh-121163 : Add support for all as an valid action for warnings.simplefilter() and warnings.filterwarnings() .

gh-121151 : Fix wrapping of long usage text of arguments inside a mutually exclusive group in argparse .

gh-121141 : Add support for copy.replace() to AST nodes. Patch by Bénédikt Tran.

gh-87744 : Fix waitpid race while calling send_signal() in asyncio. Patch by Kumar Aditya.

gh-121027 : Add a future warning in functools.partial.__get__() . In future Python versions functools.partial will be a method descriptor.

gh-121027 : Make the functools.partial object a method descriptor.

gh-117784 : CPython now detects whether its linked TLS library supports TLSv1.3 post-handshake authentication and disables that feature if support is lacking.

gh-121025 : Improve the __repr__() of functools.partialmethod . Patch by Bénédikt Tran.

gh-121018 : Fixed issues where argparse.ArgumentParser.parse_args() did not honor exit_on_error=False . Based on patch by Ben Hsing.

gh-119614 : Fix truncation of strings with embedded null characters in some internal operations in tkinter .

gh-120910 : When reading installed files from an egg, use relative_to(walk_up=True) to honor files installed outside of the installation root.

gh-61103 : Support float complex , double complex and long double complex C types in ctypes as c_float_complex , c_double_complex and c_longdouble_complex if the compiler has C11 complex arithmetic. Patch by Sergey B Kirpichev.

gh-120888 : Upgrade pip wheel bundled with ensurepip (pip 24.1.1)

gh-101830 : Accessing the tkinter object’s string representation no longer converts the underlying Tcl object to a string on Windows.

gh-120678 : Fix regression in the new REPL that meant that globals from files passed using the -i argument would not be included in the REPL’s global namespace. Patch by Alex Waygood.

gh-120811 : Fix possible memory leak in contextvars.Context.run() .

gh-120782 : Fix wrong references of the datetime types after reloading the module.

gh-120713 : datetime.datetime.strftime() now 0-pads years with less than four digits for the format specifiers %Y and %G on Linux. Patch by Ben Hsing

gh-120769 : Make empty line in pdb repeats the last command even when the command is from cmdqueue .

gh-120780 : Show string value of LOAD_SPECIAL oparg in dis output.

gh-120743 : Soft deprecate os.popen() and os.spawn* functions. They should no longer be used to write new code. The subprocess module is recommended instead. Patch by Victor Stinner.

gh-120732 : Fix name passing to unittest.mock.Mock object when using unittest.mock.create_autospec() .

gh-111259 : re now handles patterns like "[\s\S]" or "\s|\S" which match any character as effectively as a dot with the DOTALL modifier ( "(?s:.)" ).

gh-120683 : Fix an error in logging.LogRecord , when the integer part of the timestamp is rounded up, while the millisecond calculation truncates, causing the log timestamp to be wrong by up to 999 ms (affected roughly 1 in 8 million timestamps).

gh-118710 : ipaddress.IPv4Address and ipaddress.IPv6Address attributes version and max_prefixlen are now available on the class.

gh-120633 : Move scrollbar and remove tear-off menus in turtledemo.

gh-120606 : Allow users to use EOF to exit commands definition in pdb

gh-120541 : Improve the prompt in the “less” pager when help() is called with non-string argument.

gh-120495 : Fix incorrect exception handling in Tab Nanny. Patch by Wulian233.

gh-120388 : Improve a warning message when a test method in unittest returns something other than None . Now we show the returned object type and optional asyncio-related tip.

gh-120381 : Correct inspect.ismethoddescriptor() to check also for the lack of __delete__() . Patch by Jan Kaliszewski.

gh-90425 : The OS byte in gzip headers is now always set to 255 when using gzip.compress() .

gh-120343 : Fix column offset reporting for tokens that come after multiline f-strings in the tokenize module.

gh-119180 : As part of implementing PEP 649 and PEP 749 , add a new module annotationlib . Add support for unresolved forward references in annotations to dataclasses , typing.TypedDict , and typing.NamedTuple .

gh-119600 : Fix unittest.mock.patch() to not read attributes of the target when new_callable is set. Patch by Robert Collins.

gh-120289 : Fixed the use-after-free issue in cProfile by disallowing disable() and clear() in external timers.

gh-82017 : Added support for converting any objects that have the as_integer_ratio() method to a Fraction .

gh-114053 : Fix edge-case bug where typing.get_type_hints() would produce incorrect results if type parameters in a class scope were overridden by assignments in a class scope and from __future__ import annotations semantics were enabled. Patch by Alex Waygood.

gh-114053 : Fix erroneous NameError when calling inspect.get_annotations() with eval_str=True` on a class that made use of PEP 695 type parameters in a module that had from __future__ import annotations at the top of the file. Patch by Alex Waygood.

gh-120268 : Prohibit passing None to pure-Python datetime.date.fromtimestamp() to achieve consistency with C-extension implementation.

gh-120244 : Fix memory leak in re.sub() when the replacement string contains backreferences.

gh-120211 : Fix tkinter.ttk with Tcl/Tk 9.0.

gh-71587 : Fix crash in C version of datetime.datetime.strptime() when called again on the restarted interpreter.

gh-117983 : Defer the threading import in importlib.util until lazy loading is used.

gh-120157 : Remove unused constant concurrent.futures._base._FUTURE_STATES in concurrent.futures . Patch by Clinton Christian (pygeek).

gh-120161 : datetime no longer crashes in certain complex reference cycle situations.

gh-119698 : Fix symtable.Class.get_methods() and document its behaviour. Patch by Bénédikt Tran.

gh-120121 : Add concurrent.futures.InvalidStateError to module’s __all__ .

gh-119933 : Add the symtable.SymbolTableType enumeration to represent the possible outputs of the symtable.SymbolTable.get_type method. Patch by Bénédikt Tran.

gh-120029 : Expose symtable.Symbol methods is_free_class() , is_comp_iter() and is_comp_cell() . Patch by Bénédikt Tran.

gh-120108 : Fix calling copy.deepcopy() on ast trees that have been modified to have references to parent nodes. Patch by Jelle Zijlstra.

gh-120056 : Add socket.IP_RECVERR and socket.IP_RECVTTL constants (both available since Linux 2.2). And socket.IP_RECVORIGDSTADDR constant (available since Linux 2.6.29).

gh-112672 : Support building tkinter with Tcl 9.0.

gh-120057 : Added the os.environ.refresh() method to update os.environ with changes to the environment made by os.putenv() , by os.unsetenv() , or made outside Python in the same process. Patch by Victor Stinner.

gh-120029 : Expose symtable.Symbol.is_type_parameter() in the symtable module. Patch by Bénédikt Tran.

gh-119819 : Fix regression to allow logging configuration with multiprocessing queue types.

gh-65454 : unittest.mock.Mock.attach_mock() no longer triggers a call to a PropertyMock being attached.

gh-117142 : The ctypes module may now be imported in all subinterpreters, including those that have their own GIL.

gh-118835 : Fix _pyrepl crash when using custom prompt with ANSI escape codes.

gh-81936 : help() and showtopic() methods now respect a configured output argument to pydoc.Helper and not use the pager in such cases. Patch by Enrico Tröger.

gh-117398 : The _datetime module (C implementation for datetime ) now supports being imported in multiple interpreters.

gh-119824 : Print stack entry in pdb when and only when user input is needed.

gh-119838 : In mixed arithmetic operations with Fraction and complex, the fraction is now converted to float instead of complex .

gh-119770 : Make termios ioctl() constants positive. Patch by Victor Stinner.

gh-89727 : Fix issue with shutil.rmtree() where a RecursionError is raised on deep directory trees.

gh-119577 : The DeprecationWarning emitted when testing the truth value of an xml.etree.ElementTree.Element now describes unconditionally returning True in a future version rather than raising an exception in Python 3.14.

gh-89727 : Partially fix issue with shutil.rmtree() where a RecursionError is raised on deep directory trees. A recursion error is no longer raised when rmtree.avoids_symlink_attacks is false.

gh-93963 : Remove deprecated names from importlib.abc as found in importlib.resources.abc .

gh-119118 : Fix performance regression in the tokenize module by caching the line token attribute and calculating the column offset more efficiently.

gh-89727 : Fix issue with os.fwalk() where a RecursionError was raised on deep directory trees by adjusting the implementation to be iterative instead of recursive.

gh-119594 : If one calls pow(fractions.Fraction, x, module) with modulo not None, the error message now says that the types are incompatible rather than saying pow only takes 2 arguments. Patch by Wim Jeantine-Glenn and Mark Dickinson.

gh-119588 : zipfile.Path.is_symlink now assesses if the given path is a symlink.

gh-119562 : Remove ast.Num , ast.Str , ast.Bytes , ast.NameConstant and ast.Ellipsis . They had all emitted deprecation warnings since Python 3.12. Patch by Alex Waygood.

gh-119555 : Catch SyntaxError from compile() in the runsource() method of the InteractiveColoredConsole. Patch by Sergey B Kirpichev.

gh-118908 : Limit exposed globals from internal imports and definitions on new REPL startup. Patch by Eugene Triguba and Pablo Galindo.

gh-117865 : Improve the import time of the ast module by deferring the import of re . Patch by Jelle Zijlstra.

gh-113892 : Now, the method sock_connect of asyncio.ProactorEventLoop raises a ValueError if given socket is not in non-blocking mode, as well as in other loop implementations.

gh-119506 : Fix io.TextIOWrapper.write() method breaks internal buffer when the method is called again during flushing internal buffer.

gh-69214 : Fix fcntl.ioctl() request parameter: use an unsigned long instead of an unsigned int for the request parameter of fcntl.ioctl() to support requests larger than UINT_MAX . Patch by Victor Stinner.

gh-119105 : difflib ’s DIffer.compare() (and so also ndiff ) can no longer be provoked into cubic-time behavior, or into unbounded recursion, and should generally be faster in ordinary cases too. Results may change in some cases, although that should be rare. Correctness of diffs is not affected. Some similar lines far apart may be reported as deleting one and adding the other, where before they were displayed on adjacent output lines with markup showing the intraline differences.

gh-119443 : The interactive REPL no longer runs with from __future__ import annotations enabled. Patch by Jelle Zijlstra.

gh-119461 : Add socket.VMADDR_CID_LOCAL constant. Patch by Victor Stinner.

gh-117398 : Objects in the datetime C-API are now all statically allocated, which means better memory safety, especially when the module is reloaded. This should be transparent to users.

gh-118894 : asyncio REPL now has the same capabilities as PyREPL.

gh-118830 : Bump pickle default protocol to 5 .

gh-118911 : In PyREPL, updated maybe-accept ’s logic so that if the user hits Enter twice, they are able to terminate the block even if there’s trailing whitespace. Also, now when the user hits arrow up, the cursor is on the last functional line. This matches IPython’s behavior. Patch by Aya Elsayed.

gh-115225 : Raise error on certain technically valid but pathological ISO 8601 strings passed to datetime.time.fromisoformat() that were previously parsed incorrectly.

gh-111201 : Remove dependency to readline from the new Python REPL.

gh-119189 : When using the ** operator or pow() with Fraction as the base and an exponent that is not rational, a float, or a complex, the fraction is no longer converted to a float.

gh-119174 : Fix high DPI causes turtledemo(turtle-graphics examples) windows blurry Patch by Wulian233 and Terry Jan Reedy

gh-119121 : Fix a NameError happening in asyncio.staggered.staggered_race . This function is now tested.

gh-119105 : difflib.Differ is much faster for some cases of diffs where many pairs of lines are equally similar.

gh-119113 : Fix issue where pathlib.PurePath.with_suffix() didn’t raise TypeError when given None as a suffix.

gh-118643 : Fix an AttributeError in the email module when re-fold a long address list. Also fix more cases of incorrect encoding of the address separator in the address list.

gh-73991 : Add pathlib.Path.copy() , which copies a file or directory to another.

gh-58933 : Make pdb return to caller frame correctly when f_trace of the caller frame is not set

gh-82805 : Support single-dot file extensions in pathlib.PurePath.suffix and related attributes and methods. For example, the suffixes of PurePath('foo.bar.') are now ['.bar', '.'] rather than [] . This brings file extension splitting in line with os.path.splitext() .

gh-118924 : Remove version and version_info from sqlite3 . Patch by Hugo van Kemenade.

gh-118928 : Disallow using a sequence of parameters with named placeholders in sqlite3 queries. Patch by Erlend E. Aasland.

gh-118895 : Setting attributes on typing.NoDefault now raises AttributeError instead of TypeError .

gh-118868 : Fixed issue where kwargs were no longer passed to the logging handler QueueHandler

gh-118827 : Remove deprecated Quoter class from urllib.parse . It had previously raised a DeprecationWarning since Python 3.11. Patch by Nikita Sobolev.

gh-118824 : Remove deprecated pty.master_open() and pty.slave_open() . Use pty.openpty() instead. Patch by Nikita Sobolev.

gh-118851 : ctx arguments to the constructors of ast node classes now default to ast.Load() . Patch by Jelle Zijlstra.

gh-101588 : Remove copy, deepcopy, and pickle from itertools. These had previously raised a DeprecationWarning since Python 3.12.

gh-118805 : Remove type , choices , and metavar parameters of argparse.BooleanOptionalAction . They were deprecated since Python 3.12.

gh-118803 : typing.ByteString and collections.abc.ByteString are removed. They had previously raised a DeprecationWarning since Python 3.12.

gh-118798 : The isdst parameter has been removed from email.utils.localtime() . Patch by Hugo van Kemenade.

gh-118760 : Restore the default value of tkiter.wantobjects to 1 .

gh-118760 : Fix errors in calling Tkinter bindings on Windows.

gh-74033 : Drop support for passing keyword arguments to pathlib.Path .

gh-101357 : Suppress all OSError exceptions from pathlib.Path.exists() and is_*() methods, rather than a selection of more common errors. The new behaviour is consistent with os.path.exists() , os.path.isdir() , etc. Use pathlib.Path.stat() to retrieve the file status without suppressing exceptions.

gh-78707 : Drop support for passing additional positional arguments to pathlib.PurePath.relative_to() and is_relative_to() .

gh-118507 : Fix os.path.isfile() on Windows for pipes. Speedup os.path.isjunction() and os.path.lexists() on Windows with a native implementation.

gh-118772 : Allow typing.TypeVar instances without a default to follow instances without a default in some cases. Patch by Jelle Zijlstra.

gh-118714 : Allow restart in post-mortem debugging of pdb . Removed restart message when the user quits pdb from post-mortem mode.

gh-118673 : Removed executable bits and shebang from stdlib modules.

gh-110863 : os.path.realpath() now suppresses any OSError from os.readlink() when strict mode is disabled (the default).

gh-118263 : Speed up os.path.splitroot() & os.path.normpath() with a direct C call.

gh-118033 : Fix dataclasses.dataclass() not creating a __weakref__ slot when subclassing typing.Generic .

gh-106531 : In importlib.resources , sync with importlib_resources 6.3.2 , including: MultiplexedPath now expects Traversable paths, deprecating string arguments to MultiplexedPath ; Enabled support for resources in namespace packages in zip files; Fixed NotADirectoryError when calling files on a subdirectory of a namespace package.

gh-115808 : Add operator.is_none() and operator.is_not_none() functions.

gh-114264 : Improve performance of copy.deepcopy() by adding a fast path for atomic types.

gh-113978 : Ignore warnings on text completion inside REPL.

gh-108172 : webbrowser honors OS preferred browser on Linux when its desktop entry name contains the text of a known browser name.

gh-109109 : You can now get the raw TLS certificate chains from TLS connections via ssl.SSLSocket.get_verified_chain() and ssl.SSLSocket.get_unverified_chain() methods.

Contributed by Mateusz Nowak.

gh-105623 : Fix performance degradation in logging.handlers.RotatingFileHandler . Patch by Craig Robson.

gh-103956 : Fix lack of newline characters in trace module output when line tracing is enabled but source code line for current frame is not available.

gh-92081 : Fix missing spaces in email headers when the spaces are mixed with encoded 8-bit characters.

gh-103194 : Prepare Tkinter for C API changes in Tcl 8.7/9.0 to avoid _tkinter.Tcl_Obj being unexpectedly returned instead of bool , str , bytearray , or int .

gh-87106 : Fixed handling in inspect.Signature.bind() of keyword arguments having the same name as positional-only arguments when a variadic keyword argument (e.g. **kwargs ) is present.

gh-103134 : Add additional methods to proxy objects in the multiprocessing module:

clear() and copy() for proxies of list

fromkeys() , reversed(d) , d | {} , {} | d , d |= {'b': 2} for proxies of dict

bpo-45767 : Fix integer conversion in os.major() , os.minor() , and os.makedev() . Support device numbers larger than 2**63-1 . Support non-existent device number ( NODEV ).

bpo-15987 : Implemented ast.compare() for comparing two ASTs. Patch by Batuhan Taskaya with some help from Jeremy Hylton.

gh-67693 : Fix urllib.parse.urlunparse() and urllib.parse.urlunsplit() for URIs with path starting with multiple slashes and no authority. Based on patch by Ashwin Ramaswami.

gh-122482 : Change About IDLE to direct users to discuss.python.org instead of the now unused idle-dev email and mailing list.

gh-78889 : Stop Shell freezes by blocking user access to non-method sys.stdout.shell attributes, which are all private.

gh-120083 : Add explicit black IDLE Hovertip foreground color needed for recent macOS. Fixes Sonoma showing unreadable white on pale yellow. Patch by John Riggles.

gh-120104 : Fix padding in config and search dialog windows in IDLE.

Documentation ¶

gh-117765 : Improved documentation for unittest.mock.patch.dict()

gh-121749 : Fix documentation for PyModule_AddObjectRef() .

gh-120012 : Clarify the behaviours of multiprocessing.Queue.empty() and multiprocessing.SimpleQueue.empty() on closed queues. Patch by Bénédikt Tran.

gh-119574 : Added some missing environment variables to the output of --help-env .

Core and Builtins ¶

gh-123572 : Fix key mappings for various F-keys in Windows for the new REPL. Patch by devdanzin

gh-123562 : Improve SyntaxError message for using case ... as ... with not a name.

gh-123545 : Fix a double decref in rare cases on experimental JIT builds.

gh-123484 : Fix _Py_DebugOffsets for long objects to be relative to the start of the object rather than the start of a subobject.

gh-123446 : Fix empty function name in TypeError when builtin magic methods are used without the required args.

gh-123440 : Improve SyntaxError message for using except as with not a name.

gh-123344 : Add AST optimizations for type parameter defaults.

gh-123321 : Prevent Parser/myreadline race condition from segfaulting on multi-threaded use. Patch by Bar Harel and Amit Wienner.

gh-123177 : Fix a bug causing stray prompts to appear in the middle of wrapped lines in the new REPL.

gh-122982 : Extend the deprecation period for bool inversion ( ~ ) by two years.

gh-123271 : Make concurrent iterations over the same zip() iterator safe under free-threading.

gh-123177 : Deactivate line wrap in the Apple Terminal via a ANSI escape code. Patch by Pablo Galindo

gh-123229 : Fix valgrind warning by initializing the f-string buffers to 0 in the tokenizer. Patch by Pablo Galindo

gh-122298 : Restore printout of GC stats when gc.set_debug(gc.DEBUG_STATS) is called. This featue was accidentally removed when implementing incremental GC.

gh-121804 : Correctly show error locations when a SyntaxError is raised in the basic REPL. Patch by Sergey B Kirpichev.

gh-115776 : Enables inline values (Python’s equivalent of hidden classes) on any class who’s instances are of a fixed size.

gh-123142 : Fix too-wide source location in exception tracebacks coming from broken iterables in comprehensions.

gh-123048 : Fix a bug where pattern matching code could emit a JUMP_FORWARD with no source location.

gh-118093 : Break up CALL_ALLOC_AND_ENTER_INIT into micro-ops and relax requirement for exact args, in order to increase the amount of code supported by tier 2.

gh-123123 : Fix displaying SyntaxError exceptions covering multiple lines. Patch by Pablo Galindo

gh-123083 : Fix a potential use-after-free in STORE_ATTR_WITH_HINT .

gh-123022 : Fix crash in free-threaded build when calling Py_Initialize() from a non-main thread.

gh-118093 : Add three specializations for CALL_KW :

CALL_KW_PY for calls to Python functions

CALL_KW_BOUND_METHOD for calls to bound methods

CALL_KW_NON_PY for all other calls

gh-122821 : Make sure that branches in while statements have consistent offsets for sys.monitoring . while statements are now compiled with a simple jump at the end of the body, instead of duplicating the test.

gh-122907 : Building with HAVE_DYNAMIC_LOADING now works as well as it did in 3.12. Existing deficiences will be addressed separately. (See https://github.com/python/cpython/issues/122950 .)

gh-122888 : Fix crash on certain calls to str() with positional arguments of the wrong type. Patch by Jelle Zijlstra.

gh-118093 : Improve the experimental JIT’s handling of polymorphic code.

gh-122697 : Fixed memory leaks at interpreter shutdown in the free-threaded build, and also reporting of leaked memory blocks via -X showrefcount .

gh-116622 : Fix Android stdout and stderr messages being truncated or lost.

gh-122527 : Fix a crash that occurred when a PyStructSequence was deallocated after its type’s dictionary was cleared by the GC. The type’s tp_basicsize now accounts for non-sequence fields that aren’t included in the Py_SIZE of the sequence.

gh-122445 : Add only fields which are modified via self.* to __static_attributes__ .

gh-122417 : In the free-threaded build, the reference counts for heap type objects are now partially stored in a distributed manner in per-thread arrays. This reduces contention on the heap type’s reference count fields when creating or destroying instances of the same type from multiple threads concurrently.

gh-116090 : Fix an issue in JIT builds that prevented some for loops from correctly firing RAISE monitoring events.

gh-122300 : Preserve AST nodes for f-string with single-element format specifiers. Patch by Pablo Galindo

gh-120906 : frame.f_locals now supports arbitrary hashable objects as keys.

gh-122245 : Detection of writes to __debug__ is moved from the compiler’s codegen stage to the symtable. This means that these errors are now detected even in code that is optimized away before codegen (such as assertions with the -O command line option).

gh-122234 : Specializations for sums with float and complex inputs in sum() now always use compensated summation. Also, for integer items in above specializations: PyLong_AsDouble() is used, instead of PyLong_AsLongAndOverflow() . Patch by Sergey B Kirpichev.

gh-122208 : Dictionary watchers now only deliver the PyDict_EVENT_ADDED event when the insertion is in a known good state to succeed.

gh-122160 : Remove the BUILD_CONST_KEY_MAP opcode. Use BUILD_MAP instead.

gh-122029 : Emit c_call events in sys.setprofile() when a PyMethodObject pointing to a PyCFunction is called.

gh-122026 : Fix a bug that caused the tokenizer to not correctly identify mismatched parentheses inside f-strings in some situations. Patch by Pablo Galindo

gh-99108 : Python’s hashlib now unconditionally uses the vendored HACL* library for Blake2. Python no longer accepts libb2 as an optional dependency for Blake2.

We refreshed HACL* to the latest version, and now vendor HACL*’s 128-bit and 256-bit wide vector implementations for Blake2, which are used on x86/x64 toolchains when the required CPU features are available at runtime.

HACL*’s 128-bit wide vector implementation of Blake2 can also run on ARM NEON and Power8, but lacking evidence of a performance gain, these are not enabled (yet).

gh-121999 : The default extraction filter for the tarfile module is now set to 'data' .

gh-121860 : Fix crash when rematerializing a managed dictionary after it was deleted.

gh-121795 : Improve performance of set membership testing, set.remove() and set.discard() when the argument is a set.

gh-121814 : Fixed the SegFault when PyEval_SetTrace() is used with no Python frame on stack.

gh-121295 : Fix PyREPL console getting into a blocked state after interrupting a long paste

gh-121794 : Fix bug in free-threaded Python where a resurrected object could lead to a negative ref count assertion failure.

gh-121657 : Improve the SyntaxError message if the user tries to use yield from outside a function.

gh-121609 : Fix pasting of characters containing unicode character joiners in the new REPL. Patch by Marta Gomez Macias

gh-121297 : Previously, incorrect usage of await or asynchronous comprehensions in code removed by the -O option was not flagged by the Python compiler. Now, such code raises SyntaxError . Patch by Jelle Zijlstra.

gh-117482 : Unexpected slot wrappers are no longer created for builtin static types in subinterpreters.

gh-121562 : Optimized performance of hex_from_char by replacing switch-case with a lookup table

gh-121499 : Fix a bug affecting how multi-line history was being rendered in the new REPL after interacting with the new screen cache. Patch by Pablo Galindo

gh-121497 : Fix a bug that was preventing the REPL to correctly respect the history when an input hook was set. Patch by Pablo Galindo

gh-121012 : Tier 2 execution now ensures that list iterators remain exhausted, once they become exhausted.

gh-121439 : Allow tuples of length 20 in the freelist to be reused.

gh-121288 : ValueError messages for list.index() , range.index() , deque.index() , deque.remove() and ShareableList.index() no longer contain the repr of the searched value (which can be arbitrary large) and are consistent with error messages for other index() and remove() methods.

gh-121368 : Fix race condition in _PyType_Lookup in the free-threaded build due to a missing memory fence. This could lead to _PyType_Lookup returning incorrect results on arm64.

gh-121149 : Added specialization for summation of complexes, this also improves accuracy of builtin sum() for such inputs. Patch by Sergey B Kirpichev.

gh-121130 : Fix f-strings with debug expressions in format specifiers. Patch by Pablo Galindo

gh-121381 : Remove subprocess._USE_VFORK escape hatch code and documentation. It was added just in case, and doesn’t have any known cases that require it.

gh-119726 : Optimize code layout for calls to C functions from the JIT on AArch64. Patch by Diego Russo.

gh-121115 : PyLong_AsNativeBytes() no longer uses __index__() methods by default. The Py_ASNATIVEBYTES_ALLOW_INDEX flag has been added to allow it.

gh-120838 : Py_Finalize() and Py_FinalizeEx() now always run with the main interpreter active.

gh-113433 : Subinterpreters now get cleaned up automatically during runtime finalization.

gh-119726 : Improve the speed and memory use of C function calls from JIT code on AArch64. Patch by Diego Russo

gh-116017 : Simplify the warmup mechanism used for “side exits” in JIT code, resulting in slightly better performance and slightly lower memory usage for most platforms.

gh-98442 : Fix too wide source locations of the cleanup instructions of a with statement.

gh-120754 : Reduce the number of system calls invoked when reading a whole file (ex. open('a.txt').read() ). For a sample program that reads the contents of the 400+ .rst files in the cpython repository Doc folder, there is an over 10% reduction in system call count.

gh-119462 : Make sure that invariants of type versioning are maintained: * Superclasses always have their version number assigned before subclasses * The version tag is always zero if the tag is not valid. * The version tag is always non-if the tag is valid.

gh-120437 : Fix _CHECK_STACK_SPACE optimization problems introduced in gh-118322 .

gh-120722 : Correctly set the bytecode position on return instructions within lambdas. Patch by Jelle Zijlstra.

gh-120367 : Fix bug where compiler creates a redundant jump during pseudo-op replacement. Can only happen with a synthetic AST that has a try on the same line as the instruction following the exception handler.

gh-120507 : Remove the BEFORE_WITH and BEFORE_ASYNC_WITH instructions. Add the new LOAD_SPECIAL instruction. Generate code for with and async with statements using the new instruction.

gh-113993 : Strings interned with sys.intern() are again garbage-collected when no longer used, as per the documentation. Strings interned with the C function PyUnicode_InternInPlace() are still immortal. Internals of the string interning mechanism have been changed. This may affect performance and identities of str objects.

gh-120485 : Add an override of allow_reuse_port on classes subclassing socketserver.TCPServer where allow_reuse_address is also overridden.

gh-120384 : Fix an array out of bounds crash in list_ass_subscript , which could be invoked via some specifically tailored input: including concurrent modification of a list object, where one thread assigns a slice and another clears it.

gh-120367 : Fix crash in compiler on code with redundant NOPs and JUMPs which show up after exception handlers are moved to the end of the code.

gh-120380 : Fix Python implementation of pickle.Pickler for bytes and bytearray objects when using protocol version 5. Patch by Bénédikt Tran.

gh-120397 : Improve the throughput by up to two times for the str.count() , bytes.count() and bytearray.count() methods for counting single characters.

gh-120221 : Deliver real signals on Ctrl-C and Ctrl-Z in the new REPL. Patch by Pablo Galindo

gh-120346 : Respect PYTHON_BASIC_REPL when running in interactive inspect mode ( python -i ). Patch by Pablo Galindo

gh-93691 : Fix source locations of instructions generated for the iterator of a for statement.

gh-120198 : Fix a crash when multiple threads read and write to the same __class__ of an object concurrently.

gh-120298 : Fix use-after free in list_richcompare_impl which can be invoked via some specifically tailored evil input.

gh-119666 : Fix a compiler crash in the case where two comprehensions in class scope both reference __class__ .

gh-119726 : JIT: Re-use trampolines on AArch64 when creating stencils. Patch by Diego Russo

gh-120225 : Fix crash in compiler on empty block at end of exception handler.

gh-93691 : Fix source locations of instructions generated for with statements.

gh-120097 : FrameLocalsProxy now subclasses collections.abc.Mapping and can be matched as a mapping in match statements

gh-120080 : Direct call to the int.__round__() now accepts None as a valid argument.

gh-119933 : Improve SyntaxError messages for invalid expressions in a type parameters bound, a type parameter constraint tuple or a default type parameter. Patch by Bénédikt Tran.

gh-119724 : Reverted improvements to error messages for elif / else statements not matching any valid statements, which made in hard to locate the syntax errors inside those elif / else blocks.

gh-119879 : String search is now slightly faster for certain cases. It now utilizes last character gap (good suffix rule) for two-way periodic needles.

gh-119842 : Honor PyOS_InputHook() in the new REPL. Patch by Pablo Galindo

gh-119180 : classmethod() and staticmethod() now wrap the __annotations__ and __annotate__ attributes of their underlying callable lazily. See PEP 649 . Patch by Jelle Zijlstra.

gh-119821 : Fix execution of annotation scopes within classes when globals is set to a non-dict. Patch by Jelle Zijlstra.

gh-118934 : Make PyEval_GetLocals return borrowed reference

gh-119740 : Remove the previously-deprecated delegation of int() to __trunc__() .

gh-119689 : Generate stack effect metadata for pseudo instructions from bytecodes.c.

gh-109218 : complex() accepts now a string only as a positional argument. Passing a complex number as the “real” or “imag” argument is deprecated; it should only be passed as a single positional argument.

gh-119548 : Add a clear command to the REPL. Patch by Pablo Galindo

gh-111999 : Fix the signature of str.format_map() .

gh-119560 : An invalid assert in beta 1 has been removed. The assert would fail if PyState_FindModule() was used in an extension module’s init function before the module def had been initialized.

gh-119369 : Fix deadlock during thread deletion in free-threaded build, which could occur when the GIL was enabled at runtime.

gh-119525 : Fix deadlock involving _PyType_Lookup() cache in the free-threaded build when the GIL is dynamically enabled at runtime.

gh-119258 : Eliminate type version guards in the tier two interpreter.

Note that setting the tp_version_tag manually (which has never been supported) may result in crashes.

gh-119311 : Fix bug where names are unexpectedly mangled in the bases of generic classes.

gh-119395 : Fix bug where names appearing after a generic class are mangled as if they are in the generic class.

gh-119372 : Correct invalid corner cases in complex division (resulted in (nan+nanj) output), e.g. 1/complex('(inf+infj)') . Patch by Sergey B Kirpichev.

gh-119180 : Evaluation of annotations is now deferred. See PEP 649 for details.

gh-119180 : Replace LOAD_ASSERTION_ERROR opcode with LOAD_COMMON_CONSTANT and add support for NotImplementedError .

gh-119213 : Non-builtin modules built with argument clinic were crashing if used in a subinterpreter before the main interpreter. The objects that were causing the problem by leaking between interpreters carelessly have been fixed.

gh-119011 : Fixes type.__type_params__ to return an empty tuple instead of a descriptor.

gh-118692 : Avoid creating unnecessary StopIteration instances for monitoring.

gh-119180 : Add an __annotate__ attribute to functions, classes, and modules as part of PEP 649 . Patch by Jelle Zijlstra.

gh-119049 : Fix displaying the source line for warnings created by the C API if the warnings module had not yet been imported.

gh-119057 : Improve ZeroDivisionError error message. Now, all error messages are harmonized: all / , // , and % operations just use “division by zero” message. And 0 ** -1 operation uses “zero to a negative power”.

gh-118844 : Fix build failures when configuring with both --disable-gil and --enable-experimental-jit .

gh-118921 : Add copy() method for FrameLocalsProxy which returns a snapshot dict for local variables.

gh-117657 : Fix data races on the field that stores a pointer to the interpreter’s main thread that occur in free-threaded builds.

gh-118750 : If the C version of the decimal module is available, int(str) now uses it to supply an asymptotically much faster conversion. However, this only applies if the string contains over about 2 million digits.

gh-118767 : Using NotImplemented in a boolean context now raises TypeError . Contributed by Jelle Zijlstra.

gh-118561 : Fix race condition in free-threaded build where list.extend() could expose uninitialised memory to concurrent readers.

gh-117139 : Convert the Python evaluation stack to use internal stack references. The purpose is to support tagged pointers. In PEP 703 , this will allow for its form of deferred reference counting. For both the default and free-threaded builds, this sets up the infrastructure for unboxed integers in the future.

gh-115801 : Raise TypeError when passing a string to difflib.unified_diff() and difflib.context_diff() .

gh-117558 : Improve error messages when a string, bytes or bytearray object of length 1 is expected.

gh-117195 : Avoid assertion failure for debug builds when calling object.__sizeof__(1)

gh-114091 : Changed the error message for awaiting something that can’t be awaited from “object <type> can’t be used in an await expression” to “’<type>’ object can’t be awaited”.

gh-113190 : Py_Finalize() now deletes all interned strings.

gh-84978 : Add class methods float.from_number() and complex.from_number() .

gh-95144 : Improve the error message from a in b when b is not a container to mention the term “container”.

bpo-24766 : Fix handling of doc argument to subclasses of property .

gh-107954 : Add functions to configure the Python initialization ( PEP 741 ):

PyInitConfig_Create()

PyInitConfig_Free()

PyInitConfig_GetError()

PyInitConfig_GetExitCode()

PyInitConfig_HasOption()

PyInitConfig_GetInt()

PyInitConfig_GetStr()

PyInitConfig_GetStrList()

PyInitConfig_FreeStrList()

PyInitConfig_SetInt()

PyInitConfig_SetStr()

PyInitConfig_SetStrList()

PyInitConfig_AddModule()

Py_InitializeFromInitConfig()

Patch by Victor Stinner.

gh-107954 : Add functions to get and set the current runtime Python configuration:

PyConfig_Get()

PyConfig_GetInt()

PyConfig_Set()

PyConfig_Names()

gh-123465 : Py_RELATIVE_OFFSET is now allowed in PyMemberDef for the special offset member "__vectorcalloffset__" , as well as the discouraged special offset members "__dictoffset__" and "__weaklistoffset__"

gh-122854 : Add Py_HashBuffer() to compute and return the hash value of a buffer. Patch by Antoine Pitrou and Victor Stinner.

gh-122728 : Fix PyEval_GetLocals() to avoid SystemError (“bad argument to internal function”). Patch by Victor Stinner.

gh-116622 : Make PyObject_Print work around a bug in Android and OpenBSD which prevented it from throwing an exception when trying to write to a read-only stream.

gh-105201 : Add PyIter_NextItem() to replace PyIter_Next() , which has an ambiguous return value. Patch by Irit Katriel and Erlend Aasland.

gh-121489 : Export private _PyBytes_Join() again.

gh-121645 : Add PyBytes_Join(sep, iterable) function, similar to sep.join(iterable) in Python. Patch by Victor Stinner.

gh-89364 : Export the PySignal_SetWakeupFd() function. Previously, the function was documented but it couldn’t be used in 3rd party code. Patch by Victor Stinner.

gh-113993 : PyUnicode_InternInPlace() no longer prevents its argument from being garbage collected.

Several functions that take char * are now documented as possibly preventing string objects from being garbage collected; refer to their documentation for details: PyUnicode_InternFromString() , PyDict_SetItemString() , PyObject_SetAttrString() , PyObject_DelAttrString() , PyUnicode_InternFromString() , and PyModule_Add* convenience functions.

gh-113601 : Removed debug build assertions related to interning strings, which were falsely triggered by stable ABI extensions.

gh-112136 : Restore the private _PyArg_Parser structure and the private _PyArg_ParseTupleAndKeywordsFast() function, previously removed in Python 3.13 alpha 1. Patch by Victor Stinner.

gh-120642 : Remove the private _Py_CODEUNIT type from the public C API. The internal pycore_code.h header should now be used to get this internal type. Patch by Victor Stinner.

gh-120858 : PyDict_Next() no longer locks the dictionary in the free-threaded build. The locking needs to be done by the caller around the entire iteration loop.

gh-120642 : Remove the following unstable functions:

PyUnstable_Replace_Executor()

PyUnstable_SetOptimizer()

PyUnstable_GetOptimizer()

PyUnstable_GetExecutor()

PyUnstable_Optimizer_NewCounter()

PyUnstable_Optimizer_NewUOpOptimizer()

gh-120389 : Add new functions to convert C <stdint.h> numbers from/to Python int :

PyLong_FromInt32()

PyLong_FromUInt32()

PyLong_FromInt64()

PyLong_FromUInt64()

PyLong_AsInt32()

PyLong_AsUInt32()

PyLong_AsInt64()

PyLong_AsUInt64()

gh-120600 : In the limited C API 3.14 and newer, Py_TYPE() is now implemented as an opaque function call to hide implementation details. Patch by Victor Stinner.

gh-70278 : PyUnicode_FromFormat() no longer produces the ending \ufffd character for truncated C string when use precision with %s and %V . It now truncates the string before the start of truncated multibyte sequences.

gh-119182 : Add a new PyUnicodeWriter API to create a Python str object:

PyUnicodeWriter_Create() .

PyUnicodeWriter_Discard() .

PyUnicodeWriter_Finish() .

PyUnicodeWriter_WriteChar() .

PyUnicodeWriter_WriteUTF8() .

PyUnicodeWriter_WriteUCS4() .

PyUnicodeWriter_WriteWideChar() .

PyUnicodeWriter_WriteStr() .

PyUnicodeWriter_WriteRepr() .

PyUnicodeWriter_WriteSubstring() .

PyUnicodeWriter_Format() .

PyUnicodeWriter_DecodeUTF8Stateful() .

gh-119613 : Soft deprecate the Py_MEMCPY macro: use directly memcpy() instead. Patch by Victor Stinner.

gh-119775 : Creating immutable types with mutable bases was deprecated since 3.12 and now raises a TypeError .

gh-119585 : Fix crash when a thread state that was created by PyGILState_Ensure() calls a destructor that during PyThreadState_Clear() that calls back into PyGILState_Ensure() and PyGILState_Release() . This might occur when in the free-threaded build or when using thread-local variables whose destructors call PyGILState_Ensure() .

gh-119613 : Macros Py_IS_NAN , Py_IS_INFINITY and Py_IS_FINITE are soft deprecated .

gh-119336 : Restore the removed _PyLong_NumBits() function. It is used by the pywin32 project. Patch by Ethan Smith

gh-119344 : The critical section API is now public as part of the non-limited C API.

gh-119247 : Added Py_BEGIN_CRITICAL_SECTION_SEQUENCE_FAST and Py_END_CRITICAL_SECTION_SEQUENCE_FAST macros to make it possible to use PySequence_Fast APIs safely when free-threaded, and update str.join to work without the GIL using them.

gh-111389 : Add PyHASH_MULTIPLIER constant: prime multiplier used in string and various other hashes. Patch by Victor Stinner.

gh-118771 : Several C declarations with names that didn’t start with the Py or _Py prefixes, which were added by mistake in 3.13 alpha and beta releases, were moved to internal headers.

gh-116984 : Make mimalloc includes relative to the current file to avoid embedders or extensions needing to include Internal/mimalloc if they are already including internal CPython headers.

gh-118789 : Add PyUnstable_Object_ClearWeakRefsNoCallbacks() , which clears weakrefs without calling their callbacks.

gh-118789 : Restore _PyWeakref_ClearRef that was previously removed in Python 3.13 alpha 1.

gh-117511 : Make the PyMutex public in the non-limited C API.

gh-116560 : Add PyLong_GetSign() function. Patch by Sergey B Kirpichev.

gh-97588 : Fix creating bitfields in ctypes structures and unions. Fields no longer overlap.

gh-123418 : Updated Android build to use OpenSSL 3.0.15.

gh-123297 : Propagate the value of LDFLAGS to LDCXXSHARED in sysconfig . Patch by Pablo Galindo

gh-121634 : Allow for specifying the target compile triple for WASI.

gh-122578 : Use WASI SDK 24 for testing.

gh-116622 : Rename build variable MODULE_LDFLAGS back to LIBPYTHON , as it’s used by package build systems (e.g. Meson).

gh-118943 : Fix an issue where the experimental JIT could be built several times by the make regen-all target, leading to possible race conditions on heavily parallelized builds.

gh-121996 : Introduce ./configure –disable-safety and –enable-slower-safety options. Patch by Donghee Na.

gh-120522 : Added a --with-app-store-compliance option to patch out known issues with macOS/iOS App Store review processes.

gh-120371 : Support WASI SDK 22 by explicitly skipping functions that are just stubs in wasi-libc.

gh-121731 : Fix mimalloc compile error on GNU/Hurd

gh-121487 : Fix deprecation warning for ATOMIC_VAR_INIT in mimalloc.

gh-121467 : Fix a Makefile bug that prevented mimalloc header files from being installed.

gh-121103 : On POSIX systems, excluding macOS framework installs, the lib directory for the free-threaded build now includes a “t” suffix to avoid conflicts with a co-located default build installation.

gh-120831 : The default minimum iOS version was increased to 13.0.

gh-121082 : Fix build failure when the developer use --enable-pystats arguments in configuration command after #118450.

gh-120671 : Fix failing configure tests due to a missing space when appending to CFLAGS.

gh-120602 : Correctly handle LLVM installs with LLVM_VERSION_SUFFIX when building with --enable-experimental-jit .

gh-120688 : On WASI in debug mode, Python is now built with compiler flag -O3 instead of -Og , to support more recursive calls. Patch by Victor Stinner.

gh-118943 : Fix a possible race condition affecting parallel builds configured with --enable-experimental-jit , in which FileNotFoundError could be caused by another process already moving jit_stencils.h.new to jit_stencils.h .

gh-120326 : On Windows, fix build error when --disable-gil and --experimental-jit options are combined.

gh-120291 : Make the python-config shell script compatible with non-bash shells.

gh-113565 : Improve curses and curses.panel dependency checks in configure .

gh-119729 : On POSIX systems, the pkg-config ( .pc ) filenames now include the ABI flags, which may include debug (“d”) and free-threaded (“t”). For example: * python-3.14.pc (default, non-debug build) * python-3.14d.pc (default, debug build) * python-3.14t.pc (free-threaded build)

gh-115119 : Fall back to the bundled libmpdec if a system version cannot be found.

gh-119132 : Update sys.version to identify whether the build is default build or free-threading build. Patch By Donghee Na.

gh-118836 : Fix an AssertionError when building with --enable-experimental-jit and the compiler emits a SHT_NOTE section.

gh-118943 : Fix a possible race condition affecting parallel builds configured with --enable-experimental-jit , in which compilation errors could be caused by an incompletely-generated header file.

Python 3.13.0 beta 1 ¶

Release date: 2024-05-08

gh-116741 : Update bundled libexpat to 2.6.2

gh-117233 : Detect BLAKE2, SHA3, Shake, & truncated SHA512 support in the OpenSSL-ish libcrypto library at build time. This allows hashlib to be used with libraries that do not to support every algorithm that upstream OpenSSL does.

gh-118414 : Add instrumented opcodes to YIELD_VALUE assertion for tracing cases.

gh-117953 : When a builtin or extension module is imported for the first time, while a subinterpreter is active, the module’s init function is now run by the main interpreter first before import continues in the subinterpreter. Consequently, single-phase init modules now fail in an isolated subinterpreter without the init function running under that interpreter, whereas before it would run under the subinterpreter before failing, potentially leaving behind global state and callbacks and otherwise leaving the module in an inconsistent state.

gh-117549 : Don’t use designated initializer syntax in inline functions in internal headers. They cause problems for C++ or MSVC users who aren’t yet using the latest C++ standard (C++20). While internal, pycore_backoff.h, is included (indirectly, via pycore_code.h) by some key 3rd party software that does so for speed.

gh-95382 : Improve performance of json.dumps() and json.dump() when using the argument indent . Depending on the data the encoding using json.dumps() with indent can be up to 2 to 3 times faster.

gh-116322 : In --disable-gil builds, the GIL will be enabled while loading C extension modules. If the module indicates that it supports running without the GIL, the GIL will be disabled once loading is complete. Otherwise, the GIL will remain enabled for the remainder of the interpreter’s lifetime. This behavior does not apply if the GIL has been explicitly enabled or disabled with PYTHON_GIL or -Xgil .

gh-118513 : Fix incorrect UnboundLocalError when two comprehensions in the same function both reference the same name, and in one comprehension the name is bound while in the other it’s an implicit global.

gh-118518 : Allow the Linux perf support to work without frame pointers using perf’s advanced JIT support. The feature is activated when using the PYTHON_PERF_JIT_SUPPORT environment variable or when running Python with -Xperf_jit . Patch by Pablo Galindo.

gh-117514 : Add sys._is_gil_enabled() function that returns whether the GIL is currently enabled. In the default build it always returns True because the GIL is always enabled. In the free-threaded build, it may return True or False .

gh-118164 : Break a loop between the Python implementation of the decimal module and the Python code for integer to string conversion. Also optimize integer to string conversion for values in the range from 9_000 to 135_000 decimal digits.

gh-118473 : Fix sys.set_asyncgen_hooks() not to be partially set when raising TypeError .

gh-118465 : Compiler populates the new __firstlineno__ field on a class with the line number of the first line of the class definition.

gh-118492 : Fix an issue where the type cache can expose a previously accessed attribute when a finalizer is run.

gh-117714 : update async_generator.athrow().close() and async_generator.asend().close() to close their section of the underlying async generator

gh-111201 : The interactive interpreter is now implemented in Python, which allows for a number of new features like colors, multiline input, history viewing, and paste mode. Contributed by Pablo Galindo, Łukasz Langa and Lysandros Nikolaou based on code from the PyPy project.

gh-74929 : Implement PEP 667: converted FrameType.f_locals and PyFrame_GetLocals() to return a write-through proxy object when the frame refers to a function or comprehension.

gh-116767 : Fix crash in compiler on ‘async with’ that has many context managers.

gh-118335 : Change how to use the tier 2 interpreter. Instead of running Python with -X uops or setting the environment variable PYTHON_UOPS=1 , this choice is now made at build time by configuring with --enable-experimental-jit=interpreter .

Beware! This changes the environment variable to enable or disable micro-ops to PYTHON_JIT . The old PYTHON_UOPS is no longer used.

gh-118306 : Update JIT compilation to use LLVM 18

gh-118160 : Annotation scopes within classes can now contain comprehensions. However, such comprehensions are not inlined into their parent scope at runtime. Patch by Jelle Zijlstra.

gh-118272 : Fix bug where generator.close does not free the generator frame’s locals.

gh-118216 : Don’t consider __future__ imports with dots before the module name.

gh-118074 : Make sure that the Executor objects in the COLD_EXITS array aren’t assumed to be GC-able (which would access bytes outside the object).

gh-107674 : Lazy load frame line number to improve performance of tracing

gh-118082 : Improve SyntaxError message for imports without names, like in from x import and import cases. It now points out to users that import expects at least one name after it.

gh-118090 : Improve SyntaxError message for empty type param brackets.

gh-117958 : Added a get_jit_code() method to access JIT compiled machine code from the UOp Executor when the experimental JIT is enabled. Patch by Anthony Shaw.

gh-117901 : Add option for compiler’s codegen to save nested instruction sequences for introspection.

gh-116622 : Redirect stdout and stderr to system log when embedded in an Android app.

gh-109118 : annotation scope within class scopes can now contain lambdas.

gh-117894 : Prevent agen.aclose() objects being re-used after .throw() .

gh-117881 : prevent concurrent access to an async generator via athrow().throw() or asend().throw()

gh-117536 : Fix a RuntimeWarning when calling agen.aclose().throw(Exception) .

gh-117755 : Fix mimalloc allocator for huge memory allocation (around 8,589,934,592 GiB) on s390x. Patch by Victor Stinner.

gh-117750 : Fix issue where an object’s dict would get out of sync with the object’s internal values when being cleared. obj.__dict__.clear() now clears the internal values, but leaves the dict attached to the object.

gh-117431 : Improve the performance of the following bytes and bytearray methods by adapting them to the METH_FASTCALL calling convention:

gh-117709 : Speed up calls to str() with positional-only argument, by using the PEP 590 vectorcall calling convention. Patch by Erlend Aasland.

gh-117680 : Give _PyInstructionSequence a Python interface and use it in tests.

gh-115776 : Statically allocated objects are, by definition, immortal so must be marked as such regardless of whether they are in extension modules or not.

gh-117385 : Remove unhandled PY_MONITORING_EVENT_BRANCH and PY_MONITORING_EVENT_EXCEPTION_HANDLED events from sys.settrace() .

gh-116322 : Extension modules may indicate to the runtime that they can run without the GIL. Multi-phase init modules do so by calling providing Py_MOD_GIL_NOT_USED for the Py_mod_gil slot, while single-phase init modules call PyUnstable_Module_SetGIL(mod, Py_MOD_GIL_NOT_USED) from their init function.

gh-116129 : Implement PEP 696 , adding support for defaults on type parameters. Patch by Jelle Zijlstra.

gh-93502 : Add two new functions to the C-API, PyRefTracer_SetTracer() and PyRefTracer_GetTracer() , that allows to track object creation and destruction the same way the tracemalloc module does. Patch by Pablo Galindo

gh-107674 : Improved the performance of sys.settrace() significantly

gh-95754 : Improve the error message when a script shadowing a module from the standard library causes AttributeError to be raised. Similarly, improve the error message when a script shadowing a third party module attempts to access an attribute from that third party module while still initialising.

gh-99180 : Elide uninformative traceback indicators in return and simple assignment statements. Patch by Pablo Galindo.

gh-105879 : Allow the globals and locals arguments to exec() and eval() to be passed as keywords.

gh-118418 : A DeprecationWarning is now emitted if you fail to pass a value to the new type_params parameter of typing._eval_type() or typing.ForwardRef._evaluate() . (Using either of these private and undocumented functions is discouraged to begin with, but failing to pass a value to the type_params parameter may lead to incorrect behaviour on Python 3.12 or newer.)

gh-118660 : Add an optional second type parameter to typing.ContextManager and typing.AsyncContextManager , representing the return types of __exit__() and __aexit__() respectively. This parameter defaults to bool | None .

gh-118650 : The enum module allows method named _repr_* to be defined on Enum types.

gh-118648 : Add type parameter defaults to typing.Generator and typing.AsyncGenerator .

gh-101137 : Mime type text/x-rst is now supported by mimetypes .

gh-118164 : The Python implementation of the decimal module could appear to hang in relatively small power cases (like 2**117 ) if context precision was set to a very high value. A different method to check for exactly representable results is used now that doesn’t rely on computing 10**precision (which could be effectively too large to compute).

gh-111744 : breakpoint() and pdb.set_trace() now enter the debugger immediately after the call rather than before the next line is executed.

gh-118500 : Add pdb support for zipapps

gh-118406 : Add signature for sqlite3.Connection objects.

gh-101732 : Use a Y2038 compatible openssl time function when available.

gh-118404 : Fix inspect.signature() for non-comparable callables.

gh-118402 : Fix inspect.signature() for the result of the functools.cmp_to_key() call.

gh-116622 : On Android, sysconfig.get_platform now returns the format specified by PEP 738 .

gh-118285 : Allow to specify the signature of custom callable instances of extension type by the __text_signature__ attribute. Specify signatures of operator.attrgetter , operator.itemgetter , and operator.methodcaller instances.

gh-118314 : Fix an edge case in binascii.a2b_base64() strict mode, where excessive padding is not detected when no padding is necessary.

gh-118271 : Add the PhotoImage methods read() to read an image from a file and data() to get the image data. Add background and grayscale parameters to PhotoImage method write() .

gh-118225 : Add the PhotoImage method copy_replace() to copy a region from one image to other image, possibly with pixel zooming and/or subsampling. Add from_coords parameter to PhotoImage methods copy() , zoom() and subsample() . Add zoom and subsample parameters to PhotoImage method copy() .

gh-118221 : Fix a bug where sqlite3.Connection.iterdump() could fail if a custom row factory was used. Patch by Erlend Aasland.

gh-118013 : Fix regression introduced in gh-103193 that meant that calling inspect.getattr_static() on an instance would cause a strong reference to that instance’s class to persist in an internal cache in the inspect module. This caused unexpected memory consumption if the class was dynamically created, the class held strong references to other objects which took up a significant amount of memory, and the cache contained the sole strong reference to the class. The fix for the regression leads to a slowdown in getattr_static() , but the function should still be significantly faster than it was in Python 3.11. Patch by Alex Waygood.

gh-118218 : Speed up itertools.pairwise() in the common case by up to 1.8x.

gh-117486 : Improve the behavior of user-defined subclasses of ast.AST . Such classes will now require no changes in the usual case to conform with the behavior changes of the ast module in Python 3.13. Patch by Jelle Zijlstra.

gh-90848 : Fixed unittest.mock.create_autospec() to configure parent mock with keyword arguments.

gh-118168 : Fix incorrect argument substitution when typing.Unpack is used with the builtin tuple . typing.Unpack now raises TypeError when used with certain invalid types. Patch by Jelle Zijlstra.

gh-118131 : Add command-line interface for the random module. Patch by Hugo van Kemenade.

gh-118107 : Fix zipimport reading of ZIP64 files with file entries that are too big or offset too far.

gh-102511 : Fix os.path.normpath() for UNC paths on Windows. Speed up os.path.splitroot() with a native implementation.

gh-117535 : Change the unknown filename of warnings from sys to <sys> to clarify that it’s not a real filename.

gh-114053 : Fix erroneous NameError when calling typing.get_type_hints() on a class that made use of PEP 695 type parameters in a module that had from __future__ import annotations at the top of the file. Patch by Alex Waygood.

gh-116931 : Add parameter fileobj check for tarfile.TarFile.addfile()

gh-117995 : Don’t raise DeprecationWarning when a sequence of parameters is used to bind indexed, nameless placeholders. See also gh-100668 .

gh-80361 : Fix TypeError in email.message.Message.get_payload() when the charset is RFC 2231 encoded.

gh-86650 : Fix IndexError when parse some emails with invalid Message-ID (including one-off addresses generated by Microsoft Outlook).

gh-117691 : Improve the error messages emitted by tarfile deprecation warnings relating to PEP 706. If a filter argument is not provided to extract() or extractall , the deprecation warning now points to the line in the user’s code where the relevant function was called. Patch by Alex Waygood.

gh-115874 : Fixed a possible segfault during garbage collection of _asyncio.FutureIter objects. Patch by Savannah Ostrowski.

gh-115060 : Speed up pathlib.Path.glob() by omitting an initial is_dir() call. As a result of this change, glob() can no longer raise OSError .

gh-77102 : site module now parses .pth file with UTF-8 first, and locale encoding if UnicodeDecodeError happened. It supported only locale encoding before.

gh-76785 : We’ve exposed the low-level _interpreters module for the sake of the PyPI implementation of PEP 734 . It was sometimes available as the _xxsubinterpreters module and was formerly used only for testing. For the most part, it should be considered an internal module, like _thread and _imp . See https://discuss.python.org/t/pep-734-multiple-interpreters-in-the-stdlib/41147/26 .

gh-115060 : Speed up pathlib.Path.glob() by not scanning directories for non-wildcard pattern segments.

gh-117727 : Speed up pathlib.Path.iterdir() by using os.scandir() internally.

gh-117586 : Speed up pathlib.Path.walk() by working with strings internally.

gh-117722 : Change the new multi-separator support in asyncio.StreamReader.readuntil() to only accept tuples of separators rather than arbitrary iterables.

gh-117692 : Fixes a bug when doctest.DocTestFinder was failing on wrapped builtin_function_or_method .

gh-117348 : Largely restored import time performance of configparser by avoiding dataclasses.

gh-117641 : Speedup os.path.commonpath() on Unix.

gh-117663 : Fix _simple_enum to detect aliases when multiple arguments are present but only one is the member value.

gh-117636 : Speedup os.path.join() .

gh-117618 : Support package.module as filename for break command of pdb

gh-102247 : the status codes enum with constants in http.HTTPStatus are updated to include the names from RFC9110. This RFC includes some HTTP statuses previously only used for WEBDAV and assigns more generic names to them.

The old constants are preserved for backwards compatibility.

gh-117607 : Speedup os.path.relpath() .

gh-117586 : Speed up pathlib.Path.glob() by working with strings internally.

gh-117225 : Add colour to doctest output. Patch by Hugo van Kemenade.

gh-117566 : ipaddress.IPv6Address.is_loopback() will now return True for IPv4-mapped loopback addresses, i.e. addresses in the ::ffff:127.0.0.0/104 address space.

gh-117546 : Fix issue where os.path.realpath() stopped resolving symlinks after encountering a symlink loop on POSIX.

gh-116720 : Improved behavior of asyncio.TaskGroup when an external cancellation collides with an internal cancellation. For example, when two task groups are nested and both experience an exception in a child task simultaneously, it was possible that the outer task group would misbehave, because its internal cancellation was swallowed by the inner task group.

In the case where a task group is cancelled externally and also must raise an ExceptionGroup , it will now call the parent task’s cancel() method. This ensures that a asyncio.CancelledError will be raised at the next await , so the cancellation is not lost.

An added benefit of these changes is that task groups now preserve the cancellation count ( asyncio.Task.cancelling() ).

In order to handle some corner cases, asyncio.Task.uncancel() may now reset the undocumented _must_cancel flag when the cancellation count reaches zero.

gh-117516 : Add typing.TypeIs , implementing PEP 742 . Patch by Jelle Zijlstra.

gh-117503 : Fix support of non-ASCII user names in bytes paths in os.path.expanduser() on Posix.

gh-117394 : os.path.ismount() is now 2-3 times faster if the user has permissions.

gh-117313 : Only treat '\n' , '\r' and '\r\n' as line separators in re-folding the email messages. Preserve control characters '\v' , '\f' , '\x1c' , '\x1d' and '\x1e' and Unicode line separators '\x85' , '\u2028' and '\u2029' as is.

gh-117142 : Convert _ctypes to multi-phase initialisation ( PEP 489 ).

gh-66543 : Add the mimetypes.guess_file_type() function which works with file path. Passing file path instead of URL in guess_type() is soft deprecated .

gh-68583 : webbrowser CLI: replace getopt with argparse, add long options. Patch by Hugo van Kemenade.

gh-116871 : Name suggestions for AttributeError and ImportError now only include underscored names if the original name was underscored.

gh-116023 : Don’t show empty fields (value None or [] ) in ast.dump() by default. Add show_empty=False parameter to optionally show them.

gh-115961 : Added name and mode attributes for compressed and archived file-like objects in modules bz2 , lzma , tarfile and zipfile . The value of the mode attribute of gzip.GzipFile was changed from integer ( 1 or 2 ) to string ( 'rb' or 'wb' ). The value of the mode attribute of the readable file-like object returned by zipfile.ZipFile.open() was changed from 'r' to 'rb' .

gh-82062 : Fix inspect.signature() to correctly handle parameter defaults on methods in extension modules that use names defined in the module namespace.

gh-83856 : Honor atexit for all multiprocessing start methods

gh-113081 : Print colorized exception just like built-in traceback in pdb

gh-112855 : Speed up pickling of pathlib.PurePath objects. Patch by Barney Gale.

gh-111744 : Support opcode events in bdb

gh-109617 : ncurses : fixed a crash that could occur on macOS 13 or earlier when Python was built with Apple Xcode 15’s SDK.

gh-83151 : Enabled arbitrary statements and evaluations in pdb shell to access the local variables of the current frame, which made it possible for multi-scope code like generators or nested function to work.

gh-110209 : Add __class_getitem__() to types.GeneratorType and types.CoroutineType for type hinting purposes. Patch by James Hilton-Balfe.

gh-108191 : The types.SimpleNamespace now accepts an optional positional argument which specifies initial values of attributes as a dict or an iterable of key-value pairs.

gh-62090 : Fix assertion errors caused by whitespace in metavars or SUPPRESS -ed groups in argparse by simplifying usage formatting. Patch by Ali Hamdan.

gh-102402 : Adjust logging.LogRecord to use time.time_ns() and fix minor bug related to floating-point math.

gh-100242 : Bring pure Python implementation functools.partial.__new__ more in line with the C-implementation by not just always checking for the presence of the attribute 'func' on the first argument of partial . Instead, both the Python version and the C version perform an isinstance(func, partial) check on the first argument of partial .

gh-99730 : HEAD requests are no longer upgraded to GET request during redirects in urllib.

gh-66410 : Setting the tkinter module global wantobjects to 2 before creating the Tk object or call the wantobjects() method of the Tk object with argument 2 makes now arguments to callbacks registered in the tkinter module to be passed as various Python objects ( int , float , bytes , tuple ), depending on their internal representation in Tcl, instead of always str . tkinter.wantobjects is now set to 2 by default.

bpo-40943 : Fix several IndexError when parse emails with truncated Message-ID, address, routes, etc, e.g. example@ .

bpo-39324 : Add mime type mapping for .md <-> text/markdown

bpo-18108 : shutil.chown() now supports dir_fd and follow_symlinks keyword arguments.

bpo-30988 : Fix parsing of emails with invalid address headers having a leading or trailing dot. Patch by tsufeki.

bpo-32839 : Add the after_info() method for Tkinter widgets.

gh-117928 : The minimum Sphinx version required for the documentation is now 6.2.1.

gh-118734 : Fixes Windows build when invoked directly (not through the build.bat script) without specifying a value for UseTIER2 .

gh-115119 : The configure option --with-system-libmpdec now defaults to yes . The bundled copy of libmpdecimal will be removed in Python 3.15.

gh-117845 : Fix building against recent libedit versions by detecting readline hook signatures in configure .

gh-116622 : A testbed project was added to run the test suite on Android.

gh-117645 : Increase WASI stack size from 512 KiB to 8 MiB and the initial memory from 10 MiB to 20 MiB. Patch by Victor Stinner.

gh-115119 : configure now uses pkg-config to detect decimal dependencies if the --with-system-libmpdec option is given.

gh-115119 : Update Windows installer to use libmpdecimal 4.0.0.

gh-118486 : os.mkdir() now accepts mode of 0o700 to restrict the new directory to the current user.

gh-118347 : Fixes launcher updates not being installed.

gh-118293 : The multiprocessing module now passes the STARTF_FORCEOFFFEEDBACK flag when spawning processes to tell Windows not to change the mouse cursor.

gh-115009 : Update Windows installer to use SQLite 3.45.3.

gh-90329 : Suppress the warning displayed on virtual environment creation when the requested and created paths differ only by a short (8.3 style) name. Warnings will continue to be shown if a junction or symlink in the path caused the venv to be created in a different location than originally requested.

gh-117786 : Fixes virtual environments not correctly launching when created from a Store install.

gh-115119 : Update macOS installer to use libmpdecimal 4.0.0.

gh-114099 : iOS preprocessor symbol usage was made compatible with older macOS SDKs.

gh-115009 : Update macOS installer to use SQLite 3.45.3.

gh-91629 : Use ~/.config/fish/conf.d configs and fish_add_path to set PATH when installing for the Fish shell.

bpo-34774 : Use user-selected color theme for Help => IDLE Doc.

gh-118124 : Fix Py_BUILD_ASSERT and Py_BUILD_ASSERT_EXPR for non-constant expressions: use static_assert() on C11 and newer. Patch by Victor Stinner.

gh-110850 : Add “Raw” variant of PyTime functions

PyTime_MonotonicRaw()

PyTime_PerfCounterRaw()

PyTime_TimeRaw()

gh-117987 : Restore functions removed in Python 3.13 alpha 1:

Py_SetPythonHome()

Py_SetProgramName()

PySys_SetArgvEx()

PySys_SetArgv()

gh-117929 : Restore removed PyEval_InitThreads() function. Patch by Victor Stinner.

gh-117534 : Improve validation logic in the C implementation of datetime.datetime.fromisoformat() to better handle invalid years. Patch by Vlad Efanov.

gh-68114 : Fixed skipitem()’s handling of the old ‘w’ and ‘w#’ formatters. These are no longer supported and now raise an exception if used.

gh-111997 : Add a C-API for firing monitoring events.

Python 3.13.0 alpha 6 ¶

Release date: 2024-04-09

gh-117494 : Refactored the instruction sequence data structure out of compile.c into instruction_sequence.c.

gh-116968 : Introduce a unified 16-bit backoff counter type ( _Py_BackoffCounter ), shared between the Tier 1 adaptive specializer and the Tier 2 optimizer. The API used for adaptive specialization counters is changed but the behavior is (supposed to be) identical.

The behavior of the Tier 2 counters is changed:

There are no longer dynamic thresholds (we never varied these).

All counters now use the same exponential backoff.

The counter for JUMP_BACKWARD starts counting down from 16.

The temperature in side exits starts counting down from 64.

startswith()

gh-117431 : Improve the performance of the following str methods by adapting them to the METH_FASTCALL calling convention:

gh-117411 : Move PyFutureFeatures to an internal header and make it private.

gh-109120 : Added handle of incorrect star expressions, e.g f(3, *) . Patch by Grigoryev Semyon

gh-117266 : Fix crashes for certain user-created subclasses of ast.AST . Such classes are now expected to set the _field_types attribute.

gh-99108 : Updated the hashlib built-in HACL* project C code from upstream that we use for many implementations when they are not present via OpenSSL in a given build. This also avoids the rare potential for a C symbol name one definition rule linking issue.

gh-117108 : Change the old space bit of objects in the young generation from 0 to gcstate->visited, so that any objects created during GC will have the old bit set correctly if they get moved into the old generation.

gh-117108 : The cycle GC now chooses the size of increments based on the total heap size, instead of the rate of object creation. This ensures that it can keep up with growing heaps.

gh-116735 : For INSTRUMENTED_CALL_FUNCTION_EX , set arg0 to sys.monitoring.MISSING instead of None for CALL event.

gh-113964 : Starting new threads and process creation through os.fork() are now only prevented once all non-daemon threads exit.

gh-116626 : Ensure INSTRUMENTED_CALL_FUNCTION_EX always emits CALL

gh-116554 : list.sort() now exploits more cases of partial ordering, particularly those with long descending runs with sub-runs of equal values. Those are recognized as single runs now (previously, each block of repeated values caused a new run to be created).

gh-114099 : Added a Loader that can discover extension modules in an iOS-style Frameworks folder.

gh-115775 : Compiler populates the new __static_attributes__ field on a class with the names of attributes of this class which are accessed through self.X from any function in its body.

gh-115776 : The array of values, the PyDictValues struct is now embedded in the object during allocation. This provides better performance in the common case, and does not degrade as much when the object’s __dict__ is materialized.

gh-108362 : Implement an incremental cyclic garbage collector. By collecting the old generation in increments, there is no need for a full heap scan. This can hugely reduce maximum pause time for programs with large heaps.

Reduce the number of generations from three to two. The old generation is split into two spaces, “visited” and “pending”.

Collection happens in two steps:: * An increment is formed from the young generation and a small part of the pending space. * This increment is scanned and the survivors moved to the end of the visited space.

When the collecting space becomes empty, the two spaces are swapped.

gh-109870 : Dataclasses now calls exec() once per dataclass, instead of once per method being added. This can speed up dataclass creation by up to 20%.

gh-97901 : Mime type text/rtf is now supported by mimetypes .

bpo-24612 : Improve the SyntaxError that happens when ‘not’ appears after an operator. Patch by Pablo Galindo

gh-117648 : Improve performance of os.path.join() and os.path.expanduser() .

gh-117584 : Raise TypeError for non-paths in posixpath.relpath() .

gh-117467 : Preserve mailbox ownership when rewriting in mailbox.mbox.flush() . Patch by Tony Mountifield.

gh-114848 : Raise FileNotFoundError when getcwd() returns ‘(unreachable)’, which can happen on Linux >= 2.6.36 with glibc < 2.27.

gh-117459 : asyncio.asyncio.run_coroutine_threadsafe() now keeps the traceback of CancelledError , TimeoutError and InvalidStateError which are raised in the coroutine.

gh-117381 : Fix error message for ntpath.commonpath() .

gh-117337 : Deprecate undocumented glob.glob0() and glob.glob1() functions. Use glob.glob() and pass a directory to its root_dir argument instead.

gh-117349 : Optimise several functions in os.path .

gh-117348 : Refactored configparser.RawConfigParser._read() to reduce cyclometric complexity and improve comprehensibility.

gh-117335 : Raise TypeError for non-sequences for ntpath.commonpath() .

gh-66449 : configparser.ConfigParser now accepts unnamed sections before named ones, if configured to do so.

gh-88014 : In documentation of gzip.GzipFile in module gzip, explain data type of optional constructor argument mtime , and recommend mtime = 0 for generating deterministic streams.

gh-117310 : Fixed an unlikely early & extra Py_DECREF triggered crash in ssl when creating a new _ssl._SSLContext if CPython was built implausibly such that the default cipher list is empty or the SSL library it was linked against reports a failure from its C SSL_CTX_set_cipher_list() API.

gh-117294 : A DocTestCase now reports as skipped if all examples in the doctest are skipped.

gh-98966 : In subprocess , raise a more informative message when stdout=STDOUT .

gh-117225 : doctest: only print “and X failed” when non-zero, don’t pluralise “1 items”. Patch by Hugo van Kemenade.

gh-117205 : Speed up compileall.compile_dir() by 20% when using multiprocessing by increasing chunksize .

gh-117178 : Fix regression in lazy loading of self-referential modules, introduced in gh-114781 .

gh-112383 : Fix dis module’s handling of ENTER_EXECUTOR instructions.

gh-117182 : Lazy-loading of modules that modify their own __class__ no longer reverts the __class__ to types.ModuleType .

gh-117084 : Fix zipfile extraction for directory entries with the name containing backslashes on Windows.

gh-117114 : Make os.path.isdevdrive() available on all platforms. For those that do not offer Dev Drives, it will always return False .

gh-117110 : Fix a bug that prevents subclasses of typing.Any to be instantiated with arguments. Patch by Chris Fu.

gh-109653 : Deferred select imports in importlib.metadata and importlib.resources for a 14% speedup.

gh-70647 : Start the deprecation period for the current behavior of datetime.datetime.strptime() and time.strptime() which always fails to parse a date string with a ValueError involving a day of month such as strptime("02-29", "%m-%d") when a year is not specified and the date happen to be February 29th. This should help avoid users finding new bugs every four years due to a natural mistaken assumption about the API when parsing partial date values.

gh-116987 : Fixed inspect.findsource() for class code objects.

gh-114099 : Modify standard library to allow for iOS platform differences.

gh-90872 : On Windows, subprocess.Popen.wait() no longer calls WaitForSingleObject() with a negative timeout: pass 0 ms if the timeout is negative. Patch by Victor Stinner.

gh-116957 : configparser: Don’t leave ConfigParser values in an invalid state (stored as a list instead of a str) after an earlier read raised DuplicateSectionError or DuplicateOptionError.

gh-115538 : _io.WindowsConsoleIO now emit a warning if a boolean value is passed as a filedescriptor argument.

gh-90095 : Ignore empty lines and comments in .pdbrc

gh-106531 : Refreshed zipfile._path from zipp 3.18 , providing better compatibility for PyPy, better glob performance for deeply nested zipfiles, and providing internal access to CompleteDirs.inject for use in other tests (like importlib.resources).

gh-63207 : On Windows, time.time() now uses the GetSystemTimePreciseAsFileTime() clock to have a resolution better than 1 us, instead of the GetSystemTimeAsFileTime() clock which has a resolution of 15.6 ms. Patch by Victor Stinner.

gh-116764 : Restore support of None and other false values in urllib.parse functions parse_qs() and parse_qsl() . Also, they now raise a TypeError for non-zero integers and non-empty sequences.

gh-116811 : In PathFinder.invalidate_caches , delegate to MetadataPathFinder.invalidate_caches .

gh-116647 : Fix recursive child in dataclasses

gh-113171 : Fixed various false positives and false negatives in

ipaddress.IPv4Address.is_private (see these docs for details)

ipaddress.IPv4Address.is_global

ipaddress.IPv6Address.is_private

ipaddress.IPv6Address.is_global

Also in the corresponding ipaddress.IPv4Network and ipaddress.IPv6Network attributes.

gh-63283 : In encodings.idna , any capitalization of the the ACE prefix ( xn-- ) is now acceptable. Patch by Pepijn de Vos and Zackery Spytz.

gh-71042 : Add platform.android_ver() , which provides device and OS information on Android.

gh-73468 : Added new math.fma() function, wrapping C99’s fma() operation: fused multiply-add function. Patch by Mark Dickinson and Victor Stinner.

gh-116608 : The importlib.resources functions is_resource() , open_binary() , open_text() , path() , read_binary() , and read_text() are un-deprecated, and support subdirectories via multiple positional arguments. The contents() function also allows subdirectories, but remains deprecated.

gh-116484 : Change automatically generated tkinter.Checkbutton widget names to avoid collisions with automatically generated tkinter.ttk.Checkbutton widget names within the same parent widget.

gh-114314 : In ctypes , ctype data is now stored in type objects directly rather than in a dict subclass. This is an internal change that should not affect usage.

gh-116401 : Fix blocking os.fwalk() and shutil.rmtree() on opening named pipe.

gh-71052 : Implement ctypes.util.find_library() on Android.

gh-90535 : Fix support of interval values > 1 in logging.TimedRotatingFileHandler for when='MIDNIGHT' and when='Wx' .

gh-113308 : Remove some internal protected parts from uuid : _has_uuid_generate_time_safe , _netbios_getnode , _ipconfig_getnode , and _load_system_functions . They were unused.

gh-115627 : Fix the ssl module error handling of connection terminate by peer. It now throws an OSError with the appropriate error code instead of an EOFError.

gh-114847 : Speed up os.path.realpath() on non-Windows platforms.

gh-114271 : Fix a race in threading.Thread.join() .

threading._MainThread now always represents the main thread of the main interpreter.

PyThreadState.on_delete and PyThreadState.on_delete_data have been removed.

gh-113538 : Add asyncio.Server.close_clients() and asyncio.Server.abort_clients() methods which allow to more forcefully close an asyncio server.

gh-85287 : Changes Unicode codecs to return UnicodeEncodeError or UnicodeDecodeError, rather than just UnicodeError.

gh-113548 : pdb now allows CLI arguments to pdb -m .

gh-112948 : Make completion of pdb similar to Python REPL

gh-105866 : Fixed _get_slots bug which caused error when defining dataclasses with slots and a weakref_slot.

gh-96471 : Add asyncio.Queue termination with shutdown() method.

gh-89739 : The zipimport module can now read ZIP64 files.

bpo-33533 : asyncio.as_completed() now returns an object that is both an asynchronous iterator and plain iterator. The new asynchronous iteration pattern allows for easier correlation between prior tasks and their completed results. This is a closer match to concurrent.futures.as_completed() ’s iteration pattern. Patch by Justin Arthur.

bpo-27578 : inspect.getsource() (and related functions) work with empty module files, returning '\n' (or reasonable equivalent) instead of raising OSError . Patch by Kernc.

bpo-37141 : Accept an iterable of separators in asyncio.StreamReader.readuntil() , stopping when one of them is encountered.

gh-66543 : Make mimetypes.guess_type() properly parsing of URLs with only a host name, URLs containing fragment or query, and filenames with only a UNC sharepoint on Windows. Based on patch by Dong-hee Na.

bpo-15010 : unittest.TestLoader.discover() now saves the original value of unittest.TestLoader._top_level_dir and restores it at the end of the call.

gh-115977 : Remove compatibility references to Emscripten.

gh-114099 : Add an iOS platform guide, and flag modules not available on iOS.

gh-91565 : Changes to documentation files and config outputs to reflect the new location for reporting bugs - i.e. GitHub rather than bugs.python.org.

gh-83434 : Disable JUnit XML output ( --junit-xml=FILE command line option) in regrtest when hunting for reference leaks ( -R option). Patch by Victor Stinner.

gh-117187 : Fix XML tests for vanilla Expat <2.6.0.

gh-116333 : Tests of TLS related things (error codes, etc) were updated to be more lenient about specific error message strings and behaviors as seen in the BoringSSL and AWS-LC forks of OpenSSL.

gh-117089 : Consolidated tests for importlib.metadata in their own metadata package.

gh-115979 : Update test_importlib so that it passes under WASI SDK 21.

gh-112536 : Add –tsan to test.regrtest for running TSAN tests in reasonable execution times. Patch by Donghee Na.

gh-116307 : Added import helper isolated_modules as CleanImport does not remove modules imported during the context. Use it in importlib.resources tests to avoid leaving mod around to impede importlib.metadata tests.

gh-114736 : Have WASI builds use WASI SDK 21.

gh-115983 : Skip building test modules that must be built as shared under WASI.

gh-71052 : Add Android build script and instructions.

gh-117267 : Ensure DirEntry.stat().st_ctime behaves consistently with os.stat() during the deprecation period of st_ctime by containing the same value as st_birthtime . After the deprecation period, st_ctime will be the metadata change time (or unavailable through DirEntry ), and only st_birthtime will contain the creation time.

gh-116195 : Improves performance of os.getppid() by using an alternate system API when available. Contributed by vxiiduu.

gh-88494 : On Windows, time.monotonic() now uses the QueryPerformanceCounter() clock to have a resolution better than 1 us, instead of the GetTickCount64() clock which has a resolution of 15.6 ms. Patch by Victor Stinner.

gh-116773 : Fix instances of <_overlapped.Overlapped object at 0xXXX> still has pending operation at deallocation, the process may crash .

gh-91227 : Fix the asyncio ProactorEventLoop implementation so that sending a datagram to an address that is not listening does not prevent receiving any more datagrams.

gh-115119 : Switched from vendored libmpdecimal code to a separately-hosted external package in the cpython-source-deps repository when building the _decimal module.

gh-117642 : Fix PEP 737 implementation for %#T and %#N .

gh-87193 : _PyBytes_Resize() can now be called for bytes objects with reference count > 1, including 1-byte bytes objects. It creates a new bytes object and destroys the old one if it has reference count > 1.

gh-117021 : Fix integer overflow in PyLong_AsPid() on non-Windows 64-bit platforms.

gh-115756 : PyCode_GetFirstFree() is an ustable API now and has been renamed to PyUnstable_Code_GetFirstFree() . (Contributed by Bogdan Romanyuk in gh-115781 )

gh-116869 : Add test_cext test: build a C extension to check if the Python C API emits C compiler warnings. Patch by Victor Stinner.

gh-116869 : Make the C API compatible with -Werror=declaration-after-statement compiler flag again. Patch by Victor Stinner.

gh-116936 : Add PyType_GetModuleByDef() to the limited C API. Patch by Victor Stinner.

gh-116809 : Restore removed private _PyErr_ChainExceptions1() function. Patch by Victor Stinner.

gh-115754 : In the limited C API version 3.13, getting Py_None , Py_False , Py_True , Py_Ellipsis and Py_NotImplemented singletons is now implemented as function calls at the stable ABI level to hide implementation details. Getting these constants still return borrowed references. Patch by Victor Stinner.

gh-115754 : Add Py_GetConstant() and Py_GetConstantBorrowed() functions to get constants. For example, Py_GetConstant(Py_CONSTANT_ZERO) returns a strong reference to the constant zero. Patch by Victor Stinner.

gh-111696 : Add support for %T , %T# , %N and %N# formats to PyUnicode_FromFormat() : format the fully qualified name of an object type and of a type: call PyType_GetModuleName() . See PEP 737 for more information. Patch by Victor Stinner.

gh-111696 : Add PyType_GetModuleName() function to get the type’s module name. Equivalent to getting the type.__module__ attribute. Patch by Eric Snow and Victor Stinner.

gh-111696 : Add PyType_GetFullyQualifiedName() function to get the type’s fully qualified name. Equivalent to f"{type.__module__}.{type.__qualname__}" , or type.__qualname__ if type.__module__ is not a string or is equal to "builtins" . Patch by Victor Stinner.

gh-85283 : The fcntl , grp , pwd , termios , _statistics and _testconsole C extensions are now built with the limited C API . Patch by Victor Stinner.

gh-111140 : Add additional flags to PyLong_AsNativeBytes() and PyLong_FromNativeBytes() to allow the caller to determine how to handle edge cases around values that fill the entire buffer.

gh-113024 : Add PyObject_GenericHash() function.

Python 3.13.0 alpha 5 ¶

Release date: 2024-03-12

gh-115398 : Allow controlling Expat >=2.6.0 reparse deferral ( CVE-2023-52425 ) by adding five new methods:

xml.etree.ElementTree.XMLParser.flush()

xml.etree.ElementTree.XMLPullParser.flush()

xml.parsers.expat.xmlparser.GetReparseDeferralEnabled()

xml.parsers.expat.xmlparser.SetReparseDeferralEnabled()

xml.sax.expatreader.ExpatParser.flush()

gh-114572 : ssl.SSLContext.cert_store_stats() and ssl.SSLContext.get_ca_certs() now correctly lock access to the certificate store, when the ssl.SSLContext is shared across multiple threads.

gh-116604 : Respect the status of the garbage collector when indirect calls are made via PyErr_CheckSignals() and the evaluation breaker. Patch by Pablo Galindo

gh-112087 : list is now compatible with the implementation of PEP 703 .

gh-116381 : Add specialization for CONTAINS_OP .

gh-116296 : Fix possible refleak in object.__reduce__() internal error handling.

gh-115823 : Properly calculate error ranges in the parser when raising SyntaxError exceptions caused by invalid byte sequences. Patch by Pablo Galindo

gh-115778 : Add tierN annotation for instruction definition in interpreter DSL.

gh-115733 : Fix crash when calling next() on exhausted list iterators.

gh-115700 : The regen-cases build stage now works on Windows.

gh-115347 : Fix bug where docstring was replaced by a redundant NOP when Python is run with -OO .

gh-115323 : Make error message more meaningful for when bytearray.extend() is called with a str object.

gh-112175 : Every PyThreadState now has its own eval_breaker , allowing specific threads to be interrupted.

gh-115154 : Fix a bug that was causing the tokenize.untokenize() function to handle unicode named literals incorrectly. Patch by Pablo Galindo

gh-112433 : Add ability to force alignment of ctypes.Structure by way of the new _align_ attribute on the class.

gh-104090 : The multiprocessing resource tracker now exits with non-zero status code if a resource leak was detected. It still exits with status code 0 otherwise.

gh-105858 : Improve the constructors for ast nodes. Arguments of list types now default to an empty list if omitted, and optional fields default to None . AST nodes now have an __annotations__ attribute with the expected types of their attributes. Passing unrecognized extra arguments to AST nodes is deprecated and will become an error in Python 3.15. Omitting a required argument to an AST node is deprecated and will become an error in Python 3.15. Patch by Jelle Zijlstra.

gh-101860 : Expose __name__ attribute on property.

gh-96497 : Fix incorrect resolution of mangled class variables used in assignment expressions in comprehensions.

gh-116600 : Fix repr() for global Flag members.

gh-116349 : platform.java_ver() is deprecated and will be removed in 3.15. It was largely untested, had a confusing API, and was only useful for Jython support.

gh-116143 : Fix a race in pydoc _start_server , eliminating a window in which _start_server can return a thread that is “serving” but without a docserver set.

gh-116127 : typing : implement PEP 705 which adds typing.ReadOnly support to typing.TypedDict .

gh-116325 : typing : raise SyntaxError instead of AttributeError on forward references as empty strings.

gh-115957 : When asyncio.TaskGroup.create_task is called on an inactive asyncio.TaskGroup , the given coroutine will be closed (which prevents a RuntimeWarning ).

gh-115978 : Disable preadv(), readv(), pwritev(), and writev() on WASI.

Under wasmtime for WASI 0.2, these functions don’t pass test_posix ( https://github.com/bytecodealliance/wasmtime/issues/7830 ).

gh-88352 : Fix the computation of the next rollover time in the logging.TimedRotatingFileHandler handler. computeRollover() now always returns a timestamp larger than the specified time and works correctly during the DST change. doRollover() no longer overwrite the already rolled over file, saving from data loss when run at midnight or during repeated time at the DST change.

gh-87115 : Set __main__.__spec__ to None when running a script with pdb

gh-76511 : Fix UnicodeEncodeError in email.Message.as_string() that results when a message that claims to be in the ascii character set actually has non-ascii characters. Non-ascii characters are now replaced with the U+FFFD replacement character, like in the replace error handler.

gh-89547 : Add support for nested typing special forms like Final[ClassVar[int]].

gh-65824 : Improve the less prompt in pydoc .

gh-116040 : [Enum] fix by-value calls when second value is falsey; e.g. Cardinal(1, 0)

gh-115821 : [Enum] Improve error message when calling super().__new__() in custom __new__.

gh-85644 : Use the XDG_CURRENT_DESKTOP environment variable in webbrowser to check desktop. Prefer it to the deprecated GNOME_DESKTOP_SESSION_ID for GNOME detection.

gh-75988 : Fixed unittest.mock.create_autospec() to pass the call through to the wrapped object to return the real result.

gh-115881 : Fix issue where ast.parse() would incorrectly flag conditional context managers (such as with (x() if y else z()): ... ) as invalid syntax if feature_version=(3, 8) was passed. This reverts changes to the grammar made as part of gh-94949 .

gh-115886 : Fix silent truncation of the name with an embedded null character in multiprocessing.shared_memory.SharedMemory .

gh-115532 : Add kernel density estimation to the statistics module.

gh-115714 : On WASI, the time module no longer get process time using times() or CLOCK_PROCESS_CPUTIME_ID , system API is that is unreliable and is likely to be removed from WASI. The affected clock functions fall back to calling clock() .

gh-115809 : Improve algorithm for computing which rolled-over log files to delete in logging.TimedRotatingFileHandler . It is now reliable for handlers without namer and with arbitrary deterministic namer that leaves the datetime part in the file name unmodified.

gh-74668 : urllib.parse functions parse_qs() and parse_qsl() now support bytes arguments containing raw and percent-encoded non-ASCII data.

gh-67044 : csv.writer() now always quotes or escapes '\r' and '\n' , regardless of lineterminator value.

gh-115712 : Restore support of space delimiter with skipinitialspace=True in csv . csv.writer() now quotes empty fields if delimiter is a space and skipinitialspace is true and raises exception if quoting is not possible.

gh-112364 : Fixed ast.unparse() to handle format_spec with " , ' or \\ . Patched by Frank Hoffmann.

gh-112997 : Stop logging potentially sensitive callback arguments in asyncio unless debug mode is active.

gh-114914 : Fix an issue where an abandoned StreamWriter would not be garbage collected.

gh-111358 : Fix a bug in asyncio.BaseEventLoop.shutdown_default_executor() to ensure the timeout passed to the coroutine behaves as expected.

gh-115618 : Fix improper decreasing the reference count for None argument in property methods getter() , setter() and deleter() .

gh-112720 : Refactor dis.ArgResolver to make it possible to subclass and change the way jump args are interpreted.

gh-112006 : Fix inspect.unwrap() for types with the __wrapper__ data descriptor. Fix inspect.Signature.from_callable() for builtins classmethod() and staticmethod() .

gh-101293 : Support callables with the __call__() method and types with __new__() and __init__() methods set to class methods, static methods, bound methods, partial functions, and other types of methods and descriptors in inspect.Signature.from_callable() .

gh-103092 : Isolate _lsprof (apply PEP 687 ).

gh-113942 : pydoc no longer skips global functions implemented as builtin methods, such as MethodDescriptorType and WrapperDescriptorType .

gh-115256 : Added DeprecationWarning when accessing the tarfile attribute of TarInfo objects. The attribute is never used internally and is only attached to TarInfos when the tarfile is opened in write-mode, not read-mode. The attribute creates an unnecessary reference cycle which may cause corruption when not closing the handle after writing a tarfile.

gh-115197 : urllib.request no longer resolves the hostname before checking it against the system’s proxy bypass list on macOS and Windows.

gh-113812 : DatagramTransport.sendto() will now send zero-length datagrams if called with an empty bytes object. The transport flow control also now accounts for the datagram header when calculating the buffer size.

gh-114763 : Protect modules loaded with importlib.util.LazyLoader from race conditions when multiple threads try to access attributes before the loading is complete.

gh-114709 : posixpath.commonpath() now raises a ValueError exception when passed an empty iterable. Previously, IndexError was raised.

posixpath.commonpath() now raises a TypeError exception when passed None . Previously, ValueError was raised.

gh-114610 : Fix bug where pathlib.PurePath.with_stem() converted a non-empty path suffix to a stem when given an empty stem argument. It now raises ValueError , just like pathlib.PurePath.with_suffix() does when called on a path with an empty stem, given a non-empty suffix argument.

gh-107361 : Add ssl.VERIFY_X509_PARTIAL_CHAIN and VERIFY_X509_STRICT to the default SSL context created with ssl.create_default_context() .

gh-112281 : Allow creating union of types for typing.Annotated with unhashable metadata.

gh-111775 : Fix importlib.resources.simple.ResourceHandle.open() for text mode, added missed stream argument.

gh-90095 : Make .pdbrc and -c work with any valid pdb commands.

gh-107625 : Raise configparser.ParsingError from read() and read_file() methods of configparser.ConfigParser if a key without a corresponding value is continued (that is, followed by an indented line).

gh-107155 : Fix incorrect output of help(x) where x is a lambda function, which has an __annotations__ dictionary attribute with a "return" key.

gh-57141 : Add option for non-shallow comparisons to filecmp.dircmp like filecmp.cmp() . Original patch by Steven Ward. Enhanced by Tobias Rautenkranz

gh-69990 : Profile.print_stats() has been improved to accept multiple sort arguments. Patched by Chiu-Hsiang Hsu and Furkan Onder.

gh-104061 : Add socket.SO_BINDTOIFINDEX constant.

gh-60346 : Fix ArgumentParser inconsistent with parse_known_args.

gh-102389 : Add windows_31j to aliases for cp932 codec

gh-72249 : functools.partial`s of :func:`repr() has been improved to include the module name. Patched by Furkan Onder and Anilyka Barry.

gh-100985 : Update HTTPSConnection to consistently wrap IPv6 Addresses when using a proxy.

gh-100884 : email: fix misfolding of comma in address-lists over multiple lines in combination with unicode encoding.

gh-95782 : Fix io.BufferedReader.tell() , io.BufferedReader.seek() , _pyio.BufferedReader.tell() , io.BufferedRandom.tell() , io.BufferedRandom.seek() and _pyio.BufferedRandom.tell() being able to return negative offsets.

gh-96310 : Fix a traceback in argparse when all options in a mutually exclusive group are suppressed.

gh-93205 : Fixed a bug in logging.handlers.TimedRotatingFileHandler where multiple rotating handler instances pointing to files with the same name but different extensions would conflict and not delete the correct files.

bpo-31116 : Add Z85 encoding to base64 .

bpo-44865 : Add missing call to localization function in argparse .

bpo-43952 : Fix multiprocessing.connection.Listener.accept() to accept empty bytes as authkey. Not accepting empty bytes as key causes it to hang indefinitely.

bpo-42125 : linecache: get module name from __spec__ if available. This allows getting source code for the __main__ module when a custom loader is used.

bpo-41122 : Failing to pass arguments properly to functools.singledispatchmethod() now throws a TypeError instead of hitting an index out of bounds internally.

bpo-40818 : The asyncio REPL now runs sys.__interactivehook__ on startup. The default implementation of sys.__interactivehook__ provides auto-completion to the asyncio REPL. Patch contributed by Rémi Lapeyre.

bpo-33775 : Add ‘default’ and ‘version’ help text for localization in argparse.

gh-115399 : Document CVE-2023-52425 of Expat <2.6.0 under “XML vulnerabilities”.

gh-109653 : Improve import time of uuid on Linux.

gh-71052 : Add test exclusions to support running the test suite on Android.

gh-71052 : Enable test_concurrent_futures on platforms that support threading but not multiprocessing.

gh-115796 : Make ‘_testinternalcapi.assemble_code_object’ construct the exception table for the code object.

gh-115720 : Leak tests ( -R , --huntrleaks ) now show a summary of the number of leaks found in each iteration.

gh-115122 : Add --bisect option to regrtest test runner: run failed tests with test.bisect_cmd to identify failing tests. Patch by Victor Stinner.

gh-115596 : Fix ProgramPriorityTests in test_os permanently changing the process priority.

gh-115556 : On Windows, commas passed in arguments to Tools\buildbot\test.bat and PCbuild\\rt.bat are now properly handled.

gh-115420 : Fix translation of exception handler targets by _testinternalcapi.optimize_cfg .

gh-115376 : Fix segfault in _testinternalcapi.compiler_codegen on bad input.

gh-116313 : Get WASI builds to work under wasmtime 18 w/ WASI 0.2/preview2 primitives.

gh-71052 : Change Android’s sys.platform from "linux" to "android" .

gh-116117 : Backport libb2 ’s PR #42 to fix compiling CPython on 32-bit Windows with clang-cl .

gh-71052 : Fix several Android build issues

gh-114099 : A testbed project was added to run the test suite on iOS.

gh-115350 : Fix building ctypes module with -DWIN32_LEAN_AND_MEAN defined

gh-111225 : Link extension modules against libpython on Android.

gh-115737 : The install name for libPython is now correctly set for non-framework macOS builds.

gh-114099 : Makefile targets were added to support compiling an iOS-compatible framework build.

gh-116012 : Ensure the value of GetLastError() is preserved across GIL operations.

gh-115582 : Building extensions intended for free-threaded builds of CPython now require compiling with /DPy_GIL_DISABLED manually when using a regular install. This is expected to change in future releases.

gh-115554 : The installer now has more strict rules about updating the Python Launcher for Windows . In general, most users only have a single launcher installed and will see no difference. When multiple launchers have been installed, the option to install the launcher is disabled until all but one have been removed. Downgrading the launcher (which was never allowed) is now more obviously blocked.

gh-115543 : Python Launcher for Windows can now detect Python 3.13 when installed from the Microsoft Store, and will install Python 3.12 by default when PYLAUNCHER_ALLOW_INSTALL is set.

gh-116145 : Update macOS installer to Tcl/Tk 8.6.14.

gh-88516 : On macOS show a proxy icon in the title bar of editor windows to match platform behaviour.

gh-100176 : Remove outdated Tools/{io,cc,string}bench

bpo-45101 : Add consistency in usage message IO between 2 versions of python-config.

gh-114626 : Add again _PyCFunctionFastWithKeywords name, removed in Python 3.13 alpha 4 by mistake. Keep the old private _PyCFunctionFastWithKeywords name (Python 3.7) as an alias to the new public name PyCFunctionFastWithKeywords (Python 3.13a4). Patch by Victor Stinner.

gh-111418 : Add PyHASH_MODULUS , PyHASH_BITS , PyHASH_INF and PyHASH_IMAG C macros. Patch by Sergey B Kirpichev.

Python 3.13.0 alpha 4 ¶

Release date: 2024-02-15

gh-115399 : Update bundled libexpat to 2.6.0

gh-115243 : Fix possible crashes in collections.deque.index() when the deque is concurrently modified.

gh-112087 : For an empty reverse iterator for list will be reduced to reversed() . Patch by Donghee Na

gh-114570 : Add PythonFinalizationError exception. This exception derived from RuntimeError is raised when an operation is blocked during the Python finalization . Patch by Victor Stinner.

gh-114695 : Add sys._clear_internal_caches() , which clears all internal performance-related caches (and deprecate the less-general sys._clear_type_cache() function).

gh-114828 : Fix compilation crashes in uncommon code examples using super() inside a comprehension in a class body.

gh-112069 : Adapt set and frozenset methods to Argument Clinic.

gh-115011 : Setters for members with an unsigned integer type now support the same range of valid values for objects that has a __index__() method as for int .

gh-114887 : Changed socket type validation in create_datagram_endpoint() to accept all non-stream sockets. This fixes a regression in compatibility with raw sockets.

gh-114944 : Fixes a race between PyParkingLot_Park and _PyParkingLot_UnparkAll .

gh-113462 : Limit the number of versions that a single class can use. Prevents a few wayward classes using up all the version numbers.

gh-76763 : The chr() builtin function now always raises ValueError for values outside the valid range. Previously it raised OverflowError for very large or small values.

gh-114806 : No longer specialize calls to classes, if those classes have metaclasses. Fixes bug where the __call__ method of the metaclass was not being called.

gh-107944 : Improve error message for function calls with bad keyword arguments via getargs

gh-112529 : The free-threaded build no longer allocates space for the PyGC_Head structure in objects that support cyclic garbage collection. A number of other fields and data structures are used as replacements, including ob_gc_bits , ob_tid , and mimalloc internal data structures.

gh-114456 : Lower the recursion limit under a debug build of WASI.

gh-114083 : Compiler applies folding of LOAD_CONST with following instruction in a separate pass before other optimisations. This enables jump threading in certain circumstances.

gh-114388 : Fix a RuntimeWarning emitted when assign an integer-like value that is not an instance of int to an attribute that corresponds to a C struct member of type T_UINT and T_ULONG. Fix a double RuntimeWarning emitted when assign a negative integer value to an attribute that corresponds to a C struct member of type T_UINT.

gh-114265 : Compiler propagates line numbers before optimization, leading to more optimization opportunities and removing the need for the guarantee_lineno_for_exits hack.

gh-112529 : The free-threaded build now has its own thread-safe GC implementation that uses mimalloc to find GC tracked objects. It is non-generational, unlike the existing GC implementation.

gh-114050 : Fix segmentation fault caused by an incorrect format string in TypeError exception when more than two arguments are passed to int .

gh-112354 : The END_FOR instruction now pops only one value. This is to better support side exits in loops.

gh-113884 : Make queue.SimpleQueue thread safe when the GIL is disabled.

gh-114058 : Implement the foundations of the Tier 2 redundancy eliminator.

gh-113939 : frame.clear(): Clear frame.f_locals as well, and not only the fast locals. This is relevant once frame.f_locals was accessed, which would contain also references to all the locals.

gh-112050 : Convert collections.deque to use Argument Clinic.

gh-112050 : Make methods on collections.deque thread-safe when the GIL is disabled.

gh-113464 : Add an option ( --enable-experimental-jit for configure -based builds or --experimental-jit for PCbuild -based ones) to build an experimental just-in-time compiler, based on copy-and-patch

gh-113055 : Make interp->obmalloc a pointer. For interpreters that share state with the main interpreter, this points to the same static memory structure. For interpreters with their own obmalloc state, it is heap allocated. Add free_obmalloc_arenas() which will free the obmalloc arenas and radix tree structures for interpreters with their own obmalloc state.

gh-55664 : Add warning when creating type using a namespace dictionary with non-string keys. Patched by Daniel Urban and Furkan Onder.

gh-104530 : Use native Win32 condition variables.

gh-115392 : Fix a bug in doctest where incorrect line numbers would be reported for decorated functions.

gh-114563 : Fix several format() bugs when using the C implementation of Decimal : * memory leak in some rare cases when using the z format option (coerce negative 0) * incorrect output when applying the z format option to type F (fixed-point with capital NAN / INF ) * incorrect output when applying the # format option (alternate form)

gh-102840 : Fix confused traceback when floordiv, mod, or divmod operations happens between instances of fractions.Fraction and complex .

gh-115165 : Most exceptions are now ignored when attempting to set the __orig_class__ attribute on objects returned when calling typing generic aliases (including generic aliases created using typing.Annotated ). Previously only AttributeError was ignored. Patch by Dave Shawley.

gh-112903 : Fix “issubclass() arg 1 must be a class” errors in certain cases of multiple inheritance with generic aliases (regression in early 3.13 alpha releases).

gh-115133 : Fix tests for XMLPullParser with Expat 2.6.0.

gh-115059 : io.BufferedRandom.read1() now flushes the underlying write buffer.

gh-79382 : Trailing ** no longer allows to match files and non-existing paths in recursive glob() .

gh-67837 : Avoid race conditions in the creation of directories during concurrent extraction in tarfile and zipfile .

gh-115060 : Speed up pathlib.Path.glob() by removing redundant regex matching.

gh-97928 : Partially revert the behavior of tkinter.Text.count() . By default it preserves the behavior of older Python versions, except that setting wantobjects to 0 no longer has effect. Add a new parameter return_ints : specifying return_ints=True makes Text.count() always returning the single count as an integer instead of a 1-tuple or None .

gh-114628 : When csv.Error is raised when handling TypeError, do not print the TypeError traceback.

gh-85984 : Added _POSIX_VDISABLE from C’s <unistd.h> to termios .

gh-114965 : Update bundled pip to 24.0

gh-114959 : tarfile no longer ignores errors when trying to extract a directory on top of a file.

gh-114894 : Add array.array.clear() .

gh-114071 : Support tuple subclasses using auto() for enum member value.

gh-109475 : Fix support of explicit option value “–” in argparse (e.g. --option=-- ).

gh-49766 : Fix date - datetime comparison. Now the special comparison methods like __eq__ and __lt__ return NotImplemented if one of comparands is date and other is datetime instead of ignoring the time part and the time zone or forcefully return “not equal” or raise TypeError . It makes comparison of date and datetime subclasses more symmetric and allows to change the default behavior by overriding the special comparison methods in subclasses.

gh-110190 : Fix ctypes structs with array on Windows ARM64 platform by setting MAX_STRUCT_SIZE to 32 in stgdict. Patch by Diego Russo

gh-114678 : Ensure that deprecation warning for ‘N’ specifier in Decimal format is not raised for cases where ‘N’ appears in other places in the format specifier. Based on patch by Stefan Krah.

gh-70303 : Return both files and directories from pathlib.Path.glob() if a pattern ends with “ ** ”. Previously only directories were returned.

gh-109653 : Improve import time of importlib.metadata and email.utils .

gh-113280 : Fix a leak of open socket in rare cases when error occurred in ssl.SSLSocket creation.

gh-77749 : email.policy.EmailPolicy.fold() now always encodes non-ASCII characters in headers if utf8 is false.

gh-83383 : Synchronization of the dbm.dumb database is now no-op if there was no modification since opening or last synchronization. The directory file for a newly created empty dbm.dumb database is now created immediately after opening instead of deferring this until synchronizing or closing.

gh-91602 : Add filter keyword-only parameter to sqlite3.Connection.iterdump() for filtering database objects to dump. Patch by Mariusz Felisiak.

gh-112451 : Prohibit subclassing pure-Python datetime.timezone . This is consistent with C-extension implementation. Patch by Mariusz Felisiak.

gh-69893 : Add the close() method for the iterator returned by xml.etree.ElementTree.iterparse() .

gh-109653 : Reduce the import time of threading module by ~50%. Patch by Daniel Hollas.

gh-114492 : Make the result of termios.tcgetattr() reproducible on Alpine Linux. Previously it could leave a random garbage in some fields.

gh-114315 : Make threading.Lock a real class, not a factory function. Add __new__ to _thread.lock type.

gh-100414 : Add dbm.sqlite3 as a backend to dbm , and make it the new default dbm backend. Patch by Raymond Hettinger and Erlend E. Aasland.

gh-113267 : Revert changes in gh-106584 which made calls of TestResult methods startTest() and stopTest() unbalanced.

gh-75128 : Ignore an OSError in asyncio.BaseEventLoop.create_server() when IPv6 is available but the interface cannot actually support it.

gh-114423 : _DummyThread entries in threading._active are now automatically removed when the related thread dies.

gh-114257 : Dismiss the FileNotFound error in ctypes.util.find_library() and just return None on Linux.

gh-114321 : Expose more platform specific constants in the fcntl module on Linux, macOS, FreeBSD and NetBSD.

gh-114328 : The tty.setcbreak() and new tty.cfmakecbreak() no longer clears the terminal input ICRLF flag. This fixes a regression introduced in 3.12 that no longer matched how OSes define cbreak mode in their stty(1) manual pages.

gh-114281 : Remove type hints from Lib/asyncio/staggered.py . The annotations in the typeshed project should be used instead.

gh-101438 : Avoid reference cycle in ElementTree.iterparse. The iterator returned by ElementTree.iterparse may hold on to a file descriptor. The reference cycle prevented prompt clean-up of the file descriptor if the returned iterator was not exhausted.

gh-114198 : The signature for the __replace__ method on dataclasses now has the first argument named self , rather than obj .

gh-104522 : OSError raised when run a subprocess now only has filename attribute set to cwd if the error was caused by a failed attempt to change the current directory.

gh-114149 : Enum: correctly handle tuple subclasses in custom __new__ .

gh-83648 : Support deprecation of options, positional arguments and subcommands in argparse .

gh-114087 : Speed up dataclasses.asdict up to 1.35x.

gh-109534 : Fix a reference leak in asyncio.selector_events.BaseSelectorEventLoop when SSL handshakes fail. Patch contributed by Jamie Phan.

gh-79634 : Accept path-like objects as patterns in pathlib.Path.glob() and rglob() .

gh-112202 : Ensure that a asyncio.Condition.notify() call does not get lost if the awakened Task is simultaneously cancelled or encounters any other error.

gh-113951 : Fix the behavior of tag_unbind() methods of tkinter.Text and tkinter.Canvas classes with three arguments. Previously, widget.tag_unbind(tag, sequence, funcid) destroyed the current binding for sequence , leaving sequence unbound, and deleted the funcid command. Now it removes only funcid from the binding for sequence , keeping other commands, and deletes the funcid command. It leaves sequence unbound only if funcid was the last bound command.

gh-97959 : Fix rendering class methods, bound methods, method and function aliases in pydoc . Class methods no longer have “method of builtins.type instance” note. Corresponding notes are now added for class and unbound methods. Method and function aliases now have references to the module or the class where the origin was defined if it differs from the current. Bound methods are now listed in the static methods section. Methods of builtin classes are now supported as well as methods of Python classes.

gh-113796 : Add more validation checks in the csv.Dialect constructor. ValueError is now raised if the same character is used in different roles.

gh-113732 : Fix support of QUOTE_NOTNULL and QUOTE_STRINGS in csv.reader() .

gh-113225 : Speed up pathlib.Path.walk() by using os.DirEntry.path where possible.

gh-89039 : When replace() method is called on a subclass of datetime, date or time, properly call derived constructor. Previously, only the base class’s constructor was called.

Also, make sure to pass non-zero fold values when creating subclasses in various methods. Previously, fold was silently ignored.

gh-112919 : Speed-up datetime.datetime.replace() , datetime.date.replace() and datetime.time.replace() .

gh-59013 : Set breakpoint on the first executable line of the function, instead of the line of function definition when the user do break func using pdb

gh-112343 : Improve handling of pdb convenience variables to avoid replacing string contents.

gh-112240 : Add option to calendar module CLI to specify the weekday to start each week. Patch by Steven Ward.

gh-111741 : Recognise image/webp as a standard format in the mimetypes module.

gh-43457 : Fix the tkinter widget method wm_attributes() . It now accepts the attribute name without the minus prefix to get window attributes and allows to specify attributes and values to set as keyword arguments. Add new optional keyword argument return_python_dict : calling w.wm_attributes(return_python_dict=True) returns the attributes as a dict instead of a tuple. Calling w.wm_attributes() now returns a tuple instead of string if wantobjects was set to 0.

gh-82626 : Many functions now emit a warning if a boolean value is passed as a file descriptor argument.

gh-111051 : Added check for file modification during debugging with pdb

gh-110345 : Show the Tcl/Tk patchlevel (rather than version) in tkinter._test() .

gh-38807 : Fix race condition in trace . Instead of checking if a directory exists and creating it, directly call os.makedirs() with the kwarg exist_ok=True .

gh-75705 : Set unixfrom envelope in mailbox.mbox and mailbox.MMDF .

gh-106233 : Fix stacklevel in InvalidTZPathWarning during zoneinfo module import.

gh-105102 : Allow ctypes.Union to be nested in ctypes.Structure when the system endianness is the opposite of the classes.

gh-104282 : Fix null pointer dereference in lzma._decode_filter_properties() due to improper handling of BCJ filters with properties of zero length. Patch by Radislav Chugunov.

gh-96471 : Add queue.Queue termination with shutdown() .

gh-101599 : Changed argparse flag options formatting to remove redundancy.

gh-85984 : Add POSIX pseudo-terminal functions os.posix_openpt() , os.grantpt() , os.unlockpt() , and os.ptsname() .

gh-102512 : When os.fork() is called from a foreign thread (aka _DummyThread ), the type of the thread in a child process is changed to _MainThread . Also changed its name and daemonic status, it can be now joined.

gh-88569 : Add os.path.isreserved() , which identifies reserved pathnames such as “NUL”, “AUX” and “CON”. This function is only available on Windows.

Deprecate pathlib.PurePath.is_reserved() .

bpo-38364 : The inspect functions isgeneratorfunction , iscoroutinefunction , isasyncgenfunction now support functools.partialmethod wrapped functions the same way they support functools.partial .

gh-115233 : Fix an example for LoggerAdapter in the Logging Cookbook.

gh-114123 : Move the csv module docstring to the csv module instead of reexporting it from the internal _csv module, and remove __doc__ from csv.__all__ .

Move csv.__version__ to the csv module instead of reexporting it from the internal _csv module, and remove __version__ from csv.__all__ .

gh-114099 : Added test exclusions required to run the test suite on iOS.

gh-105089 : Fix test.test_zipfile.test_core.TestWithDirectory.test_create_directory_with_write test in AIX by doing a bitwise AND of 0xFFFF on mode , so that it will be in sync with zinfo.external_attr

gh-115167 : Avoid vendoring vcruntime140_threads.dll when building with Visual Studio 2022 version 17.8.

gh-113632 : Promote WASI to a tier 2 platform and drop Emscripten from tier 3 in configure.ac.

gh-114099 : configure and Makefile were refactored to accommodate framework builds on Apple platforms other than macOS.

gh-114875 : Add getgrent() as a prerequisite for building the grp module.

gh-115049 : Fixes py.exe launcher failing when run as users without user profiles.

gh-115009 : Update Windows installer to use SQLite 3.45.1.

gh-109991 : Update Windows build to use OpenSSL 3.0.13.

gh-111239 : Update Windows builds to use zlib v1.3.1.

gh-100107 : The py.exe launcher will no longer attempt to run the Microsoft Store redirector when launching a script containing a /usr/bin/env shebang

gh-112984 : Adds free-threaded binaries to Windows installer as an optional component.

gh-89240 : Allows multiprocessing to create pools of greater than 62 processes.

gh-115009 : Update macOS installer to use SQLite 3.45.1.

gh-109991 : Update macOS installer to use OpenSSL 3.0.13.

gh-114490 : Add Mach-O linkage support for platform.architecture() .

gh-87804 : On macOS the result of os.statvfs and os.fstatvfs now correctly report the size of very large disks, in previous versions the reported number of blocks was wrong for disks with at least 2**32 blocks.

gh-96905 : In idlelib code, stop redefining built-ins ‘dict’ and ‘object’.

gh-103820 : Revise IDLE bindings so that events from mouse button 4/5 on non-X11 windowing systems (i.e. Win32 and Aqua) are not mistaken for scrolling.

gh-113516 : Don’t set LDSHARED when building for WASI.

gh-109991 : Update GitHub CI workflows to use OpenSSL 3.0.13 and multissltests to use 1.1.1w, 3.0.13, 3.1.5, and 3.2.1.

gh-115015 : Fix a bug in Argument Clinic that generated incorrect code for methods with no parameters that use the METH_METHOD | METH_FASTCALL | METH_KEYWORDS calling convention. Only the positional parameter count was checked; any keyword argument passed would be silently accepted.

gh-111140 : Adds PyLong_AsNativeBytes() , PyLong_FromNativeBytes() and PyLong_FromUnsignedNativeBytes() functions.

gh-114685 : PyBuffer_FillInfo() now raises a SystemError if called with PyBUF_READ or PyBUF_WRITE as flags. These flags should only be used with the PyMemoryView_* C API.

gh-114685 : PyObject_GetBuffer() now raises a SystemError if called with PyBUF_READ or PyBUF_WRITE as flags. These flags should only be used with the PyMemoryView_* C API.

gh-114626 : Add PyCFunctionFast and PyCFunctionFastWithKeywords typedefs (identical to the existing _PyCFunctionFast and _PyCFunctionFastWithKeywords typedefs, just without a leading _ prefix).

gh-114329 : Add PyList_GetItemRef() , which is similar to PyList_GetItem() but returns a strong reference instead of a borrowed reference .

gh-110850 : Add PyTime C API:

PyTime_t type.

PyTime_MIN and PyTime_MAX constants.

PyTime_AsSecondsDouble() , PyTime_Monotonic() , PyTime_PerfCounter() , and PyTime_Time() functions.

gh-112066 : Add PyDict_SetDefaultRef() : insert a key and value into a dictionary if the key is not already present. This is similar to dict.setdefault() , but returns an integer value indicating if the key was already present. It is also similar to PyDict_SetDefault() , but returns a strong reference instead of a borrowed reference.

Python 3.13.0 alpha 3 ¶

Release date: 2024-01-17

gh-113659 : Skip .pth files with names starting with a dot or hidden file attribute.

gh-112302 : Created a Software Bill-of-Materials document and tooling for tracking dependencies.

gh-107901 : Compiler duplicates basic blocks that have an eval breaker check, no line number, and multiple predecessors.

gh-107901 : A jump leaving an exception handler back to normal code no longer checks the eval breaker.

gh-113655 : Set the C recursion limit to 4000 on Windows, and 10000 on Linux/OSX. This seems to be near the sweet spot to maintain safety, but not compromise backwards compatibility.

gh-113710 : Add typed stack effects to the interpreter DSL, along with various instruction annotations.

gh-77046 : On Windows, file descriptors wrapping Windows handles are now created non inheritable by default ( PEP 446 ). Patch by Zackery Spytz and Victor Stinner.

gh-113853 : Guarantee that all executors make progress. This then guarantees that tier 2 execution always makes progress.

gh-113753 : Fix an issue where the finalizer of PyAsyncGenASend objects might not be called if they were allocated from a free list.

gh-107901 : Compiler changed so that synthetic jumps which are not at loop end no longer check the eval breaker.

gh-113703 : Fix a regression in the codeop module that was causing it to incorrectly identify incomplete f-strings. Patch by Pablo Galindo

gh-89811 : Check for a valid tp_version_tag before performing bytecode specializations that rely on this value being usable.

gh-111488 : Changed error message in case of no ‘in’ keyword after ‘for’ in list comprehensions

gh-113657 : Fix an issue that caused important instruction pointer updates to be optimized out of tier two traces.

gh-113603 : Fixed bug where a redundant NOP is not removed, causing an assertion to fail in the compiler in debug mode.

gh-113602 : Fix an error that was causing the parser to try to overwrite existing errors and crashing in the process. Patch by Pablo Galindo

gh-113486 : No longer issue spurious PY_UNWIND events for optimized calls to classes.

gh-113297 : Fix segfault in the compiler on with statement with 19 context managers.

gh-113212 : Improve super error messages.

gh-111375 : Only use NULL in the exception stack to indicate an exception was handled. Patch by Carey Metcalfe.

gh-112215 : Increase the C recursion limit by a factor of 3 for non-debug builds, except for webassembly and s390 platforms which are unchanged. This mitigates some regressions in 3.12 with deep recursion mixing builtin (C) and Python code.

gh-113054 : Fixed bug where a redundant NOP is not removed, causing an assertion to fail in the compiler in debug mode.

gh-106905 : Use per AST-parser state rather than global state to track recursion depth within the AST parser to prevent potential race condition due to simultaneous parsing.

The issue primarily showed up in 3.11 by multithreaded users of ast.parse() . In 3.12 a change to when garbage collection can be triggered prevented the race condition from occurring.

gh-108866 : Change the API and contract of _PyExecutorObject to return the next_instr pointer, instead of the frame, and to always execute at least one instruction.

gh-90350 : Optimize builtin functions min() and max() .

gh-112943 : Correctly compute end column offsets for multiline tokens in the tokenize module. Patch by Pablo Galindo

gh-112125 : Fix None.__ne__(None) returning NotImplemented instead of False .

gh-74616 : input() now raises a ValueError when output on the terminal if the prompt contains embedded null characters instead of silently truncating it.

gh-112716 : Fix SystemError in the import statement and in __reduce__() methods of builtin types when __builtins__ is not a dict.

gh-112730 : Use color to highlight error locations in tracebacks. Patch by Pablo Galindo

gh-112625 : Fixes a bug where a bytearray object could be cleared while iterating over an argument in the bytearray.join() method that could result in reading memory after it was freed.

gh-112660 : Do not clear unexpected errors during formatting error messages for ImportError and AttributeError for modules.

gh-105967 : Workaround a bug in Apple’s macOS platform zlib library where zlib.crc32() and binascii.crc32() could produce incorrect results on multi-gigabyte inputs. Including when using zipfile on zips containing large data.

gh-95754 : Provide a better error message when accessing invalid attributes on partially initialized modules. The origin of the module being accessed is now included in the message to help with the common issue of shadowing other modules.

gh-112217 : Add check for the type of __cause__ returned from calling the type T in raise from T .

gh-111058 : Change coro.cr_frame/gen.gi_frame to return None after the coroutine/generator has been closed. This fixes a bug where getcoroutinestate() and getgeneratorstate() return the wrong state for a closed coroutine/generator.

gh-112388 : Fix an error that was causing the parser to try to overwrite tokenizer errors. Patch by pablo Galindo

gh-112387 : Fix error positions for decoded strings with backwards tokenize errors. Patch by Pablo Galindo

gh-99606 : Make code generated for an empty f-string identical to the code of an empty normal string.

gh-112367 : Avoid undefined behaviour when using the perf trampolines by not freeing the code arenas until shutdown. Patch by Pablo Galindo

gh-112320 : The Tier 2 translator now tracks the confidence level for staying “on trace” (i.e. not exiting back to the Tier 1 interpreter) for branch instructions based on the number of bits set in the branch “counter”. Trace translation ends when the confidence drops below 1/3rd.

gh-109598 : PyComplex_RealAsDouble() / PyComplex_ImagAsDouble() now tries to convert an object to a complex instance using its __complex__() method before falling back to the __float__() method. Patch by Sergey B Kirpichev.

gh-94606 : Fix UnicodeEncodeError when email.message.get_payload() reads a message with a Unicode surrogate character and the message content is not well-formed for surrogateescape encoding. Patch by Sidney Markowitz.

bpo-21861 : Use the object’s actual class name in _io.FileIO.__repr__() , _io._WindowsConsoleIO() and _io.TextIOWrapper.__repr__() , to make these methods subclass friendly.

bpo-45369 : Remove LibreSSL workarounds as per PEP 644 .

bpo-34392 : Added sys._is_interned() .

gh-114077 : Fix possible OverflowError in socket.socket.sendfile() when pass count larger than 2 GiB on 32-bit platform.

gh-111803 : plistlib now supports loading more deeply nested lists in binary format.

gh-114014 : Fixed a bug in fractions.Fraction where an invalid string using d in the decimals part creates a different error compared to other invalid letters/characters. Patch by Jeremiah Gabriel Pascual.

gh-108364 : sqlite3.Connection.iterdump() now ensures that foreign key support is disabled before dumping the database schema, if there is any foreign key violation. Patch by Erlend E. Aasland and Mariusz Felisiak.

gh-113971 : The zipfile.ZipInfo previously protected ._compresslevel attribute has been made public as .compress_level with the old _compresslevel name remaining available as a property to retain compatibility.

gh-113877 : Fix tkinter method winfo_pathname() on 64-bit Windows.

gh-113868 : Added mmap.MAP_NORESERVE , mmap.MAP_NOEXTEND , mmap.MAP_HASSEMAPHORE , mmap.MAP_NOCACHE , mmap.MAP_JIT , mmap.MAP_RESILIENT_CODESIGN , mmap.MAP_RESILIENT_MEDIA , mmap.MAP_32BIT , mmap.MAP_TRANSLATED_ALLOW_EXECUTE , mmap.MAP_UNIX03 and mmap.MAP_TPRO . All of them are mmap(2) flags on macOS.

gh-113848 : asyncio.TaskGroup() and asyncio.timeout() context managers now handle CancelledError subclasses as well as exact CancelledError .

gh-113661 : unittest runner: Don’t exit 5 if tests were skipped. The intention of exiting 5 was to detect issues where the test suite wasn’t discovered at all. If we skipped tests, it was correctly discovered.

gh-96037 : Insert TimeoutError in the context of the exception that was raised during exiting an expired asyncio.timeout() block.

gh-113781 : Silence unraisable AttributeError when warnings are emitted during Python finalization.

gh-113238 : Add Anchor to importlib.resources (in order for the code to comply with the documentation)

gh-111693 : asyncio.Condition.wait() now re-raises the same CancelledError instance that may have caused it to be interrupted. Fixed race condition in asyncio.Semaphore.acquire() when interrupted with a CancelledError .

gh-113791 : Add CLOCK_MONOTONIC_RAW_APPROX and CLOCK_UPTIME_RAW_APPROX to time on macOS. These are clocks available on macOS 10.12 or later.

gh-112932 : Restore the ability for zipfile to extractall from zip files with a “/” directory entry in them as is commonly added to zips by some wiki or bug tracker data exporters.

gh-113568 : Raise deprecation warnings from pathlib.PurePath and not its private base class PurePathBase .

gh-113594 : Fix UnicodeEncodeError in email when re-fold lines that contain unknown-8bit encoded part followed by non-unknown-8bit encoded part.

gh-113538 : In asyncio.StreamReaderProtocol.connection_made() , there is callback that logs an error if the task wrapping the “connected callback” fails. This callback would itself fail if the task was cancelled. Prevent this by checking whether the task was cancelled first. If so, close the transport but don’t log an error.

gh-113626 : Add support for the allow_code argument in the marshal module. Passing allow_code=False prevents serialization and de-serialization of code objects which is incompatible between Python versions.

gh-85567 : Fix resource warnings for unclosed files in pickle and pickletools command line interfaces.

gh-113537 : Support loads str in plistlib.loads() .

gh-89850 : Add default implementations of pickle.Pickler.persistent_id() and pickle.Unpickler.persistent_load() methods in the C implementation. Calling super().persistent_id() and super().persistent_load() in subclasses of the C implementation of pickle.Pickler and pickle.Unpickler classes no longer causes infinite recursion.

gh-113569 : Indicate if there were no actual calls in unittest assert_has_calls() failure.

gh-101225 : Increase the backlog for multiprocessing.connection.Listener objects created by multiprocessing.manager and multiprocessing.resource_sharer to significantly reduce the risk of getting a connection refused error when creating a multiprocessing.connection.Connection to them.

gh-113568 : Raise audit events from pathlib.Path and not its private base class PathBase .

gh-113543 : Make sure that webbrowser.MacOSXOSAScript sends webbrowser.open audit event.

gh-113028 : When a second reference to a string appears in the input to pickle , and the Python implementation is in use, we are guaranteed that a single copy gets pickled and a single object is shared when reloaded. Previously, in protocol 0, when a string contained certain characters (e.g. newline) it resulted in duplicate objects.

gh-113421 : Fix multiprocessing logger for %(filename)s .

gh-111784 : Fix segfaults in the _elementtree module. Fix first segfault during deallocation of _elementtree.XMLParser instances by keeping strong reference to pyexpat module in module state for capsule lifetime. Fix second segfault which happens in the same deallocation process by keeping strong reference to _elementtree module in XMLParser structure for _elementtree module lifetime.

gh-113407 : Fix import of unittest.mock when CPython is built without docstrings.

gh-113320 : Fix regression in Python 3.12 where Protocol classes that were not marked as runtime-checkable would be unnecessarily introspected, potentially causing exceptions to be raised if the protocol had problematic members. Patch by Alex Waygood.

gh-53502 : Add a new option aware_datetime in plistlib to loads or dumps aware datetime.

gh-113358 : Fix rendering tracebacks with exceptions with a broken __getattr__

gh-113214 : Fix an AttributeError during asyncio SSL protocol aborts in SSL-over-SSL scenarios.

gh-113246 : Update bundled pip to 23.3.2.

gh-87264 : Fixed tarfile list() method to show file type.

gh-112182 : asyncio.futures.Future.set_exception() now transforms StopIteration into RuntimeError instead of hanging or other misbehavior. Patch contributed by Jamie Phan.

gh-113225 : Speed up pathlib.Path.glob() by using os.DirEntry.path where possible.

gh-113149 : Improve error message when a JSON array or object contains a trailing comma. Patch by Carson Radtke.

gh-113117 : The subprocess module can now use the os.posix_spawn() function with close_fds=True on platforms where posix_spawn_file_actions_addclosefrom_np is available. Patch by Jakub Kulik.

gh-113199 : Make http.client.HTTPResponse.read1 and http.client.HTTPResponse.readline close IO after reading all data when content length is known. Patch by Illia Volochii.

gh-113191 : Add support of os.fchmod() and a file descriptor in os.chmod() on Windows.

gh-113188 : Fix shutil.copymode() and shutil.copystat() on Windows. Previously they worked differently if dst is a symbolic link: they modified the permission bits of dst itself rather than the file it points to if follow_symlinks is true or src is not a symbolic link, and did not modify the permission bits if follow_symlinks is false and src is a symbolic link.

gh-113119 : os.posix_spawn() now accepts env=None , which makes the newly spawned process use the current process environment. Patch by Jakub Kulik.

gh-113202 : Add a strict option to batched() in the itertools module.

gh-61648 : Detect line numbers of properties in doctests.

gh-113175 : Sync with importlib_metadata 7.0, including improved type annotations, fixed issue with symlinked packages in package_distributions , added EntryPoints.__repr__ , introduced the diagnose script, added Distribution.origin property, and removed deprecated EntryPoint access by numeric index (tuple behavior).

gh-59616 : Add support of os.lchmod() and the follow_symlinks argument in os.chmod() on Windows. Note that the default value of follow_symlinks in os.lchmod() is False on Windows.

gh-112559 : signal.signal() and signal.getsignal() no longer call repr on callable handlers. asyncio.run() and asyncio.Runner.run() no longer call repr on the task results. Patch by Yilei Yang.

gh-112962 : dis module functions add cache information to the Instruction instance rather than creating fake Instruction instances to represent the cache entries.

gh-112989 : Reduce overhead to connect sockets with asyncio SelectorEventLoop.

gh-112970 : Use closefrom() on Linux where available (e.g. glibc-2.34), rather than only FreeBSD.

gh-110190 : Fix ctypes structs with array on PPC64LE platform by setting MAX_STRUCT_SIZE to 64 in stgdict. Patch by Diego Russo.

gh-112540 : The statistics.geometric_mean() function now returns zero for datasets containing a zero. Formerly, it would raise an exception.

gh-87286 : Added LOG_FTP , LOG_NETINFO , LOG_REMOTEAUTH , LOG_INSTALL , LOG_RAS , and LOG_LAUNCHD tot the syslog module, all of them constants on used on macOS.

gh-112800 : Fix asyncio SubprocessTransport.close() not to throw PermissionError when used with setuid executables.

gh-51944 : Add the following constants to the termios module. These values are present in macOS system headers: ALTWERASE , B14400 , B28800 , B7200 , B76800 , CCAR_OFLOW , CCTS_OFLOW , CDSR_OFLOW , CDTR_IFLOW , CIGNORE , CRTS_IFLOW , EXTPROC , IUTF8 , MDMBUF , NL2 , NL3 , NOKERNINFO , ONOEOT , OXTABS , VDSUSP , VSTATUS .

gh-79325 : Fix an infinite recursion error in tempfile.TemporaryDirectory() cleanup on Windows.

gh-94692 : shutil.rmtree() now only catches OSError exceptions. Previously a symlink attack resistant version of shutil.rmtree() could ignore or pass to the error handler arbitrary exception when invalid arguments were provided.

gh-112736 : The use of del-safe symbols in subprocess was refactored to allow for use in cross-platform build environments.

gh-112727 : Speed up pathlib.Path.absolute() . Patch by Barney Gale.

gh-74690 : Speedup issubclass() checks against simple runtime-checkable protocols by around 6%. Patch by Alex Waygood.

gh-74690 : Speedup isinstance() checks by roughly 20% for runtime-checkable protocols that only have one callable member. Speedup issubclass() checks for these protocols by roughly 10%. Patch by Alex Waygood.

gh-112645 : Remove deprecation error on passing onerror to shutil.rmtree() .

gh-112640 : Add kwdefaults parameter to types.FunctionType to set default keyword argument values.

gh-112622 : Ensure name parameter is passed to event loop in asyncio.create_task() .

gh-112618 : Fix a caching bug relating to typing.Annotated . Annotated[str, True] is no longer identical to Annotated[str, 1] .

gh-112334 : Fixed a performance regression in 3.12’s subprocess on Linux where it would no longer use the fast-path vfork() system call when it could have due to a logic bug, instead falling back to the safe but slower fork() .

Also fixed a second 3.12.0 potential security bug. If a value of extra_groups=[] was passed to subprocess.Popen or related APIs, the underlying setgroups(0, NULL) system call to clear the groups list would not be made in the child process prior to exec() .

This was identified via code inspection in the process of fixing the first bug.

gh-110190 : Fix ctypes structs with array on Arm platform by setting MAX_STRUCT_SIZE to 32 in stgdict. Patch by Diego Russo.

gh-81194 : Fix a crash in socket.if_indextoname() with specific value (UINT_MAX). Fix an integer overflow in socket.if_indextoname() on 64-bit non-Windows platforms.

gh-112578 : Fix a spurious RuntimeWarning when executing the zipfile module.

gh-112516 : Update the bundled copy of pip to version 23.3.1.

gh-112510 : Add readline.backend for the backend readline uses ( editline or readline )

gh-112328 : [Enum] Make EnumDict , EnumDict.member_names , EnumType._add_alias_ and EnumType._add_value_alias_ public.

gh-112509 : Fix edge cases that could cause a key to be present in both the __required_keys__ and __optional_keys__ attributes of a typing.TypedDict . Patch by Jelle Zijlstra.

gh-101336 : Add keep_alive keyword parameter for AbstractEventLoop.create_server() and BaseEventLoop.create_server() .

gh-63284 : Added support for TLS-PSK (pre-shared key) mode to the ssl module.

gh-112414 : Fix regression in Python 3.12 where calling repr() on a module that had been imported using a custom loader could fail with AttributeError . Patch by Alex Waygood.

gh-112358 : Revert change to struct.Struct initialization that broke some cases of subclassing.

gh-112405 : Optimize pathlib.PurePath.relative_to() . Patch by Alex Waygood.

gh-94722 : Fix bug where comparison between instances of DocTest fails if one of them has None as its lineno.

gh-112361 : Speed up a small handful of pathlib methods by removing some temporary objects.

gh-112345 : Improve error message when trying to call issubclass() against a typing.Protocol that has non-method members. Patch by Randolf Scholz.

gh-112137 : Change dis output to display no-lineno as “–” instead of “None”.

gh-112332 : Deprecate the exc_type field of traceback.TracebackException . Add exc_type_str to replace it.

gh-81620 : Add extra tests for random.binomialvariate()

gh-112292 : Fix a crash in readline when imported from a sub interpreter. Patch by Anthony Shaw

gh-77621 : Slightly improve the import time of the pathlib module by deferring some imports. Patch by Barney Gale.

gh-112137 : Change dis output to display logical labels for jump targets instead of offsets.

gh-112139 : Add Signature.format() to format signatures to string with extra options. And use it in pydoc to render more readable signatures that have new lines between parameters.

gh-112105 : Make readline.set_completer_delims() work with libedit

gh-106922 : Display multiple lines with traceback when errors span multiple lines.

gh-111874 : When creating a typing.NamedTuple class, ensure __set_name__() is called on all objects that define __set_name__ and exist in the values of the NamedTuple class’s class dictionary. Patch by Alex Waygood.

gh-68166 : Add support of the “vsapi” element type in tkinter.ttk.Style.element_create() .

gh-110275 : Named tuple’s methods _replace() and __replace__() now raise TypeError instead of ValueError for invalid keyword arguments.

gh-99367 : Do not mangle sys.path[0] in pdb if safe_path is set

gh-111615 : Fix a regression caused by a fix to gh-93162 whereby you couldn’t configure a QueueHandler without specifying handlers.

gh-75666 : Fix the behavior of tkinter widget’s unbind() method with two arguments. Previously, widget.unbind(sequence, funcid) destroyed the current binding for sequence , leaving sequence unbound, and deleted the funcid command. Now it removes only funcid from the binding for sequence , keeping other commands, and deletes the funcid command. It leaves sequence unbound only if funcid was the last bound command.

gh-67790 : Implement basic formatting support (minimum width, alignment, fill) for fractions.Fraction .

gh-111049 : Fix crash during garbage collection of the io.BytesIO buffer object.

gh-102980 : Redirect the output of interact command of pdb to the same channel as the debugger. Add tests and improve docs.

gh-102988 : email.utils.getaddresses() and email.utils.parseaddr() now return ('', '') 2-tuples in more situations where invalid email addresses are encountered instead of potentially inaccurate values. Add optional strict parameter to these two functions: use strict=False to get the old behavior, accept malformed inputs. getattr(email.utils, 'supports_strict_parsing', False) can be use to check if the strict parameter is available. Patch by Thomas Dwyer and Victor Stinner to improve the CVE-2023-27043 fix.

gh-52161 : cmd.Cmd.do_help() now cleans docstrings with inspect.cleandoc() before writing them. Patch by Filip Łapkiewicz.

gh-82300 : Add track parameter to multiprocessing.shared_memory.SharedMemory that allows using shared memory blocks without having to register with the POSIX resource tracker that automatically releases them upon process exit.

gh-110109 : Add private pathlib._PurePathBase class: a base class for pathlib.PurePath that omits certain magic methods. It may be made public (along with _PathBase ) in future.

gh-109858 : Protect zipfile from “quoted-overlap” zipbomb. It now raises BadZipFile when try to read an entry that overlaps with other entry or central directory.

gh-109786 : Fix possible reference leaks and crash when re-enter the __next__() method of itertools.pairwise .

gh-91539 : Small (10 - 20%) and trivial performance improvement of urllib.request.getproxies_environment() , typically useful when there are many environment variables to go over.

gh-103363 : Add follow_symlinks keyword-only argument to pathlib.Path.owner() and group() , defaulting to True .

gh-102130 : Support tab completion in cmd for editline .

gh-99437 : runpy.run_path() now decodes path-like objects, making sure __file__ and sys.argv[0] of the module being run are always strings.

gh-104003 : Add warnings.deprecated() , a decorator to mark deprecated functions to static type checkers and to warn on usage of deprecated classes and functions. See PEP 702 . Patch by Jelle Zijlstra.

gh-103708 : Make hardcoded python name, a configurable parameter so that different implementations of python can override it instead of making huge diffs in sysconfig.py

gh-66515 : mailbox.MH now supports folders that do not contain a .mh_sequences file (e.g. Claws Mail IMAP-cache folders). Patch by Serhiy Storchaka.

gh-83162 : Renamed re.error to PatternError for clarity, and kept re.error for backward compatibility. Patch by Matthias Bussonnier and Adam Chhina.

gh-91133 : Fix a bug in tempfile.TemporaryDirectory cleanup, which now no longer dereferences symlinks when working around file system permission errors.

bpo-43153 : On Windows, tempfile.TemporaryDirectory previously masked a PermissionError with NotADirectoryError during directory cleanup. It now correctly raises PermissionError if errors are not ignored. Patch by Andrei Kulakov and Ken Jin.

bpo-32731 : getpass.getuser() now raises OSError for all failures rather than ImportError on systems lacking the pwd module or KeyError if the password database is empty.

bpo-34321 : mmap.mmap now has a trackfd parameter on Unix; if it is False , the file descriptor specified by fileno will not be duplicated.

bpo-35332 : The shutil.rmtree() function now ignores errors when calling os.close() when ignore_errors is True , and os.close() no longer retried after error.

bpo-35928 : io.TextIOWrapper now correctly handles the decoding buffer after read() and write() .

bpo-26791 : shutil.move() now moves a symlink into a directory when that directory is the target of the symlink. This provides the same behavior as the mv shell command. The previous behavior raised an exception. Patch by Jeffrey Kintscher.

bpo-41422 : Fixed memory leaks of pickle.Pickler and pickle.Unpickler involving cyclic references via the internal memo mapping.

bpo-19821 : The pydoc.ispackage() function has been deprecated.

bpo-40262 : The ssl.SSLSocket.recv_into() method no longer requires the buffer argument to implement __len__ and supports buffers with arbitrary item size.

bpo-39912 : warnings.filterwarnings() and warnings.simplefilter() now raise appropriate exceptions instead of AssertionError . Patch contributed by Rémi Lapeyre.

bpo-37260 : Fixed a race condition in shutil.rmtree() in which directory entries removed by another process or thread while shutil.rmtree() is running can cause it to raise FileNotFoundError. Patch by Jeffrey Kintscher.

bpo-36959 : Fix some error messages for invalid ISO format string combinations in strptime() that referred to directives not contained in the format string. Patch by Gordon P. Hemsley.

bpo-18060 : Fixed a class inheritance issue that can cause segfaults when deriving two or more levels of subclasses from a base class of Structure or Union.

bpo-29779 : Add a new PYTHON_HISTORY environment variable to set the location of a .python_history file.

bpo-21360 : mailbox.Maildir now ignores files with a leading dot.

gh-111699 : Relocate smtpd deprecation notice to its own section rather than under locale in What’s New in Python 3.12 document

gh-110746 : Improved markup for valid options/values for methods ttk.treeview.column and ttk.treeview.heading, and for Layouts.

gh-95649 : Document that the asyncio module contains code taken from v0.16.0 of the uvloop project , as well as the required MIT licensing information.

gh-111798 : Disable test_super_deep() from test_call under pydebug builds on WASI; the stack depth is too small to make the test useful.

gh-111801 : Lower the recursion limit in test_isinstance for test_infinitely_many_bases() . This prevents a stack overflow on a pydebug build of WASI.

gh-111802 : Specify a low recursion depth for test_bad_getattr() in test.pickletester to avoid exhausting the stack under a pydebug build for WASI.

gh-44626 : Fix os.path.isabs() incorrectly returning True when given a path that starts with exactly one (back)slash on Windows.

Fix pathlib.PureWindowsPath.is_absolute() incorrectly returning False for some paths beginning with two (back)slashes.

gh-113633 : Use module state for the _testcapi extension module.

gh-109980 : Fix test_tarfile_vs_tar in test_shutil for macOS, where system tar can include more information in the archive than shutil.make_archive .

gh-112769 : The tests now correctly compare zlib version when zlib.ZLIB_RUNTIME_VERSION contains non-integer suffixes. For example zlib-ng defines the version as 1.3.0.zlib-ng .

gh-112334 : Adds a regression test to verify that vfork() is used when expected by subprocess on vfork enabled POSIX systems (Linux).

gh-108927 : Fixed order dependence in running tests in the same process when a test that has submodules (e.g. test_importlib) follows a test that imports its submodule (e.g. test_importlib.util) and precedes a test (e.g. test_unittest or test_compileall) that uses that submodule.

bpo-40648 : Test modes that file can get with chmod() on Windows.

gh-114013 : Fix Tools/wasm/wasi.py to not include the path to python.wasm as part of HOSTRUNNER . The environment variable is meant to specify how to run the WASI host only, having python.wasm and relevant flags appended to the HOSTRUNNER . This fixes make test work.

gh-113258 : Changed the Windows build to write out generated frozen modules into the build tree instead of the source tree.

gh-112305 : Fixed the check-clean-src step performed on out of tree builds to detect errant $(srcdir)/Python/frozen_modules/*.h files and recommend appropriate source tree cleanup steps to get a working build again.

gh-112536 : Add support for thread sanitizer (TSAN)

gh-112867 : Fix the build for the case that WITH_PYMALLOC_RADIX_TREE=0 set.

gh-103065 : Introduce Tools/wasm/wasi.py to simplify doing a WASI build.

bpo-11102 : The os.major() , os.makedev() , and os.minor() functions are now available on HP-UX v3.

bpo-36351 : Do not set ipv6type when cross-compiling.

gh-114096 : Process privileges that are activated for creating directory junctions are now restored afterwards, avoiding behaviour changes in other parts of the program.

gh-111877 : os.stat() calls were returning incorrect time values for files that could not be accessed directly.

gh-111973 : Update Windows installer to use SQLite 3.44.2.

gh-113009 : multiprocessing : On Windows, fix a race condition in Process.terminate() : no longer set the returncode attribute to always call WaitForSingleObject() in Process.wait() . Previously, sometimes the process was still running after TerminateProcess() even if GetExitCodeProcess() is not STILL_ACTIVE . Patch by Victor Stinner.

gh-86179 : Fixes path calculations when launching Python on Windows through a symlink.

gh-71383 : Update Tcl/Tk in Windows installer to 8.6.13 with a patch to suppress incorrect ThemeChanged warnings.

gh-111650 : Ensures the Py_GIL_DISABLED preprocessor variable is defined in pyconfig.h so that extension modules written in C are able to use it.

gh-112278 : Reduce the time cost for some functions in platform on Windows if current user has no permission to the WMI.

gh-73427 : Deprecate sys._enablelegacywindowsfsencoding() . Use PYTHONLEGACYWINDOWSFSENCODING instead. Patch by Inada Naoki.

gh-87868 : Correctly sort and remove duplicate environment variables in _winapi.CreateProcess() .

bpo-37308 : Fix mojibake in mmap.mmap when using a non-ASCII tagname argument on Windows.

gh-113666 : Add the following constants to module stat : UF_SETTABLE , UF_TRACKED , UF_DATAVAULT , SF_SUPPORTED , SF_SETTABLE , SF_SYNTHETIC , SF_RESTRICTED , SF_FIRMLINK and SF_DATALESS . The values UF_SETTABLE , SF_SUPPORTED , SF_SETTABLE and SF_SYNTHETIC are only available on macOS.

gh-113536 : os.waitid() is now available on macOS

gh-110459 : Running configure ... --with-openssl-rpath=X/Y/Z no longer fails to detect OpenSSL on macOS.

gh-74573 : Document that dbm.ndbm can silently corrupt DBM files on updates when exceeding undocumented platform limits, and can crash (segmentation fault) when reading such a corrupted file. (FB8919203)

gh-65701 : The freeze tool doesn’t work with framework builds of Python. Document this and bail out early when running the tool with such a build.

gh-87277 : webbrowser: Don’t look for X11 browsers on macOS. Those are generally not used and probing for them can result in starting XQuartz even if it isn’t used otherwise.

gh-111973 : Update macOS installer to use SQLite 3.44.2.

gh-108269 : Set CFBundleAllowMixedLocalizations to true in the Info.plist for the framework, embedded Python.app and IDLE.app with framework installs on macOS. This allows applications to pick up the user’s preferred locale when that’s different from english.

gh-102362 : Make sure the result of sysconfig.get_plaform() includes at least a major and minor versions, even if MACOSX_DEPLOYMENT_TARGET is set to only a major version during build to match the format expected by pip.

gh-110017 : Disable a signal handling stress test on macOS due to a bug in macOS (FB13453490).

gh-110820 : Make sure the preprocessor definitions for ALIGNOF_MAX_ALIGN_T , SIZEOF_LONG_DOUBLE and HAVE_GCC_ASM_FOR_X64 are correct for Universal 2 builds on macOS.

gh-109981 : Use /dev/fd on macOS to determine the number of open files in test.support.os_helper.fd_count to avoid a crash with “guarded” file descriptors when probing for open files.

gh-72284 : Improve the lists of features, editor key bindings, and shell key bingings in the IDLE doc.

gh-113903 : Fix rare failure of test.test_idle, in test_configdialog.

gh-113729 : Fix the “Help -> IDLE Doc” menu bug in 3.11.7 and 3.12.1.

gh-113269 : Fix test_editor hang on macOS Catalina.

gh-112898 : Fix processing unsaved files when quitting IDLE on macOS.

bpo-13586 : Enter the selected text when opening the “Replace” dialog.

gh-106560 : Fix redundant declarations in the public C API. Declare PyBool_Type, PyLong_Type and PySys_Audit() only once. Patch by Victor Stinner.

gh-112438 : Fix support of format units “es”, “et”, “es#”, and “et#” in nested tuples in PyArg_ParseTuple() -like functions.

gh-111545 : Add Py_HashPointer() function to hash a pointer. Patch by Victor Stinner.

gh-65210 : Change the declaration of the keywords parameter of PyArg_ParseTupleAndKeywords() and PyArg_VaParseTupleAndKeywords() for better compatibility with C++.

Python 3.13.0 alpha 2 ¶

Release date: 2023-11-22

gh-112243 : Don’t include comments in f-string debug expressions. Patch by Pablo Galindo

gh-112287 : Slightly optimize the Tier 2 (uop) interpreter by only loading oparg and operand when needed. Also double the trace size limit again, to 512 this time.

gh-112266 : Change docstrings of __dict__ and __weakref__ .

gh-111807 : Lower the max parser stack depth to 1000 under WASI debug builds.

gh-111798 : When Python is built in debug mode, set the C recursion limit to 500 instead of 1500. A debug build is likely built with low optimization level which implies higher stack memory usage than a release build. Patch by Victor Stinner.

gh-106529 : Enable translating unspecialized FOR_ITER to Tier 2.

gh-111916 : Make hashlib related modules thread-safe without the GIL

gh-81137 : Deprecate assignment to a function’s __code__ field when the new code object is of a mismatched type (e.g., from a generator to a plain function).

gh-79932 : Raise exception if frame.clear() is called on a suspended frame.

gh-81925 : Implement native thread ids for GNU KFreeBSD.

gh-111843 : Use exponential backoff to reduce the number of failed tier 2 optimization attempts by over 99%.

gh-110829 : Joining a thread now ensures the underlying OS thread has exited. This is required for safer fork() in multi-threaded processes.

gh-109369 : Make sure that tier 2 traces are de-optimized if the code is instrumented

gh-111772 : Specialize slot loads and stores for _Py_T_OBJECT as well as Py_T_OBJECT_EX

gh-111666 : Speed up BaseExceptionGroup.derive() , BaseExceptionGroup.subgroup() , and BaseExceptionGroup.split() by changing how they parse passed arguments.

gh-111654 : Fix runtime crash when some error happens in opcode LOAD_FROM_DICT_OR_DEREF .

gh-111623 : Add support for sharing tuples between interpreters using the cross-interpreter API. Patch by Anthony Shaw.

gh-111354 : The oparg of YIELD_VALUE is now 1 if the instruction is part of a yield-from or await, and 0 otherwise.

The SUSPENDED frame state is now split into SUSPENDED and SUSPENDED_YIELD_FROM . This simplifies the code in _PyGen_yf .

gh-111520 : Merge the Tier 1 (bytecode) and Tier 2 (micro-ops) interpreters together, moving the Tier 2 interpreter loop and switch into _PyEval_EvalFrameDefault() in Python/ceval.c . The Python/executor.c file is gone. Also the TIER_ONE and TIER_TWO macros are now handled by the code generator.

Beware! This changes the environment variables to enable micro-ops and their debugging to PYTHON_UOPS and PYTHON_LLTRACE .

gh-109181 : Speed up Traceback object creation by lazily compute the line number. Patch by Pablo Galindo

gh-111420 : Allow type comments in parenthesized with statements

gh-111438 : Add support for sharing floats between interpreters using the cross-interpreter API. Patch by Anthony Shaw.

gh-111435 : Add support for sharing of True and False between interpreters using the cross-interpreter API. Patch by Anthony Shaw.

gh-102388 : Fix a bug where iso2022_jp_3 and iso2022_jp_2004 codecs read out of bounds

gh-111366 : Fix an issue in the codeop that was causing SyntaxError exceptions raised in the presence of invalid syntax to not contain precise error messages. Patch by Pablo Galindo

gh-111380 : Fix a bug that was causing SyntaxWarning to appear twice when parsing if invalid syntax is encountered later. Patch by Pablo galindo

gh-111374 : Added a new environment variable PYTHON_FROZEN_MODULES . It determines whether or not frozen modules are ignored by the import machinery, equivalent of the -X frozen_modules command-line option.

gh-111354 : Remove oparg from YIELD_VALUE . Change oparg of RESUME to include information about the except-depth. These changes make it possible to simplify the code in generator close.

gh-94438 : Fix a regression that prevented jumping across is None and is not None when debugging. Patch by Savannah Ostrowski.

gh-67224 : Show source lines in tracebacks when using the -c option when running Python. Patch by Pablo Galindo

gh-111123 : Fix a bug where a global declaration in an except block is rejected when the global is used in the else block.

gh-110938 : Fix error messages for indented blocks with functions and classes with generic type parameters. Patch by Pablo Galindo

gh-109214 : Remove unnecessary instruction pointer updates before returning from frames.

gh-110912 : Correctly display the traceback for MemoryError exceptions using the traceback module. Patch by Pablo Galindo

gh-109894 : Fixed crash due to improperly initialized static MemoryError in subinterpreter.

gh-110892 : Return NULL for PyTrace_RETURN events caused by an exception

gh-110864 : Fix argument parsing by _PyArg_UnpackKeywordsWithVararg for functions defining pos-or-keyword, vararg, and kw-only parameters.

gh-109094 : Replace prev_instr on the interpreter frame by instr_ptr which points to the beginning of the instruction that is currently executing (or will execute once the frame resumes).

gh-110805 : Allow the repl to show source code and complete tracebacks. Patch by Pablo Galindo

gh-110722 : Add PYTHON_PRESITE=package.module to import a module early in the interpreter lifecycle before site.py is executed. Python needs to be built in debug mode for this option to exist.

gh-110481 : Implement biased reference counting in --disable-gil builds.

gh-110543 : Fix regression in Python 3.12 where types.CodeType.replace() would produce a broken code object if called on a module or class code object that contains a comprehension. Patch by Jelle Zijlstra.

gh-89519 : Removed chained classmethod descriptors (introduced in bpo-19072 ). This can no longer be used to wrap other descriptors such as property . The core design of this feature was flawed and caused a number of downstream problems. To “pass-through” a classmethod , consider using the __wrapped__ attribute that was added in Python 3.10.

gh-103615 : Use local events for opcode tracing

bpo-46657 : Add mimalloc memory allocator support.

gh-106718 : When PyConfig.stdlib_dir is explicitly set, it’s now respected and won’t be overridden by PyConfig.home.

gh-106905 : Fix incorrect SystemError about AST constructor recursion depth mismatch.

gh-100445 : Improve error message for unterminated strings with escapes.

bpo-45759 : Improved error messages for elif / else statements not matching any valid statements. Patch by Jeremiah Vivian.

gh-111942 : Fix SystemError in the TextIOWrapper constructor with non-encodable “errors” argument in non-debug mode.

gh-111995 : Added the NI_IDN constant to the socket module when present in C at build time for use with socket.getnameinfo() .

gh-109538 : Issue warning message instead of having RuntimeError be displayed when event loop has already been closed at StreamWriter.__del__() .

gh-111942 : Fix crashes in io.TextIOWrapper.reconfigure() when pass invalid arguments, e.g. non-string encoding.

gh-111460 : curses : restore wide character support (including curses.unget_wch() and get_wch() ) on macOS, which was unavailable due to a regression in Python 3.12.

gh-103791 : contextlib.suppress now supports suppressing exceptions raised as part of a BaseExceptionGroup , in addition to the recent support for ExceptionGroup .

gh-111835 : The mmap.mmap class now has an seekable() method that can be used when a seekable file-like object is required. The seek() method now returns the new absolute position. Patch by Donghee Na.

gh-111804 : Remove posix.fallocate() under WASI as the underlying posix_fallocate() is not available in WASI preview2.

gh-111841 : Fix truncating arguments on an embedded null character in os.putenv() and os.unsetenv() on Windows.

gh-111768 : wsgiref.util.is_hop_by_hop() is now exposed correctly in __all__ .

gh-80731 : Avoid executing the default function in cmd.Cmd in an except block

gh-111541 : Fix doctest for SyntaxError not-builtin subclasses.

gh-111719 : Add extra argument validation for alias command in pdb

gh-111482 : time : Make time.clock_gettime() and time.clock_gettime_ns() functions up to 2x faster by faster calling convention. Patch by Victor Stinner.

gh-110894 : Call loop exception handler for exceptions in client_connected_cb of asyncio.start_server() so that applications can handle it. Patch by Kumar Aditya.

gh-111531 : Fix reference leaks in bind_class() and bind_all() methods of tkinter widgets.

gh-111246 : asyncio.loop.create_unix_server() will now automatically remove the Unix socket when the server is closed.

gh-111356 : Added io.text_encoding() , io.DEFAULT_BUFFER_SIZE , and io.IncrementalNewlineDecoder to io.__all__ .

gh-66425 : Remove the code to set the REMOTE_HOST header from wsgiref module, as it is unreachable. This header is used for performance reasons, which is not necessary in the wsgiref module.

gh-111429 : Speed up pathlib.PurePath.relative_to() and is_relative_to() .

gh-111342 : Fixed typo in math.sumprod() .

gh-68166 : Remove mention of not supported “vsapi” element type in tkinter.ttk.Style.element_create() . Add tests for element_create() and other ttk.Style methods. Add examples for element_create() in the documentation.

gh-111388 : Add show_group parameter to traceback.format_exception_only() , which allows to format ExceptionGroup instances.

gh-79033 : Another attempt at fixing asyncio.Server.wait_closed() . It now blocks until both conditions are true: the server is closed, and there are no more active connections. (This means that in some cases where in 3.12.0 this function would incorrectly have returned immediately, it will now block; in particular, when there are no active connections but the server hasn’t been closed yet.)

gh-111259 : Optimize recursive wildcards in pathlib .

gh-111295 : Fix time not checking for errors when initializing.

gh-111253 : Add error checking during _socket module init.

gh-111251 : Fix _blake2 not checking for errors when initializing.

gh-111233 : Fix select not checking for errors when initializing.

gh-111230 : Fix ssl not checking for errors when initializing.

gh-111174 : Fix crash in io.BytesIO.getbuffer() called repeatedly for empty BytesIO.

gh-111187 : Postpone removal version for locale.getdefaultlocale() to Python 3.15.

gh-111159 : Fix doctest output comparison for exceptions with notes.

gh-110910 : Fix invalid state handling in asyncio.TaskGroup and asyncio.Timeout . They now raise proper RuntimeError if they are improperly used and are left in consistent state after this.

gh-111092 : Make turtledemo run without default root enabled.

gh-110944 : Support alias and convenience vars for pdb completion

gh-110745 : Added newline parameter to pathlib.Path.read_text() . Patch by Junya Okabe.

gh-84583 : Make pdb enter post-mortem mode even for SyntaxError

gh-80675 : Set f_trace_lines = True on all frames upon pdb.set_trace()

gh-110771 : Expose the setup and cleanup portions of asyncio.run_forever() as the standalone methods asyncio.run_forever_setup() and asyncio.run_forever_cleanup() . This allows for tighter integration with GUI event loops.

gh-110774 : Support setting the asyncio.Runner loop_factory kwarg in unittest.IsolatedAsyncioTestCase

gh-110392 : Fix tty.setraw() and tty.setcbreak() : previously they returned partially modified list of the original tty attributes. tty.cfmakeraw() and tty.cfmakecbreak() now make a copy of the list of special characters before modifying it.

gh-59013 : Make line number of function breakpoint more precise in pdb

gh-88434 : Emit deprecation warning for non-integer numbers in gettext functions and methods that consider plural forms even if the translation was not found.

gh-110395 : Ensure that select.kqueue() objects correctly appear as closed in forked children, to prevent operations on an invalid file descriptor.

gh-110196 : Add __reduce__ method to IPv6Address in order to keep scope_id

gh-109747 : Improve errors for unsupported look-behind patterns. Now re.error is raised instead of OverflowError or RuntimeError for too large width of look-behind pattern.

gh-109466 : Add the ipaddress.IPv4Address.ipv6_mapped property, which returns the IPv4-mapped IPv6 address.

gh-85098 : Implement the CLI of the symtable module and improve the repr of Symbol .

gh-108791 : Improved error handling in pdb command line interface, making it produce more concise error messages.

gh-105931 : Change compileall to only strip the stripdir prefix from the full path recorded in the compiled .pyc file, when the prefix matches the start of the full path in its entirety. When the prefix does not match, no stripping is performed and a warning to this effect is displayed.

Previously all path components of the stripdir prefix that matched the full path were removed, while those that did not match were left alone (including ones interspersed between matching components).

gh-107431 : Make the DictProxy and ListProxy types in multiprocessing.managers Generic Alias Types for [] use in typing contexts.

gh-72904 : Add glob.translate() . This function converts a pathname with shell-style wildcards to a regular expression.

gh-90026 : Define USE_XATTRS on Cygwin so that XATTR-related functions in the os module become available.

gh-90890 : New methods mailbox.Maildir.get_info() , mailbox.Maildir.set_info() , mailbox.Maildir.get_flags() , mailbox.Maildir.set_flags() , mailbox.Maildir.add_flag() , mailbox.Maildir.remove_flag() . These methods speed up accessing a message’s info and/or flags and are useful when it is not necessary to access the message’s contents, as when iterating over a Maildir to find messages with specific flags.

gh-102956 : Fix returning of empty byte strings after seek in zipfile module

gh-102895 : Added a parameter local_exit for code.interact() to prevent exit() and quit from closing sys.stdin and raise SystemExit .

gh-97928 : Change the behavior of tkinter.Text.count() . It now always returns an integer if one or less counting options are specified. Previously it could return a single count as a 1-tuple, an integer (only if option "update" was specified) or None if no items found. The result is now the same if wantobjects is set to 0 .

gh-96954 : Switch the storage of the unicode codepoint names to use a different data-structure, a directed acyclic word graph . This makes the unicodedata shared library about 440 KiB smaller. Contributed by Carl Friedrich Bolz-Tereick using code from the PyPy project.

gh-73561 : Omit the interface scope from an IPv6 address when used as Host header by http.client .

gh-86826 : zipinfo now supports the full range of values in the TZ string determined by RFC 8536 and detects all invalid formats. Both Python and C implementations now raise exceptions of the same type on invalid data.

gh-111808 : Make the default value of test.support.infinite_recursion() to be conditional based on whether optimizations were used when compiling the interpreter. This helps with platforms like WASI whose stack size is greatly restricted in debug builds.

gh-110722 : Gathering line coverage of standard libraries within the regression test suite is now precise, as well as much faster. Patch by Łukasz Langa.

gh-110367 : Make regrtest --verbose3 option compatible with --huntrleaks -jN options. The ./python -m test -j1 -R 3:3 --verbose3 command now works as expected. Patch by Victor Stinner.

gh-111165 : Remove no longer used functions run_unittest() and run_doctest() from the test.support module.

gh-110932 : Fix regrtest if the SOURCE_DATE_EPOCH environment variable is defined: use the variable value as the random seed. Patch by Victor Stinner.

gh-110995 : test_gdb: Fix detection of gdb built without Python scripting support. Patch by Victor Stinner.

gh-110918 : Test case matching patterns specified by options --match , --ignore , --matchfile and --ignorefile are now tested in the order of specification, and the last match determines whether the test case be run or ignored.

gh-108747 : Add unit test for usercustomize and sitecustomize hooks from site .

gh-96954 : Make make regen-unicodedata work for out-of-tree builds of CPython.

gh-112088 : Add Tools/build/regen-configure.sh script to regenerate the configure with an Ubuntu container image. The quay.io/tiran/cpython_autoconf:271 container image ( tiran/cpython_autoconf ) is no longer used. Patch by Victor Stinner.

gh-111046 : For wasi-threads, memory is now exported to fix compatibility issues with some wasm runtimes.

gh-110828 : AIX 32bit needs -latomic to build the _testcapi extension module.

gh-85283 : The errno , md5 , resource , winsound , _ctypes_test , _multiprocessing.posixshmem , _scproxy , _stat , _testimportmultiple and _uuid C extensions are now built with the limited C API . Patch by Victor Stinner.

gh-111856 : Fixes fstat() on file systems that do not support file ID requests. This includes FAT32 and exFAT.

gh-111293 : Fix os.DirEntry.inode dropping higher 64 bits of a file id on some filesystems on Windows.

gh-110913 : WindowsConsoleIO now correctly chunks large buffers without splitting up UTF-8 sequences.

gh-59703 : For macOS framework builds, in getpath.c use the system dladdr function to find the path to the shared library rather than depending on deprecated macOS APIs.

gh-110950 : Update macOS installer to include an upstream Tcl/Tk fix for the Secure coding is not enabled for restorable state! warning encountered in Tkinter on macOS 14 Sonoma.

gh-111015 : Ensure that IDLE.app and Python Launcher.app are installed with appropriate permissions on macOS builds.

gh-71383 : Update macOS installer to include an upstream Tcl/Tk fix for the ttk::ThemeChanged error encountered in Tkinter.

gh-92603 : Update macOS installer to include a fix accepted by upstream Tcl/Tk for a crash encountered after the first tkinter.Tk() instance is destroyed.

bpo-35668 : Add docstrings to the IDLE debugger module. Fix two bugs: initialize Idb.botframe (should be in Bdb); in Idb.in_rpc_code , check whether prev_frame is None before trying to use it. Greatly expand test_debugger.

gh-111903 : Argument Clinic now supports the @critical_section directive that instructs Argument Clinic to generate a critical section around the function call, which locks the self object in --disable-gil builds. Patch by Sam Gross.

gh-112026 : Add again the private _PyThreadState_UncheckedGet() function as an alias to the new public PyThreadState_GetUnchecked() function. Patch by Victor Stinner.

gh-112026 : Restore the removed _PyDict_GetItemStringWithError() function. It is used by numpy. Patch by Victor Stinner.

gh-112026 : Restore removed private C API functions, macros and structures which have no simple replacement for now:

_PyDict_GetItem_KnownHash()

_PyDict_NewPresized()

_PyHASH_BITS

_PyHASH_IMAG

_PyHASH_INF

_PyHASH_MODULUS

_PyHASH_MULTIPLIER

_PyLong_Copy()

_PyLong_FromDigits()

_PyLong_New()

_PyLong_Sign()

_PyObject_CallMethodId()

_PyObject_CallMethodNoArgs()

_PyObject_CallMethodOneArg()

_PyObject_CallOneArg()

_PyObject_EXTRA_INIT

_PyObject_FastCallDict()

_PyObject_GetAttrId()

_PyObject_Vectorcall()

_PyObject_VectorcallMethod()

_PyStack_AsDict()

_PyThread_CurrentFrames()

_PyUnicodeWriter structure

_PyUnicodeWriter_Dealloc()

_PyUnicodeWriter_Finish()

_PyUnicodeWriter_Init()

_PyUnicodeWriter_Prepare()

_PyUnicodeWriter_PrepareKind()

_PyUnicodeWriter_WriteASCIIString()

_PyUnicodeWriter_WriteChar()

_PyUnicodeWriter_WriteLatin1String()

_PyUnicodeWriter_WriteStr()

_PyUnicodeWriter_WriteSubstring()

_PyUnicode_AsString()

_PyUnicode_FromId()

_PyVectorcall_Function()

_Py_IDENTIFIER()

_Py_c_abs()

_Py_c_diff()

_Py_c_neg()

_Py_c_pow()

_Py_c_prod()

_Py_c_quot()

_Py_c_sum()

_Py_static_string()

_Py_static_string_init()

gh-112026 : Add again <ctype.h> and <unistd.h> includes in Python.h , but don’t include them in the limited C API version 3.13 and newer. Patch by Victor Stinner.

gh-111956 : Add internal-only one-time initialization API: _PyOnceFlag and _PyOnceFlag_CallOnce .

gh-111262 : Add PyDict_Pop() and PyDict_PopString() functions: remove a key from a dictionary and optionally return the removed value. This is similar to dict.pop() , but without the default value and not raising KeyError if the key missing. Patch by Stefan Behnel and Victor Stinner.

gh-111863 : Rename Py_NOGIL to Py_GIL_DISABLED . Patch by Hugo van Kemenade.

gh-111138 : Add PyList_Extend() and PyList_Clear() functions: similar to Python list.extend() and list.clear() methods. Patch by Victor Stinner.

gh-108765 : On Windows, Python.h no longer includes the <stddef.h> standard header file. If needed, it should now be included explicitly. Patch by Victor Stinner.

gh-111569 : Implement “Python Critical Sections” from PEP 703 . These are macros to help replace the GIL with per-object locks in the --disable-gil build of CPython. The macros are no-ops in the default build.

gh-111506 : In the limited C API version 3.13, Py_SET_REFCNT() function is now implemented as an opaque function call. Patch by Victor Stinner.

gh-108082 : Add PyErr_FormatUnraisable() function.

gh-110964 : Move the undocumented private _PyArg functions and _PyArg_Parser structure to internal C API ( pycore_modsupport.h ). Patch by Victor Stinner.

gh-110815 : Support non-ASCII keyword names in PyArg_ParseTupleAndKeywords() .

gh-109587 : Introduced PyUnstable_PerfTrampoline_CompileCode() , PyUnstable_PerfTrampoline_SetPersistAfterFork() and PyUnstable_CopyPerfMapFile() . These functions allow extension modules to initialize trampolines eagerly, after the application is “warmed up”. This makes it possible to have perf-trampolines running in an always-enabled fashion.

gh-85283 : Add the PySys_Audit() function to the limited C API. Patch by Victor Stinner.

gh-85283 : Add PyMem_RawMalloc() , PyMem_RawCalloc() , PyMem_RawRealloc() and PyMem_RawFree() to the limited C API. Patch by Victor Stinner.

gh-106672 : Functions PyDict_GetItem() , PyDict_GetItemString() , PyMapping_HasKey() , PyMapping_HasKeyString() , PyObject_HasAttr() , PyObject_HasAttrString() , and PySys_GetObject() , which clear all errors occurred during calling the function, report now them using sys.unraisablehook() .

gh-67565 : Remove redundant C-contiguity check in getargs.c , binascii , ssl and Argument Clinic. Patched by Stefan Krah and Furkan Onder

Python 3.13.0 alpha 1 ¶

Release date: 2023-10-13

gh-108310 : Fixed an issue where instances of ssl.SSLSocket were vulnerable to a bypass of the TLS handshake and included protections (like certificate verification) and treating sent unencrypted data as if it were post-handshake TLS encrypted data. Security issue reported as CVE-2023-40217 by Aapo Oksman. Patch by Gregory P. Smith.

gh-107774 : PEP 669 specifies that sys.monitoring.register_callback will generate an audit event. Pre-releases of Python 3.12 did not generate the audit event. This is now fixed.

gh-102988 : Reverted the email.utils security improvement change released in 3.12beta4 that unintentionally caused email.utils.getaddresses to fail to parse email addresses with a comma in the quoted name field. See gh-106669 .

gh-99108 : Refresh our new HACL* built-in hashlib code from upstream. Built-in SHA2 should be faster and an issue with SHA3 on 32-bit platforms is fixed.

gh-102509 : Start initializing ob_digit during creation of PyLongObject objects. Patch by Illia Volochii.

gh-110782 : Fix crash when typing.TypeVar is constructed with a keyword argument. Patch by Jelle Zijlstra.

gh-110752 : Reset ceval.eval_breaker in interpreter_clear()

gh-110721 : Use the traceback implementation for the default PyErr_Display() functionality. Patch by Pablo Galindo

gh-110696 : Fix incorrect error message for invalid argument unpacking. Patch by Pablo Galindo

gh-104169 : Split the tokenizer into two separate directories: - One part includes the actual lexeme producing logic and lives in Parser/lexer . - The second part wraps the lexer according to the different tokenization modes we have (string, utf-8, file, interactive, readline) and lives in Parser/tokenizer .

gh-110688 : Remove undocumented test_c_api method from set , which was only defined for testing purposes under Py_DEBUG . Now we have proper CAPI tests.

gh-104584 : Fix a reference leak when running with PYTHONUOPS or -X uops enabled.

gh-110514 : Add PY_THROW to sys.setprofile() events

gh-110489 : Optimise math.ceil() when the input is exactly a float, resulting in about a 10% improvement.

gh-110455 : Guard assert(tstate->thread_id > 0) with #ifndef HAVE_PTHREAD_STUBS . This allows for for pydebug builds to work under WASI which (currently) lacks thread support.

gh-110309 : Remove unnecessary empty constant nodes in the ast of f-string specs.

gh-110259 : Correctly identify the format spec in f-strings (with single or triple quotes) that have multiple lines in the expression part and include a formatting spec. Patch by Pablo Galindo

gh-110237 : Fix missing error checks for calls to PyList_Append in _PyEval_MatchClass .

gh-110164 : regrtest: If the SOURCE_DATE_EPOCH environment variable is defined, regrtest now disables tests randomization. Patch by Victor Stinner.

gh-109889 : Fix the compiler’s redundant NOP detection algorithm to skip over NOPs with no line number when looking for the next instruction’s lineno.

gh-109853 : sys.path[0] is now set correctly for subinterpreters.

gh-109923 : Set line number on the POP_TOP that follows a RETURN_GENERATOR .

gh-105716 : Subinterpreters now correctly handle the case where they have threads running in the background. Before, such threads would interfere with cleaning up and destroying them, as well as prevent running another script.

gh-109369 : The internal eval_breaker and supporting flags, plus the monitoring version have been merged into a single atomic integer to speed up checks.

gh-109823 : Fix bug where compiler does not adjust labels when removing an empty basic block which is a jump target.

gh-109793 : The main thread no longer exits prematurely when a subinterpreter is cleaned up during runtime finalization. The bug was a problem particularly because, when triggered, the Python process would always return with a 0 exitcode, even if it failed.

gh-109719 : Fix missing jump target labels when compiler reorders cold/warm blocks.

gh-109595 : Add -X cpu_count command line option to override return results of os.cpu_count() and os.process_cpu_count() . This option is useful for users who need to limit CPU resources of a container system without having to modify the container (application code). Patch by Donghee Na.

gh-109627 : Fix bug where the compiler does not assign a new jump target label to a duplicated small exit block.

gh-109596 : Fix some tokens in the grammar that were incorrectly marked as soft keywords. Also fix some repeated rule names and ensure that repeated rules are not allowed. Patch by Pablo Galindo

gh-109496 : On a Python built in debug mode, Py_DECREF() now calls _Py_NegativeRefcount() if the object is a dangling pointer to deallocated memory: memory filled with 0xDD “dead byte” by the debug hook on memory allocators. The fix is to check the reference count before checking for _Py_IsImmortal() . Patch by Victor Stinner.

gh-107265 : Deopt opcodes hidden by the executor when base opcode is needed

gh-109371 : Deopted instructions correctly for tool initialization and modified the incorrect assertion in instrumentation, when a previous tool already sets INSTRUCTION events

gh-105658 : Fix bug where the line trace of an except block ending with a conditional includes an excess event with the line of the conditional expression.

gh-109219 : Fix compiling type param scopes that use a name which is also free in an inner scope.

gh-109351 : Fix crash when compiling an invalid AST involving a named (walrus) expression.

gh-109341 : Fix crash when compiling an invalid AST involving a ast.TypeAlias .

gh-109195 : Fix source location for the LOAD_* instruction preceding a LOAD_SUPER_ATTR to load the super global (or shadowing variable) so that it encompasses only the name super and not the following parentheses.

gh-109256 : Opcode IDs for specialized opcodes are allocated in their own range to improve stability of the IDs for the ‘real’ opcodes.

gh-109216 : Fix possible memory leak in BUILD_MAP .

gh-109207 : Fix a SystemError in __repr__ of symtable entry object.

gh-109179 : Fix bug where the C traceback display drops notes from SyntaxError .

gh-109118 : Disallow nested scopes (lambdas, generator expressions, and comprehensions) within PEP 695 annotation scopes that are nested within classes.

gh-109156 : Add tests for de-instrumenting instructions while keeping the instrumentation for lines

gh-109114 : Relax the detection of the error message for invalid lambdas inside f-strings to not search for arbitrary replacement fields to avoid false positives. Patch by Pablo Galindo

gh-105848 : Add a new CALL_KW opcode, used for calls containing keyword arguments. Also, fix a possible crash when jumping over method calls in a debugger.

gh-109052 : Use the base opcode when comparing code objects to avoid interference from instrumentation

gh-109118 : Fix interpreter crash when a NameError is raised inside the type parameters of a generic class.

gh-88943 : Improve syntax error for non-ASCII character that follows a numerical literal. It now points on the invalid non-ASCII character, not on the valid numerical literal.

gh-108976 : Fix crash that occurs after de-instrumenting a code object in a monitoring callback.

gh-108732 : Make iteration variables of module- and class-scoped comprehensions visible to pdb and other tools that use frame.f_locals again.

gh-108959 : Fix caret placement for error locations for subscript and binary operations that involve non-semantic parentheses and spaces. Patch by Pablo Galindo

gh-104584 : Fix a crash when running with PYTHONUOPS or -X uops enabled and an error occurs during optimization.

gh-108727 : Define tp_dealloc for CounterOptimizer_Type . This fixes a segfault on deallocation.

gh-108520 : Fix multiprocessing.synchronize.SemLock.__setstate__() to properly initialize multiprocessing.synchronize.SemLock._is_fork_ctx . This fixes a regression when passing a SemLock across nested processes.

Rename multiprocessing.synchronize.SemLock.is_fork_ctx to multiprocessing.synchronize.SemLock._is_fork_ctx to avoid exposing it as public API.

gh-108654 : Restore locals shadowed by an inlined comprehension if the comprehension raises an exception.

gh-108488 : Change the initialization of inline cache entries so that the cache entry for JUMP_BACKWARD is initialized to zero, instead of the adaptive_counter_warmup() value used for all other instructions. This counter, unique among instructions, counts up from zero.

gh-108716 : Turn off deep-freezing of code objects. Modules are still frozen, so that a file system search is not needed for common modules.

gh-108614 : Add RESUME_CHECK instruction, to avoid having to handle instrumentation, signals, and contexts switches in the tier 2 execution engine.

gh-108487 : Move an assert that would cause a spurious crash in a devious case that should only trigger deoptimization.

gh-106176 : Use a WeakValueDictionary to track the lists containing the modules each thread is currently importing. This helps avoid a reference leak from keeping the list around longer than necessary. Weakrefs are used as GC can’t interrupt the cleanup.

gh-105481 : The regen-opcode build stage was removed and its work is now done in regen-cases.

gh-107901 : Fix missing line number on JUMP_BACKWARD at the end of a for loop.

gh-108113 : The compile() built-in can now accept a new flag, ast.PyCF_OPTIMIZED_AST , which is similar to ast.PyCF_ONLY_AST except that the returned AST is optimized according to the value of the optimize argument.

ast.parse() now accepts an optional argument optimize which is passed on to the compile() built-in. This makes it possible to obtain an optimized AST .

gh-107971 : Opcode IDs are generated from bytecodes.c instead of being hard coded in opcode.py.

gh-107944 : Improve error message for function calls with bad keyword arguments. Patch by Pablo Galindo

gh-108390 : Raise an exception when setting a non-local event ( RAISE , EXCEPTION_HANDLED , etc.) in sys.monitoring.set_local_events .

Fixes crash when tracing in recursive calls to Python classes.

gh-108035 : Remove the _PyCFrame struct, moving the pointer to the current interpreter frame back to the threadstate, as it was for 3.10 and earlier. The _PyCFrame existed as a performance optimization for tracing. Since PEP 669 has been implemented, this optimization no longer applies.

gh-91051 : Fix abort / segfault when using all eight type watcher slots, on platforms where char is signed by default.

gh-106581 : Fix possible assertion failures and missing instrumentation events when PYTHONUOPS or -X uops is enabled.

gh-107526 : Revert converting vars , dir , next , getattr , and iter to argument clinic.

gh-84805 : Autogenerate signature for METH_NOARGS and METH_O extension functions.

gh-107758 : Make the dump_stack() routine used by the lltrace feature (low-level interpreter debugging) robust against recursion by ensuring that it never calls a __repr__ method implemented in Python. Also make the similar output for Tier-2 uops appear on stdout (instead of stderr ), to match the lltrace code in ceval.c.

gh-107659 : Add docstrings for ctypes.pointer() and ctypes.POINTER() .

gh-105848 : Modify the bytecode so that the actual callable for a CALL is at a consistent position on the stack (regardless of whether or not bound-method-calling optimizations are active).

gh-107674 : Fixed performance regression in sys.settrace .

gh-107724 : In pre-release versions of 3.12, up to rc1, the sys.monitoring callback function for the PY_THROW event was missing the third, exception argument. That is now fixed.

gh-84436 : Skip reference count modifications for many known immortal objects.

gh-107596 : Specialize subscripting str objects by int indexes.

gh-107080 : Trace refs builds ( --with-trace-refs ) were crashing when used with isolated subinterpreters. The problematic global state has been isolated to each interpreter. Other fixing the crashes, this change does not affect users.

gh-107557 : Generate the cases needed for the barebones tier 2 abstract interpreter for optimization passes in CPython.

gh-106608 : Make _PyUOpExecutorObject variable length.

gh-100964 : Clear generators’ exception state after return to break reference cycles.

gh-107455 : Improve error messages when converting an incompatible type to ctypes.c_char_p , ctypes.c_wchar_p and ctypes.c_void_p .

gh-107263 : Increase C recursion limit for functions other than the main interpreter from 800 to 1500. This should allow functions like list.__repr__ and json.dumps to handle all the inputs that they could prior to 3.12

gh-104584 : Fix an issue which caused incorrect inline caches to be read when running with PYTHONUOPS or -X uops enabled.

gh-104432 : Fix potential unaligned memory access on C APIs involving returned sequences of char * pointers within the grp and socket modules. These were revealed using a -fsaniziter=alignment build on ARM macOS. Patch by Christopher Chavez.

gh-106078 : Isolate _decimal (apply PEP 687 ). Patch by Charlie Zhao.

gh-106898 : Add the exception as the third argument to PY_UNIND callbacks in sys.monitoring . This makes the PY_UNWIND callback consistent with the other exception handling callbacks.

gh-106895 : Raise a ValueError when a monitoring callback function returns DISABLE for events that cannot be disabled locally.

gh-106897 : Add a RERAISE event to sys.monitoring , which occurs when an exception is reraise, either explicitly by a plain raise statement, or implicitly in an except or finally block.

gh-77377 : Ensure that multiprocessing synchronization objects created in a fork context are not sent to a different process created in a spawn context. This changes a segfault into an actionable RuntimeError in the parent process.

gh-106931 : Statically allocated string objects are now interned globally instead of per-interpreter. This fixes a situation where such a string would only be interned in a single interpreter. Normal string objects are unaffected.

gh-104621 : Unsupported modules now always fail to be imported.

gh-107122 : Add dbm.ndbm.ndbm.clear() to dbm.ndbm . Patch By Donghee Na.

gh-107122 : Add dbm.gnu.gdbm.clear() to dbm.gnu . Patch By Donghee Na.

gh-107015 : The ASYNC and AWAIT tokens are removed from the Grammar, which removes the possibility of making async and await soft keywords when using feature_version<7 in ast.parse() .

gh-106917 : Fix classmethod-style super() method calls (i.e., where the second argument to super() , or the implied second argument drawn from self/cls in the case of zero-arg super, is a type) when the target of the call is not a classmethod.

gh-105699 : Python no longer crashes due an infrequent race when initializing per-interpreter interned strings. The crash would manifest when the interpreter was finalized.

gh-105699 : Python no longer crashes due to an infrequent race in setting Py_FileSystemDefaultEncoding and Py_FileSystemDefaultEncodeErrors (both deprecated), when simultaneously initializing two isolated subinterpreters. Now they are only set during runtime initialization.

gh-106908 : Fix various hangs, reference leaks, test failures, and tracing/introspection bugs when running with PYTHONUOPS or -X uops enabled.

gh-106092 : Fix a segmentation fault caused by a use-after-free bug in frame_dealloc when the trashcan delays the deallocation of a PyFrameObject .

gh-106485 : Reduce the number of materialized instances dictionaries by dematerializing them when possible.

gh-106719 : No longer suppress arbitrary errors in the __annotations__ getter and setter in the type and module types.

gh-106723 : Propagate frozen_modules to multiprocessing spawned process interpreters.

gh-104909 : Split LOAD_ATTR_INSTANCE_VALUE into micro-ops.

gh-104909 : Split LOAD_GLOBAL specializations into micro-ops.

gh-106597 : A new debug structure of offsets has been added to the _PyRuntimeState that will help out-of-process debuggers and profilers to obtain the offsets to relevant interpreter structures in a way that is agnostic of how Python was compiled and that doesn’t require copying the headers. Patch by Pablo Galindo

gh-106487 : Allow the count argument of str.replace() to be a keyword. Patch by Hugo van Kemenade.

gh-96844 : Improve error message of list.remove() . Patch by Donghee Na.

gh-81283 : Compiler now strips indents from docstrings. It reduces pyc file size 5% when the module is heavily documented. This change affects to __doc__ so tools like doctest will be affected.

gh-106396 : When the format specification of an f-string expression is empty, the parser now generates an empty ast.JoinedStr node for it instead of an one-element ast.JoinedStr with an empty string ast.Constant .

gh-100288 : Specialize LOAD_ATTR for non-descriptors on the class. Adds LOAD_ATTR_NONDESCRIPTOR_WITH_VALUES and LOAD_ATTR_NONDESCRIPTOR_NO_DICT .

gh-106008 : Fix possible reference leaks when failing to optimize comparisons with None in the bytecode compiler.

gh-106145 : Make end_lineno and end_col_offset required on type_param ast nodes.

gh-106213 : Changed the way that Emscripten call trampolines work for compatibility with Wasm/JS Promise integration.

gh-106182 : sys.getfilesystemencoding() and sys.getfilesystemencodeerrors now return interned Unicode object.

gh-106210 : Removed Emscripten import trampoline as it was no longer necessary for Pyodide.

gh-104584 : Added a new, experimental, tracing optimizer and interpreter (a.k.a. “tier 2”). This currently pessimizes, so don’t use yet – this is infrastructure so we can experiment with optimizing passes. To enable it, pass -Xuops or set PYTHONUOPS=1 . To get debug output, set PYTHONUOPSDEBUG=N where N is a debug level (0-4, where 0 is no debug output and 4 is excessively verbose).

gh-105775 : LOAD_CLOSURE is now a pseudo-op.

gh-105730 : Allow any callable other than type objects as the condition predicate in BaseExceptionGroup.split() and BaseExceptionGroup.subgroup() .

gh-105979 : Fix crash in _imp.get_frozen_object() due to improper exception handling.

gh-106003 : Add a new TO_BOOL instruction, which performs boolean conversions for POP_JUMP_IF_TRUE , POP_JUMP_IF_FALSE , and UNARY_NOT (which all expect exact bool values now). Also, modify the oparg of COMPARE_OP to include an optional “boolean conversion” flag.

gh-98931 : Ensure custom SyntaxError error messages are raised for invalid imports with multiple targets. Patch by Pablo Galindo

gh-105724 : Improve assert error messages by providing exact error range.

gh-105908 : Fixed bug where gh-99111 breaks future import barry_as_FLUFL in the Python REPL.

gh-105840 : Fix possible crashes when specializing function calls with too many __defaults__ .

gh-105831 : Fix an f-string bug, where using a debug expression (the = sign) that appears in the last line of a file results to the debug buffer that holds the expression text being one character too small.

gh-105800 : Correctly issue SyntaxWarning in f-strings if invalid sequences are used. Patch by Pablo Galindo

gh-105340 : Include the comprehension iteration variable in locals() inside a module- or class-scope comprehension.

gh-105331 : Raise ValueError if the delay argument to asyncio.sleep() is a NaN (matching time.sleep() ).

gh-105587 : The runtime can’t guarantee that immortal objects will not be mutated by Extensions. Thus, this modifies _PyStaticObject_CheckRefcnt to warn instead of asserting.

gh-105564 : Don’t include artificil newlines in the line attribute of tokens in the APIs of the tokenize module. Patch by Pablo Galindo

gh-105549 : Tokenize separately NUMBER and NAME tokens that are not ambiguous. Patch by Pablo Galindo.

gh-105588 : Fix an issue that could result in crashes when compiling malformed ast nodes.

gh-100987 : Allow objects other than code objects as the “executable” in internal frames. In the long term, this can help tools like Cython and PySpy interact more efficiently. In the shorter term, it allows us to perform some optimizations more simply.

gh-105375 : Fix bugs in the builtins module where exceptions could end up being overwritten.

gh-105375 : Fix bug in the compiler where an exception could end up being overwritten.

gh-105375 : Improve error handling in PyUnicode_BuildEncodingMap() where an exception could end up being overwritten.

gh-105486 : Change the repr of ParamSpec list of args in types.GenericAlias .

gh-105678 : Break the MAKE_FUNCTION instruction into two parts, MAKE_FUNCTION which makes the function and SET_FUNCTION_ATTRIBUTE which sets the attributes on the function. This makes the stack effect of MAKE_FUNCTION regular to ease optimization and code generation.

gh-105435 : Fix spurious newline character if file ends on a comment without a newline. Patch by Pablo Galindo

gh-105390 : Correctly raise tokenize.TokenError exceptions instead of SyntaxError for tokenize errors such as incomplete input. Patch by Pablo Galindo

gh-105259 : Don’t include newline character for trailing NEWLINE tokens emitted in the tokenize module. Patch by Pablo Galindo

gh-104635 : Eliminate redundant STORE_FAST instructions in the compiler. Patch by Donghee Na and Carl Meyer.

gh-105324 : Fix the main function of the tokenize module when reading from sys.stdin . Patch by Pablo Galindo

gh-33092 : Simplify and speed up interpreter for f-strings. Removes FORMAT_VALUE opcode. Add CONVERT_VALUE , FORMAT_SIMPLE and FORMAT_WITH_SPEC opcode. Compiler emits more efficient sequence for each format expression.

gh-105229 : Remove remaining two-codeunit superinstructions. All remaining superinstructions only take a single codeunit, simplifying instrumentation and quickening.

gh-105235 : Prevent out-of-bounds memory access during mmap.find() calls.

gh-98963 : Restore the ability for a subclass of property to define __slots__ or otherwise be dict-less by ignoring failures to set a docstring on such a class. This behavior had regressed in 3.12beta1. An AttributeError where there had not previously been one was disruptive to existing code.

gh-104812 : The “pending call” machinery now works for all interpreters, not just the main interpreter, and runs in all threads, not just the main thread. Some calls are still only done in the main thread, ergo in the main interpreter. This change does not affect signal handling nor the existing public C-API ( Py_AddPendingCall() ), which both still only target the main thread. The new functionality is meant strictly for internal use for now, since consequences of its use are not well understood yet outside some very restricted cases. This change brings the capability in line with the intention when the state was made per-interpreter several years ago.

gh-105194 : Do not escape with backslashes f-string format specifiers. Patch by Pablo Galindo

gh-105229 : Replace some dynamic superinstructions with single instruction equivalents.

gh-105162 : Fixed bug in generator.close()/throw() where an inner iterator would be ignored when the outer iterator was instrumented.

gh-105164 : Ensure annotations are set up correctly if the only annotation in a block is within a match block. Patch by Jelle Zijlstra.

gh-105148 : Make _PyASTOptimizeState internal to ast_opt.c. Make _PyAST_Optimize take two integers instead of a pointer to this struct. This avoids the need to include pycore_compile.h in ast_opt.c.

gh-104799 : Attributes of ast nodes that are lists now default to the empty list if omitted. This means that some code that previously raised TypeError when the AST node was used will now proceed with the empty list instead. Patch by Jelle Zijlstra.

gh-105111 : Remove the old trashcan macros Py_TRASHCAN_SAFE_BEGIN and Py_TRASHCAN_SAFE_END . They should be replaced by the new macros Py_TRASHCAN_BEGIN and Py_TRASHCAN_END .

gh-105035 : Fix super() calls on types with custom tp_getattro implementation (e.g. meta-types.)

gh-105017 : Show CRLF lines in the tokenize string attribute in both NL and NEWLINE tokens. Patch by Marta Gómez.

gh-105013 : Fix handling of multiline parenthesized lambdas in inspect.getsource() . Patch by Pablo Galindo

gh-105017 : Do not include an additional final NL token when parsing files having CRLF lines. Patch by Marta Gómez.

gh-104976 : Ensure that trailing DEDENT tokenize.TokenInfo objects emitted by the tokenize module are reported as in Python 3.11. Patch by Pablo Galindo

gh-104972 : Ensure that the line attribute in tokenize.TokenInfo objects in the tokenize module are always correct. Patch by Pablo Galindo

gh-104955 : Fix signature for the new __release_buffer__() slot. Patch by Jelle Zijlstra.

gh-104690 : Starting new threads and process creation through os.fork() during interpreter shutdown (such as from atexit handlers) is no longer supported. It can lead to race condition between the main Python runtime thread freeing thread states while internal threading routines are trying to allocate and use the state of just created threads. Or forked children trying to use the mid-shutdown runtime and thread state in the child process.

gh-104879 : Fix crash when accessing the __module__ attribute of type aliases defined outside a module. Patch by Jelle Zijlstra.

gh-104825 : Tokens emitted by the tokenize module do not include an implicit \n character in the line attribute anymore. Patch by Pablo Galindo

gh-104770 : If a generator returns a value upon being closed, the value is now returned by generator.close() .

gh-89091 : Raise RuntimeWarning for unawaited async generator methods like asend() , athrow() and aclose() . Patch by Kumar Aditya.

gh-96663 : Add a better, more introspect-able error message when setting attributes on classes without a __dict__ and no slot member for the attribute.

gh-93627 : Update the Python pickle module implementation to match the C implementation of the pickle module. For objects setting reduction methods like __reduce_ex__() or __reduce__() to None , pickling will result in a TypeError .

gh-101006 : Improve error handling when read marshal data.

gh-91095 : Specializes calls to most Python classes. Specifically, any class that inherits from object , or another Python class, and does not override __new__ .

The specialized instruction does the following:

Creates the object (by calling object.__new__ )

Pushes a shim frame to the frame stack (to cleanup after __init__ )

Pushes the frame for __init__ to the frame stack

Speeds up the instantiation of most Python classes.

gh-110786 : sysconfig ’s CLI now ignores BrokenPipeError , making it exit normally if its output is being piped and the pipe closes.

gh-103480 : The sysconfig module is now a package, instead of a single-file module.

gh-110733 : Micro-optimization: Avoid calling min() , max() in BaseEventLoop._run_once() .

gh-94597 : Added asyncio.EventLoop for use with the asyncio.run() loop_factory kwarg to avoid calling the asyncio policy system.

gh-110682 : runtime-checkable protocols used to consider __match_args__ a protocol member in __instancecheck__ if it was present on the protocol. Now, this attribute is ignored if it is present.

gh-110488 : Fix a couple of issues in pathlib.PurePath.with_name() : a single dot was incorrectly considered a valid name, and in PureWindowsPath , a name with an NTFS alternate data stream, like a:b , was incorrectly considered invalid.

gh-110590 : Fix a bug in _sre.compile() where TypeError would be overwritten by OverflowError when the code argument was a list of non-ints.

gh-65052 : Prevent pdb from crashing when trying to display undisplayable objects

gh-110519 : Deprecation warning about non-integer number in gettext now always refers to the line in the user code where gettext function or method is used. Previously it could refer to a line in gettext code.

gh-89902 : Deprecate non-standard format specifier “N” for decimal.Decimal . It was not documented and only supported in the C implementation.

gh-110378 : contextmanager() and asynccontextmanager() context managers now close an invalid underlying generator object that yields more then one value.

gh-106670 : In pdb , set convenience variable $_exception for post mortem debugging.

gh-110365 : Fix termios.tcsetattr() bug that was overwriting existing errors during parsing integers from term list.

gh-109653 : Slightly improve the import time of several standard-library modules by deferring imports of warnings within those modules. Patch by Alex Waygood.

gh-110273 : dataclasses.replace() now raises TypeError instead of ValueError if specify keyword argument for a field declared with init=False or miss keyword argument for required InitVar field.

gh-110249 : Add --inline-caches flag to dis command line.

gh-109653 : Fix a Python 3.12 regression in the import time of random . Patch by Alex Waygood.

gh-110222 : Add support of struct sequence objects in copy.replace() . Patched by Xuehai Pan.

gh-109649 : multiprocessing , concurrent.futures , compileall : Replace os.cpu_count() with os.process_cpu_count() to select the default number of worker threads and processes. Get the CPU affinity if supported. Patch by Victor Stinner.

gh-110150 : Fix base case handling in statistics.quantiles. Now allows a single data point.

gh-110036 : On Windows, multiprocessing Popen.terminate() now catches PermissionError and get the process exit code. If the process is still running, raise again the PermissionError . Otherwise, the process terminated as expected: store its exit code. Patch by Victor Stinner.

gh-110038 : Fixed an issue that caused KqueueSelector.select() to not return all the ready events in some cases when a file descriptor is registered for both read and write.

gh-110045 : Update the symtable module to support the new scopes introduced by PEP 695 .

gh-88402 : Add new variables to sysconfig.get_config_vars() on Windows: LIBRARY , LDLIBRARY , LIBDIR , SOABI , and Py_NOGIL .

gh-109631 : re functions such as re.findall() , re.split() , re.search() and re.sub() which perform short repeated matches can now be interrupted by user.

gh-109653 : Reduce the import time of email.utils by around 43%. This results in the import time of email.message falling by around 18%, which in turn reduces the import time of importlib.metadata by around 6%. Patch by Alex Waygood.

gh-109818 : Fix reprlib.recursive_repr() not copying __type_params__ from decorated function.

gh-109047 : concurrent.futures : The executor manager thread now catches exceptions when adding an item to the call queue . During Python finalization, creating a new thread can now raise RuntimeError . Catch the exception and call terminate_broken() in this case. Patch by Victor Stinner.

gh-109782 : Ensure the signature of os.path.isdir() is identical on all platforms. Patch by Amin Alaee.

gh-109653 : Improve import time of functools by around 13%. Patch by Alex Waygood.

gh-109590 : shutil.which() will prefer files with an extension in PATHEXT if the given mode includes os.X_OK on win32. If no PATHEXT match is found, a file without an extension in PATHEXT can be returned. This change will have shutil.which() act more similarly to previous behavior in Python 3.11.

gh-109653 : Reduce the import time of enum by over 50%. Patch by Alex Waygood.

gh-109593 : Avoid deadlocking on a reentrant call to the multiprocessing resource tracker. Such a reentrant call, though unlikely, can happen if a GC pass invokes the finalizer for a multiprocessing object such as SemLock.

gh-109653 : Reduce the import time of typing by around a third. Patch by Alex Waygood.

gh-109649 : Add os.process_cpu_count() function to get the number of logical CPUs usable by the calling thread of the current process. Patch by Victor Stinner.

gh-74481 : Add set_error_mode related constants in msvcrt module in Python debug build.

gh-109613 : Fix os.stat() and os.DirEntry.stat() : check for exceptions. Previously, on Python built in debug mode, these functions could trigger a fatal Python error (and abort the process) when a function succeeded with an exception set. Patch by Victor Stinner.

gh-109599 : Expose the type of PyCapsule objects as types.CapsuleType .

gh-109559 : Update unicodedata database to Unicode 15.1.0.

gh-109543 : Remove unnecessary hasattr() check during typing.TypedDict creation.

gh-109495 : Remove unnecessary extra __slots__ in datetime 's pure python implementation to reduce memory size, as they are defined in the superclass. Patch by James Hilton-Balfe

gh-109461 : logging : Use a context manager for lock acquisition.

gh-109096 : http.server.CGIHTTPRequestHandler has been deprecated for removal in 3.15. Its design is old and the web world has long since moved beyond CGI.

gh-109409 : Fix error when it was possible to inherit a frozen dataclass from multiple parents some of which were possibly not frozen.

gh-109375 : The pdb alias command now prevents registering aliases without arguments.

gh-109319 : Deprecate the dis.HAVE_ARGUMENT field in favour of dis.hasarg .

gh-107219 : Fix a race condition in concurrent.futures . When a process in the process pool was terminated abruptly (while the future was running or pending), close the connection write end. If the call queue is blocked on sending bytes to a worker process, closing the connection write end interrupts the send, so the queue can be closed. Patch by Victor Stinner.

gh-66143 : The codecs.CodecInfo object has been made copyable and pickleable. Patched by Robert Lehmann and Furkan Onder.

gh-109187 : pathlib.Path.resolve() now treats symlink loops like other errors: in strict mode, OSError is raised, and in non-strict mode, no exception is raised.

gh-50644 : Attempts to pickle or create a shallow or deep copy of codecs streams now raise a TypeError. Previously, copying failed with a RecursionError, while pickling produced wrong results that eventually caused unpickling to fail with a RecursionError.

gh-109174 : Add support of types.SimpleNamespace in copy.replace() .

gh-109164 : pdb : Replace getopt with argparse for parsing command line arguments.

gh-109151 : Enable readline editing features in the sqlite3 command-line interface ( python -m sqlite3 ).

gh-108987 : Fix _thread.start_new_thread() race condition. If a thread is created during Python finalization, the newly spawned thread now exits immediately instead of trying to access freed memory and lead to a crash. Patch by Victor Stinner.

gh-108682 : Enum: require names=() or type=... to create an empty enum using the functional syntax.

gh-109033 : Exceptions raised by os.utime builtin function now include the related filename

gh-108843 : Fix an issue in ast.unparse() when unparsing f-strings containing many quote types.

gh-108469 : ast.unparse() now supports new f-string syntax introduced in Python 3.12. Note that the f-string quotes are reselected for simplicity under the new syntax. (Patch by Steven Sun)

gh-108751 : Add copy.replace() function which allows to create a modified copy of an object. It supports named tuples, dataclasses, and many other objects.

gh-108682 : Enum: raise TypeError if super().__new__() is called from a custom __new__ .

gh-108278 : Deprecate passing the callback callable by keyword for the following sqlite3.Connection APIs:

set_authorizer()

set_progress_handler()

set_trace_callback()

The affected parameters will become positional-only in Python 3.15.

Patch by Erlend E. Aasland.

gh-105829 : Fix concurrent.futures.ProcessPoolExecutor deadlock

gh-108295 : Fix crashes related to use of weakrefs on typing.TypeVar .

gh-108463 : Make expressions/statements work as expected in pdb

gh-108277 : Add os.timerfd_create() , os.timerfd_settime() , os.timerfd_gettime() , os.timerfd_settime_ns() , and os.timerfd_gettime_ns() to provide a low level interface for Linux’s timer notification file descriptor.

gh-107811 : tarfile : extraction of members with overly large UID or GID (e.g. on an OS with 32-bit id_t ) now fails in the same way as failing to set the ID.

gh-64662 : Fix support for virtual tables in sqlite3.Connection.iterdump() . Patch by Aviv Palivoda.

gh-108111 : Fix a regression introduced in gh-101251 for 3.12, resulting in an incorrect offset calculation in gzip.GzipFile.seek() .

gh-108294 : time.sleep() now raises an auditing event.

gh-108278 : Deprecate passing name, number of arguments, and the callable as keyword arguments, for the following sqlite3.Connection APIs:

create_function()

create_aggregate()

gh-108322 : Speed-up NormalDist.samples() by using the inverse CDF method instead of calling random.gauss().

gh-83417 : Add the ability for venv to create a .gitignore file which causes the created environment to be ignored by Git. It is on by default when venv is called via its CLI.

gh-105736 : Harmonized the pure Python version of OrderedDict with the C version. Now, both versions set up their internal state in __new__ . Formerly, the pure Python version did the set up in __init__ .

gh-108083 : Fix bugs in the constructor of sqlite3.Connection and sqlite3.Connection.close() where exceptions could be leaked. Patch by Erlend E. Aasland.

gh-107932 : Fix dis module to properly report and display bytecode that do not have source lines.

gh-105539 : sqlite3 now emits an ResourceWarning if a sqlite3.Connection object is not closed explicitly. Patch by Erlend E. Aasland.

gh-107995 : The __module__ attribute on instances of functools.cached_property is now set to the name of the module in which the cached_property is defined, rather than “functools”. This means that doctests in cached_property docstrings are now properly collected by the doctest module. Patch by Tyler Smart.

gh-107963 : Fix multiprocessing.set_forkserver_preload() to check the given list of modules names. Patch by Donghee Na.

gh-106242 : Fixes os.path.normpath() to handle embedded null characters without truncating the path.

gh-81555 : xml.dom.minidom now only quotes " in attributes.

gh-50002 : xml.dom.minidom now preserves whitespaces in attributes.

gh-93057 : Passing more than one positional argument to sqlite3.connect() and the sqlite3.Connection constructor is deprecated. The remaining parameters will become keyword-only in Python 3.15. Patch by Erlend E. Aasland.

gh-76913 : Add merge_extra parameter/feature to logging.LoggerAdapter

gh-107913 : Fix possible losses of errno and winerror values in OSError exceptions if they were cleared or modified by the cleanup code before creating the exception object.

gh-107845 : tarfile.data_filter() now takes the location of symlinks into account when determining their target, so it will no longer reject some valid tarballs with LinkOutsideDestinationError .

gh-107812 : Extend socket’s netlink support to the FreeBSD platform.

gh-107805 : Fix signatures of module-level generated functions in turtle .

gh-107782 : pydoc is now able to show signatures which are not representable in Python, e.g. for getattr and dict.pop .

gh-56166 : Deprecate passing optional arguments maxsplit , count and flags in module-level functions re.split() , re.sub() and re.subn() as positional. They should only be passed by keyword.

gh-107710 : Speed up logging.getHandlerNames() .

gh-107715 : Fix doctest.DocTestFinder.find() in presence of class names with special characters. Patch by Gertjan van Zwieten.

gh-100814 : Passing a callable object as an option value to a Tkinter image now raises the expected TclError instead of an AttributeError.

gh-72684 : Add tkinter widget methods: tk_busy_hold() , tk_busy_configure() , tk_busy_cget() , tk_busy_forget() , tk_busy_current() , and tk_busy_status() .

gh-106684 : Raise ResourceWarning when asyncio.StreamWriter is not closed leading to memory leaks. Patch by Kumar Aditya.

gh-107465 : Add pathlib.Path.from_uri() classmethod.

gh-107077 : Seems that in some conditions, OpenSSL will return SSL_ERROR_SYSCALL instead of SSL_ERROR_SSL when a certification verification has failed, but the error parameters will still contain ERR_LIB_SSL and SSL_R_CERTIFICATE_VERIFY_FAILED . We are now detecting this situation and raising the appropriate ssl.SSLCertVerificationError . Patch by Pablo Galindo

gh-107576 : Fix types.get_original_bases() to only return __orig_bases__ if it is present on cls directly. Patch by James Hilton-Balfe.

gh-105481 : Remove opcode.is_pseudo , opcode.MIN_PSEUDO_OPCODE and opcode.MAX_PSEUDO_OPCODE , which were added in 3.12, were never documented and were not intended to be used externally.

gh-105481 : opcode.ENABLE_SPECIALIZATION (which was added in 3.12 but never documented or intended for external usage) is moved to _opcode.ENABLE_SPECIALIZATION where tests can access it.

gh-107396 : tarfiles; Fixed use before assignment of self.exception for gzip decompression

gh-107409 : Set __wrapped__ attribute in reprlib.recursive_repr() .

gh-107406 : Implement new __repr__() method for struct.Struct . Now it returns Struct(<format repr>) .

gh-107369 : Optimize textwrap.indent() . It is ~30% faster for large input. Patch by Inada Naoki.

gh-78722 : Fix issue where pathlib.Path.iterdir() did not raise OSError until iterated.

gh-105578 : Deprecate typing.AnyStr in favor of the new Type Parameter syntax. See PEP 695.

gh-62519 : Make gettext.pgettext() search plural definitions when translation is not found.

gh-107089 : Shelves opened with shelve.open() have a much faster clear() method. Patch by James Cave.

gh-82500 : Fix overflow on 32-bit systems with asyncio os.sendfile() implementation.

gh-83006 : Document behavior of shutil.disk_usage() for non-mounted filesystems on Unix.

gh-65495 : Use lowercase mail from and rcpt to in smptlib.SMTP .

gh-106186 : Do not report MultipartInvariantViolationDefect defect when the email.parser.Parser class is used to parse emails with headersonly=True .

gh-105002 : Fix invalid result from PurePath.relative_to() method when attempting to walk a “ .. ” segment in other with walk_up enabled. A ValueError exception is now raised in this case.

gh-106739 : Add the rtype_cache to the warning message (as an addition to the type of leaked objects and the number of leaked objects already included in the message) to make debugging leaked objects easier when the multiprocessing resource tracker process finds leaked objects at shutdown. This helps more quickly identify what was leaked and/or why the leaked object was not properly cleaned up.

gh-106751 : Optimize SelectSelector.select() for many iteration case. Patch By Donghee Na.

gh-106751 : Optimize _PollLikeSelector.select() for many iteration case.

gh-106751 : Optimize KqueueSelector.select() for many iteration case. Patch By Donghee Na.

gh-106831 : Fix potential missing NULL check of d2i_SSL_SESSION result in _ssl.c .

gh-105481 : The various opcode lists in the dis module are now generated from bytecodes.c instead of explicitly constructed in opcode.py.

gh-106727 : Make inspect.getsource() smarter for class for same name definitions

gh-106789 : Remove import of pprint from sysconfig .

gh-105726 : Added __slots__ to contextlib.AbstractContextManager and contextlib.AbstractAsyncContextManager so that child classes can use __slots__ .

gh-106774 : Update the bundled copy of pip to version 23.2.1.

gh-106751 : selectors : Optimize EpollSelector.select() code by moving some code outside of the loop.

gh-106752 : Fixed several bugs in zipfile.Path, including: in zipfile.Path.match() , Windows separators are no longer honored (and never were meant to be); Fixed name / suffix / suffixes / stem operations when no filename is present and the Path is not at the root of the zipfile; Reworked glob for performance and more correct matching behavior.

gh-105293 : Remove call to SSL_CTX_set_session_id_context during client side context creation in the ssl module.

gh-106734 : Disable tab completion in multiline mode of pdb

gh-105481 : Expose opcode metadata through _opcode .

gh-106670 : Add the new exceptions command to the Pdb debugger. It makes it possible to move between chained exceptions when using post mortem debugging.

gh-106602 : Add __copy__ and __deepcopy__ in enum

gh-106664 : selectors : Add _SelectorMapping.get() method and optimize _SelectorMapping.__getitem__() .

gh-106628 : Speed up parsing of emails by about 20% by not compiling a new regular expression for every single email.

gh-89427 : Set the environment variable VIRTUAL_ENV_PROMPT at venv activation, even when VIRTUAL_ENV_DISABLE_PROMPT is set.

gh-106530 : Revert a change to colorsys.rgb_to_hls() that caused division by zero for certain almost-white inputs. Patch by Terry Jan Reedy.

gh-106584 : Fix exit code for unittest if all tests are skipped. Patch by Egor Eliseev.

gh-106566 : Optimize (?!) (pattern which always fails) in regular expressions.

gh-106554 : selectors : Reduce Selector overhead by using a dict.get() to lookup file descriptors.

gh-106558 : Remove ref cycle in callers of convert_to_error() by deleting result from scope in a finally block.

gh-100502 : Add pathlib.PurePath.pathmod class attribute that stores the implementation of os.path used for low-level path operations: either posixpath or ntpath .

gh-106527 : Reduce overhead to add and remove asyncio readers and writers.

gh-106524 : Fix crash in _sre.template() with templates containing invalid group indices.

gh-106531 : Removed _legacy and the names it provided from importlib.resources : Resource , contents , is_resource , open_binary , open_text , path , read_binary , and read_text .

gh-106052 : re module: fix the matching of possessive quantifiers in the case of a subpattern containing backtracking.

gh-106510 : Improve debug output for atomic groups in regular expressions.

gh-106503 : Fix ref cycle in asyncio._SelectorSocketTransport by removing _write_ready in close .

gh-105497 : Fix flag mask inversion when unnamed flags exist.

gh-90876 : Prevent multiprocessing.spawn from failing to import in environments where sys.executable is None . This regressed in 3.11 with the addition of support for path-like objects in multiprocessing.

gh-106403 : Instances of typing.TypeVar , typing.ParamSpec , typing.ParamSpecArgs , typing.ParamSpecKwargs , and typing.TypeVarTuple once again support weak references, fixing a regression introduced in Python 3.12.0 beta 1. Patch by Jelle Zijlstra.

gh-89812 : Add private pathlib._PathBase class, which provides experimental support for virtual filesystems, and may be made public in a future version of Python.

gh-106292 : Check for an instance-dict cached value in the __get__() method of functools.cached_property() . This better matches the pre-3.12 behavior and improves compatibility for users subclassing functools.cached_property() and adding a __set__() method.

gh-106350 : Detect possible memory allocation failure in the libtommath function mp_init() used by the _tkinter module.

gh-106330 : Fix incorrect matching of empty paths in pathlib.PurePath.match() . This bug was introduced in Python 3.12.0 beta 1.

gh-106309 : Deprecate typing.no_type_check_decorator() . No major type checker ever added support for this decorator. Patch by Alex Waygood.

gh-102541 : Make pydoc.doc() catch bad module ImportError when output stream is not None .

gh-106263 : Fix crash when calling repr with a manually constructed SignalDict object. Patch by Charlie Zhao.

gh-106236 : Replace assert statements with raise RuntimeError in threading , so that _DummyThread cannot be joined even with -OO .

gh-106238 : Fix rare concurrency bug in lock acquisition by the logging package.

gh-106152 : Added PY_THROW event hook for cProfile for generators

gh-106075 : Added asyncio.taskgroups.__all__ to asyncio.__all__ for export in star imports.

gh-104527 : Zipapp will now skip over appending an archive to itself.

gh-106046 : Improve the error message from os.fspath() if called on an object where __fspath__ is set to None . Patch by Alex Waygood.

gh-105987 : Fix crash due to improper reference counting in asyncio eager task factory internal routines.

gh-105974 : Fix bug where a typing.Protocol class that had one or more non-callable members would raise TypeError when issubclass() was called against it, even if it defined a custom __subclasshook__ method. The behaviour in Python 3.11 and lower – which has now been restored – was not to raise TypeError in these situations if a custom __subclasshook__ method was defined. Patch by Alex Waygood.

gh-96145 : Reverted addition of json.AttrDict .

gh-89812 : Add pathlib.UnsupportedOperation , which is raised instead of NotImplementedError when a path operation isn’t supported.

gh-105808 : Fix a regression introduced in gh-101251 for 3.12, causing gzip.GzipFile.flush() to not flush the compressor (nor pass along the zip_mode argument).

gh-105481 : stack_effect() no longer raises an exception if an oparg is provided for an opcode that doesn’t use its arg, or when it is not provided for an opcode that does use it. In the latter case, the stack effect is returned for oparg=0 .

gh-104799 : Enable ast.unparse() to unparse function and class definitions created without the new type_params field from PEP 695 . Patch by Jelle Zijlstra.

gh-105793 : Add follow_symlinks keyword-only argument to pathlib.Path.is_dir() and is_file() , defaulting to True .

gh-105570 : Deprecate two methods of creating typing.TypedDict classes with 0 fields using the functional syntax: TD = TypedDict("TD") and TD = TypedDict("TD", None) . Both will be disallowed in Python 3.15. To create a TypedDict class with 0 fields, either use class TD(TypedDict): pass or TD = TypedDict("TD", {}) .

gh-105745 : Fix webbrowser.Konqueror.open method.

gh-105733 : ctypes : Deprecate undocumented ctypes.SetPointerType() and ctypes.ARRAY() functions. Patch by Victor Stinner.

gh-105687 : Remove deprecated re.template , re.T , re.TEMPLATE , sre_constans.SRE_FLAG_TEMPLATE .

gh-105684 : Supporting asyncio.Task.set_name() is now mandatory for third party task implementations. The undocumented _set_task_name() function (deprecated since 3.8) has been removed. Patch by Kumar Aditya.

gh-105375 : Fix a bug in _Unpickler_SetInputStream() where an exception could end up being overwritten in case of failure.

gh-105626 : Change the default return value of http.client.HTTPConnection.get_proxy_response_headers() to be None and not {} .

gh-105375 : Fix bugs in sys where exceptions could end up being overwritten because of deferred error handling.

gh-105605 : Harden pyexpat error handling during module initialisation to prevent exceptions from possibly being overwritten, and objects from being dereferenced twice.

gh-105375 : Fix bug in decimal where an exception could end up being overwritten.

gh-105375 : Fix bugs in _datetime where exceptions could be overwritten in case of module initialisation failure.

gh-105375 : Fix bugs in _ssl initialisation which could lead to leaked references and overwritten exceptions.

gh-105375 : Fix a bug in array.array where an exception could end up being overwritten.

gh-105375 : Fix bugs in _ctypes where exceptions could end up being overwritten.

gh-105375 : Fix a bug in the posix module where an exception could be overwritten.

gh-105375 : Fix bugs in _elementtree where exceptions could be overwritten.

gh-105375 : Fix bugs in zoneinfo where exceptions could be overwritten.

gh-105375 : Fix bugs in errno where exceptions could be overwritten.

gh-105566 : Deprecate creating a typing.NamedTuple class using keyword arguments to denote the fields ( NT = NamedTuple("NT", x=int, y=str) ). This will be disallowed in Python 3.15. Use the class-based syntax or the functional syntax instead.

Two methods of creating NamedTuple classes with 0 fields using the functional syntax are also deprecated, and will be disallowed in Python 3.15: NT = NamedTuple("NT") and NT = NamedTuple("NT", None) . To create a NamedTuple class with 0 fields, either use class NT(NamedTuple): pass or NT = NamedTuple("NT", []) .

gh-105545 : Remove deprecated in 3.11 webbrowser.MacOSXOSAScript._name attribute.

gh-105497 : Fix flag inversion when alias/mask members exist.

gh-105509 : typing.Annotated is now implemented as an instance of typing._SpecialForm rather than a class. This should have no user-facing impact for users of the typing module public API.

gh-105375 : Fix bugs in pickle where exceptions could be overwritten.

gh-70303 : Emit FutureWarning from pathlib.Path.glob() and rglob() if the given pattern ends with “ ** ”. In a future Python release, patterns with this ending will match both files and directories. Add a trailing slash to only match directories.

gh-105375 : Fix a bug in sqlite3 where an exception could be overwritten in the collation callback.

gh-105382 : Remove cafile , capath and cadefault parameters of the urllib.request.urlopen() function, deprecated in Python 3.6. Patch by Victor Stinner.

gh-105376 : logging : Remove undocumented and untested Logger.warn() and LoggerAdapter.warn() methods and logging.warn() function. Deprecated since Python 3.3, they were aliases to the logging.Logger.warning() method, logging.LoggerAdapter.warning() method and logging.warning() function. Patch by Victor Stinner.

gh-105332 : Revert pickling method from by-name back to by-value.

gh-104554 : Add RTSPS scheme support in urllib.parse

gh-105292 : Add option to traceback.format_exception_only() to recurse into the nested exception of a BaseExceptionGroup .

gh-105280 : Fix bug where isinstance([], collections.abc.Mapping) could evaluate to True if garbage collection happened at the wrong time. The bug was caused by changes to the implementation of typing.Protocol in Python 3.12.

gh-80480 : array : Add 'w' typecode that represents Py_UCS4 .

gh-105239 : Fix longstanding bug where issubclass(object, typing.Protocol) would evaluate to True in some edge cases. Patch by Alex Waygood.

gh-104310 : In the beta 1 release we added a utility function for extension module authors, to use when testing their module for support in multiple interpreters or under a per-interpreter GIL. The name of that function has changed from allowing_all_extensions to _incompatible_extension_module_restrictions . The default for the “disable_check” argument has change from True to False , to better match the new function name.

gh-105080 : Fixed inconsistent signature on derived classes for inspect.signature()

gh-105144 : Fix a recent regression in the typing module. The regression meant that doing class Foo(X, typing.Protocol) , where X was a class that had abc.ABCMeta as its metaclass, would then cause subsequent isinstance(1, X) calls to erroneously raise TypeError . Patch by Alex Waygood.

gh-62948 : The io.IOBase finalizer now logs the close() method errors with sys.unraisablehook . Previously, errors were ignored silently by default, and only logged in Python Development Mode or on Python built on debug mode . Patch by Victor Stinner.

gh-105096 : wave : Deprecate the getmark() , setmark() and getmarkers() methods of the wave.Wave_read and wave.Wave_write classes. They will be removed in Python 3.15. Patch by Victor Stinner.

gh-104992 : Remove the untested and undocumented unittest.TestProgram.usageExit() method, deprecated in Python 3.11. Patch by Hugo van Kemenade.

gh-104996 : Improve performance of pathlib.PurePath initialisation by deferring joining of paths when multiple arguments are given.

gh-101588 : Deprecate undocumented copy/deepcopy/pickle support for itertools.

gh-103631 : Fix pathlib.PurePosixPath(pathlib.PureWindowsPath(...)) not converting path separators to restore 3.11 compatible behavior.

gh-104947 : Make comparisons between pathlib.PureWindowsPath objects consistent across Windows and Posix to match 3.11 behavior.

gh-104773 : PEP 594 : Remove the audioop module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the aifc module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the uu module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104935 : Fix bugs with the interaction between typing.runtime_checkable() and typing.Generic that were introduced by the PEP 695 implementation. Patch by Jelle Zijlstra.

gh-104773 : PEP 594 : Remove the crypt module and its private _crypt extension, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the nis module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104898 : Add missing __slots__ to os.PathLike .

gh-104773 : PEP 594 : Remove the xdrlib module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the nntplib module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104886 : Remove the undocumented configparser.LegacyInterpolation class, deprecated in the docstring since Python 3.2, and with a deprecation warning since Python 3.11. Patch by Hugo van Kemenade.

gh-104786 : Remove kwargs-based typing.TypedDict creation

gh-104876 : Remove the turtle.RawTurtle.settiltangle() method, deprecated in docs since Python 3.1 and with a deprecation warning since Python 3.11. Patch by Hugo van Kemenade.

gh-104773 : PEP 594 : Removed the msilib package, deprecated in Python 3.11.

gh-104773 : PEP 594 : Remove the spwd module, deprecated in Python 3.11: the python-pam project can be used instead. Patch by Victor Stinner.

gh-75552 : Removed the tkinter.tix module, deprecated since Python 3.6.

gh-104773 : PEP 594 : Remove the chunk module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the mailcap module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the sunau module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104780 : PEP 594 : Remove the ossaudiodev module, deprecated in Python 3.11. Patch Victor Stinner.

gh-104773 : PEP 594 : Remove the pipes module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104873 : Add typing.get_protocol_members() to return the set of members defining a typing.Protocol . Add typing.is_protocol() to check whether a class is a typing.Protocol . Patch by Jelle Zijlstra.

gh-104874 : Document the __name__ and __supertype__ attributes of typing.NewType . Patch by Jelle Zijlstra.

gh-104835 : Removed the following unittest functions, deprecated in Python 3.11:

unittest.findTestCases()

unittest.makeSuite()

unittest.getTestCaseNames()

Use TestLoader methods instead:

unittest.TestLoader.loadTestsFromModule()

unittest.TestLoader.loadTestsFromTestCase()

unittest.TestLoader.getTestCaseNames()

Patch by Hugo van Kemenade.

gh-104804 : Remove the untested and undocumented webbrowser MacOSX class, deprecated in Python 3.11. Patch by Hugo van Kemenade.

gh-83863 : Support for using pathlib.Path objects as context managers has been removed. Before Python 3.9, exiting the context manager marked a path as “closed”, which caused some (but not all!) methods to raise when called. Since Python 3.9, using a path as a context manager does nothing.

gh-104799 : Adjust the location of the (see PEP 695 ) type_params field on ast.ClassDef , ast.AsyncFunctionDef , and ast.FunctionDef to better preserve backward compatibility. Patch by Jelle Zijlstra

gh-104797 : Allow typing.Protocol classes to inherit from collections.abc.Buffer . Patch by Jelle Zijlstra.

gh-104783 : Remove locale.resetlocale() function deprecated in Python 3.11. Patch by Victor Stinner.

gh-104780 : Remove the 2to3 program and the lib2to3 module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the telnetlib module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the imghdr module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the cgi and cgitb modules, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104773 : PEP 594 : Remove the sndhdr module, deprecated in Python 3.11. Patch by Victor Stinner.

gh-104372 : On Linux where subprocess can use the vfork() syscall for faster spawning, prevent the parent process from blocking other threads by dropping the GIL while it waits for the vfork’ed child process exec() outcome. This prevents spawning a binary from a slow filesystem from blocking the rest of the application.

gh-99108 : We now release the GIL around built-in hashlib computations of reasonable size for the SHA families and MD5 hash functions, matching what our OpenSSL backed hash computations already does.

gh-102613 : Improve performance of pathlib.Path.glob() when expanding a pattern with a non-terminal “ ** ” component by filtering walked paths through a regular expression, rather than calling os.scandir() more than once on each directory.

gh-104399 : Prepare the _tkinter module for building with Tcl 9.0 and future libtommath by replacing usage of deprecated functions mp_to_unsigned_bin_n() and mp_unsigned_bin_size() when necessary.

gh-102676 : Add fields start_offset , cache_offset , end_offset , baseopname , baseopcode , jump_target and oparg to dis.Instruction .

gh-103558 : Fixed parent argument validation mechanism of argparse . Improved test coverage.

gh-103464 : Provide helpful usage messages when parsing incorrect pdb commands.

gh-103384 : Generalize the regex pattern BaseConfigurator.INDEX_PATTERN to allow spaces and non-alphanumeric characters in keys.

gh-103124 : Added multiline statement support for pdb

gh-101162 : Forbid using builtins.issubclass() with types.GenericAlias as the first argument.

gh-103200 : Fix cache repopulation semantics of zipimport.invalidate_caches(). The cache is now repopulated upon retrieving files with an invalid cache, not when the cache is invalidated.

gh-100061 : Fix a bug that causes wrong matches for regular expressions with possessive qualifier.

gh-77609 : Add follow_symlinks argument to pathlib.Path.glob() and rglob() , defaulting to false.

gh-102541 : Hide traceback in help() prompt, when import failed.

gh-102120 : Added a stream mode to tarfile that allows for reading archives without caching info about the inner files.

gh-102029 : Deprecate passing any arguments to threading.RLock() .

gh-88233 : Refactored zipfile._strip_extra to use higher level abstractions for extras instead of a heavy-state loop.

gh-102024 : Reduce calls of _idle_semaphore.release() in concurrent.futures.thread._worker() .

gh-73435 : Add support for recursive wildcards in pathlib.PurePath.match() .

gh-84867 : unittest.TestLoader no longer loads test cases from exact unittest.TestCase and unittest.FunctionTestCase classes.

gh-99203 : Restore following CPython <= 3.10.5 behavior of shutil.make_archive() : do not create an empty archive if root_dir is not a directory, and, in that case, raise FileNotFoundError or NotADirectoryError regardless of format choice. Beyond the brought-back behavior, the function may now also raise these exceptions in dry_run mode.

gh-80480 : Emit DeprecationWarning for array ’s 'u' type code, deprecated in docs since Python 3.3.

gh-94924 : unittest.mock.create_autospec() now properly returns coroutine functions compatible with inspect.iscoroutinefunction()

gh-94777 : Fix hanging multiprocessing ProcessPoolExecutor when a child process crashes while data is being written in the call queue.

gh-92871 : Remove the typing.io and typing.re namespaces, deprecated since Python 3.8. All items are still available from the main typing module.

bpo-43633 : Improve the textual representation of IPv4-mapped IPv6 addresses ( RFC 4291 Sections 2.2, 2.5.5.2) in ipaddress . Patch by Oleksandr Pavliuk.

bpo-44850 : Improve performance of operator.methodcaller() using the PEP 590 vectorcall convention. Patch by Anthony Lee and Pieter Eendebak.

bpo-44185 : unittest.mock.mock_open() will call the close() method of the file handle mock when it is exiting from the context manager. Patch by Samet Yaslan.

bpo-40988 : Improve performance of functools.singledispatchmethod by caching the generated dispatch wrapper. Optimization suggested by frederico. Patch by @mental32, Alex Waygood and Pieter Eendebak.

bpo-41768 : unittest.mock speccing no longer calls class properties. Patch by Melanie Witt.

bpo-18319 : Ensure gettext(msg) retrieve translations even if a plural form exists. In other words: gettext(msg) == ngettext(msg, '', 1) .

bpo-17013 : Add ThreadingMock to unittest.mock that can be used to create Mock objects that can wait until they are called. Patch by Karthikeyan Singaravelan and Mario Corchero.

gh-109209 : The minimum Sphinx version required for the documentation is now 4.2.

gh-108826 : dis module command-line interface is now mentioned in documentation.

gh-107305 : Add documentation for PyInterpreterConfig and Py_NewInterpreterFromConfig() . Also clarify some of the nearby docs relative to per-interpreter GIL.

gh-107008 : Document the curses module variables LINES and COLS .

gh-106948 : Add a number of standard external names to nitpick_ignore .

gh-106232 : Make timeit doc command lines compatible with Windows by using double quotes for arguments. This works on linux and macOS also.

gh-105172 : Fixed functools.lru_cache() docstring accounting for typed argument’s different handling of str and int. Patch by Bar Harel.

gh-105052 : Update timeit doc to specify that time in seconds is just the default.

gh-89455 : Add missing documentation for the max_group_depth and max_group_width parameters and the exceptions attribute of the traceback.TracebackException class.

gh-89412 : Add missing documentation for the end_lineno and end_offset attributes of the traceback.TracebackException class.

gh-104943 : Remove mentions of old Python versions in typing.NamedTuple .

gh-54738 : Add documentation on how to localize the argparse module.

gh-102823 : Document the return type of x // y when x and y have type float .

gh-102759 : Align function signature for functools.reduce in documentation and docstring with the C implementation.

gh-110647 : Fix test_stress_modifying_handlers() of test_signal. Patch by Victor Stinner.

gh-103053 : Fix test_tools.test_freeze on FreeBSD: run “make distclean” instead of “make clean” in the copied source directory to remove also the “python” program. Patch by Victor Stinner.

gh-110167 : Fix a deadlock in test_socket when server fails with a timeout but the client is still running in its thread. Don’t hold a lock to call cleanup functions in doCleanups(). One of the cleanup function waits until the client completes, whereas the client could deadlock if it called addCleanup() in such situation. Patch by Victor Stinner.

gh-110388 : Add tests for tty .

gh-81002 : Add tests for termios .

gh-110367 : regrtest: When using worker processes (-jN) with –verbose3 option, regrtest can now display the worker output even if a worker process does crash. Previously, sys.stdout and sys.stderr were replaced and so the worker output was lost on a crash. Patch by Victor Stinner.

gh-110267 : Add tests for pickling and copying PyStructSequence objects. Patched by Xuehai Pan.

gh-110171 : libregrtest now always sets and shows random.seed , so tests are more reproducible. Use --randseed flag to pass the explicit random seed for tests.

gh-110152 : Remove Tools/scripts/run_tests.py and make hostrunnertest . Just run ./python -m test --slow-ci , make buildbottest or make test instead. Python test runner (regrtest) now handles cross-compilation and HOSTRUNNER. It also adds options to Python such fast -u -E -W default -bb when --fast-ci or --slow-ci option is used. Patch by Victor Stinner.

gh-110031 : Skip test_threading tests using thread+fork if Python is built with Address Sanitizer (ASAN). Patch by Victor Stinner.

gh-110088 : Fix test_asyncio timeouts: don’t measure the maximum duration, a test should not measure a CI performance. Only measure the minimum duration when a task has a timeout or delay. Add CLOCK_RES to test_asyncio.utils . Patch by Victor Stinner.

gh-109974 : Fix race conditions in test_threading lock tests. Wait until a condition is met rather than using time.sleep() with a hardcoded number of seconds. Patch by Victor Stinner.

gh-110033 : Fix test_interprocess_signal() of test_signal . Make sure that the subprocess.Popen object is deleted before the test raising an exception in a signal handler. Otherwise, Popen.__del__() can get the exception which is logged as Exception ignored in: ... and the test fails. Patch by Victor Stinner.

gh-109594 : Fix test_timeout() of test_concurrent_futures.test_wait. Remove the future which may or may not complete depending if it takes longer than the timeout or not. Keep the second future which does not complete before wait() timeout. Patch by Victor Stinner.

gh-109972 : Split test_gdb.py file into a test_gdb package made of multiple tests, so tests can now be run in parallel. Patch by Victor Stinner.

gh-109566 : regrtest: When --fast-ci or --slow-ci option is used, regrtest now replaces the current process with a new process to add -u -W default -bb -E options to Python. Patch by Victor Stinner.

gh-109748 : Fix test_zippath_from_non_installed_posix() of test_venv: don’t copy __pycache__/ sub-directories, because they can be modified by other Python tests running in parallel. Patch by Victor Stinner.

gh-109739 : regrtest: Fix reference leak check on Windows. Disable the load tracker on Windows in the reference leak check mode (-R option). Patch by Victor Stinner.

gh-109276 : regrtest: When a test fails with “env changed” and the –rerun option is used, the test is now re-run in verbose mode in a fresh process. Patch by Victor Stinner.

gh-103053 : Skip test_freeze_simple_script() of test_tools.test_freeze if Python is built with ./configure --enable-optimizations , which means with Profile Guided Optimization (PGO): it just makes the test too slow. The freeze tool is tested by many other CIs with other (faster) compiler flags. Patch by Victor Stinner.

gh-109580 : Skip test_perf_profiler if Python is built with ASAN, MSAN or UBSAN sanitizer. Python does crash randomly in this test on such build. Patch by Victor Stinner.

gh-109566 : regrtest: Add --fast-ci and --slow-ci options. --fast-ci uses a default timeout of 10 minutes and -u all,-cpu (skip slowest tests). --slow-ci uses a default timeout of 20 minutes and -u all (run all tests). Patch by Victor Stinner.

gh-109425 : libregrtest now decodes stdout of test worker processes with the “backslashreplace” error handler to log corrupted stdout, instead of failing with an error and not logging the stdout. Patch by Victor Stinner.

gh-109396 : Fix test_socket.test_hmac_sha1() in FIPS mode. Use a longer key: FIPS mode requires at least of at least 112 bits. The previous key was only 32 bits. Patch by Victor Stinner.

gh-104736 : Fix test_gdb on Python built with LLVM clang 16 on Linux ppc64le (ex: Fedora 38). Search patterns in gdb “bt” command output to detect when gdb fails to retrieve the traceback. For example, skip a test if Backtrace stopped: frame did not save the PC is found. Patch by Victor Stinner.

gh-109276 : libregrtest now calls random.seed() before running each test file when -r/--randomize command line option is used. Moreover, it’s also called in worker processes. It should help to make tests more deterministic. Previously, it was only called once in the main process before running all test files and it was not called in worker processes. Patch by Victor Stinner.

gh-109276 : libregrtest now uses a separated file descriptor to write test result as JSON. Previously, if a test wrote debug messages late around the JSON, the main test process failed to parse JSON. Patch by Victor Stinner.

gh-108996 : Fix and enable test_msvcrt .

gh-109237 : Fix test_site.test_underpth_basic() when the working directory contains at least one non-ASCII character: encode the ._pth file to UTF-8 and enable the UTF-8 Mode to use UTF-8 for the child process stdout. Patch by Victor Stinner.

gh-109230 : Fix test_pyexpat.test_exception() : it can now be run from a directory different than Python source code directory. Before, the test failed in this case. Skip the test if Modules/pyexpat.c source is not available. Skip also the test on Python implementations other than CPython. Patch by Victor Stinner.

gh-108996 : Add tests for msvcrt .

gh-109015 : Fix test_asyncio, test_imaplib and test_socket tests on FreeBSD if the TCP blackhole is enabled ( sysctl net.inet.tcp.blackhole ). Skip the few tests which failed with ETIMEDOUT which such non standard configuration. Currently, the FreeBSD GCP image enables TCP and UDP blackhole ( sysctl net.inet.tcp.blackhole=2 and sysctl net.inet.udp.blackhole=1 ). Patch by Victor Stinner.

gh-91960 : Skip test_gdb if gdb is unable to retrieve Python frame objects: if a frame is <optimized out> . When Python is built with “clang -Og”, gdb can fail to retrieve the frame parameter of _PyEval_EvalFrameDefault() . In this case, tests like py_bt() are likely to fail. Without getting access to Python frames, python-gdb.py is mostly clueless on retrieving the Python traceback. Moreover, test_gdb is no longer skipped on macOS if Python is built with Clang. Patch by Victor Stinner.

gh-108962 : Skip test_tempfile.test_flags() if chflags() fails with “OSError: [Errno 45] Operation not supported” (ex: on FreeBSD 13). Patch by Victor Stinner.

gh-91960 : FreeBSD 13.2 CI coverage for pull requests is now provided by Cirrus-CI (a hosted CI service that supports Linux, macOS, Windows, and FreeBSD).

gh-89392 : Removed support of test_main() function in tests. They now always use normal unittest test runner.

gh-108851 : Fix test_tomllib recursion tests for WASI buildbots: reduce the recursion limit and compute the maximum nested array/dict depending on the current available recursion limit. Patch by Victor Stinner.

gh-108851 : Add get_recursion_available() and get_recursion_depth() functions to the test.support module. Patch by Victor Stinner.

gh-108834 : Add --fail-rerun option option to regrtest: if a test failed when then passed when rerun in verbose mode, exit the process with exit code 2 (error), instead of exit code 0 (success). Patch by Victor Stinner.

gh-108834 : Rename regrtest --verbose2 option ( -w ) to --rerun . Keep --verbose2 as a deprecated alias. Patch by Victor Stinner.

gh-108834 : When regrtest reruns failed tests in verbose mode ( ./python -m test --rerun ), tests are now rerun in fresh worker processes rather than being executed in the main process. If a test does crash or is killed by a timeout, the main process can detect and handle the killed worker process. Tests are rerun in parallel if the -jN option is used to run tests in parallel. Patch by Victor Stinner.

gh-108822 : regrtest now computes statistics on all tests: successes, failures and skipped. test_netrc , test_pep646_syntax and test_xml_etree now return results in their test_main() function. Patch by Victor Stinner and Alex Waygood.

gh-108794 : The doctest.DocTestRunner.run() method now counts the number of skipped tests. Add doctest.DocTestRunner.skips and doctest.TestResults.skipped attributes. Patch by Victor Stinner.

gh-108388 : Convert test_concurrent_futures to a package of 7 sub-tests. Patch by Victor Stinner.

gh-108388 : Split test_multiprocessing_fork, test_multiprocessing_forkserver and test_multiprocessing_spawn into test packages. Each package is made of 4 sub-tests: processes, threads, manager and misc. It allows running more tests in parallel and so reduce the total test duration. Patch by Victor Stinner.

gh-105776 : Fix test_cppext when the C compiler command -std=c11 option: remove -std= options from the compiler command. Patch by Victor Stinner.

gh-107652 : Set up CIFuzz to run fuzz targets in GitHub Actions. Patch by Illia Volochii.

gh-107237 : test_logging : Fix test_udp_reconnection() by increasing the timeout from 100 ms to 5 minutes (LONG_TIMEOUT). Patch by Victor Stinner.

gh-107178 : Add the C API test for functions in the Mapping Protocol, the Sequence Protocol and some functions in the Object Protocol.

gh-106714 : test_capi: Fix test_no_FatalError_infinite_loop() to no longer write a coredump, by using test.support.SuppressCrashReport. Patch by Victor Stinner.

gh-104090 : Avoid creating a reference to the test object in collectedDurations() .

gh-106752 : Moved tests for zipfile.Path into Lib/test/test_zipfile/_path . Made zipfile._path a package.

gh-106690 : Add .coveragerc to cpython repository for use with coverage package.

gh-101634 : When running the Python test suite with -jN option, if a worker stdout cannot be decoded from the locale encoding report a failed testn so the exitcode is non-zero. Patch by Victor Stinner.

gh-105084 : When the Python build is configured --with-wheel-pkg-dir , tests requiring the setuptools and wheel wheels will search for the wheels in WHEEL_PKG_DIR .

gh-81005 : String tests are modified to reflect that str and unicode are merged in Python 3. Patch by Daniel Fortunov.

gh-103186 : Suppress and assert expected RuntimeWarnings in test_sys_settrace.py

gh-69714 : Add additional tests to calendar to achieve full test coverage.

gh-103053 : “make check-clean-src” now also checks if the “python” program is found in the source directory: fail with an error if it does exist. Patch by Victor Stinner.

gh-109191 : Fix compile error when building with recent versions of libedit.

gh-110276 : No longer ignore PROFILE_TASK failure silently: command used by Profile Guided Optimization (PGO). Patch by Victor Stinner.

gh-109566 : Remove make testall target: use make buildbottest instead. Patch by Victor Stinner.

gh-109740 : The experimental --disable-gil configure flag now includes “t” (for “threaded”) in extension ABI tags.

gh-109054 : Fix building the _testcapi extension on Linux AArch64 which requires linking to libatomic when <cpython/pyatomic.h> is used: the _Py_atomic_or_uint64() function requires libatomic __atomic_fetch_or_8() on this platform. The configure script now checks if linking to libatomic is needed and generates a new LIBATOMIC variable used to build the _testcapi extension. Patch by Victor Stinner.

gh-63760 : Fix Solaris build: no longer redefine the gethostname() function. Solaris defines the function since 2005. Patch by Victor Stinner, original patch by Jakub Kulík.

gh-108740 : Fix a race condition in make regen-all . The deepfreeze.c source and files generated by Argument Clinic are now generated or updated before generating “global objects”. Previously, some identifiers may miss depending on the order in which these files were generated. Patch by Victor Stinner.

gh-108634 : Python built with configure --with-trace-refs (tracing references) is now ABI compatible with Python release build and debug build . Patch by Victor Stinner.

gh-85283 : The _stat C extension is now built with the limited C API . Patch by Victor Stinner.

gh-108447 : Fix x86_64 GNU/Hurd build

gh-107814 : When calling find_python.bat with -q it did not properly silence the output of nuget. That is now fixed.

gh-105481 : Remove the make target regen-opcode-targets , merge its work into regen-opcode which repeats most of the calculation. This simplifies the code for the build and reduces code duplication.

gh-106881 : Check for linux/limits.h before including it in Modules/posixmodule.c .

gh-95855 : Refactor platform triplet detection code and add detection for MIPS soft float and musl libc.

gh-106962 : Detect MPI compilers in configure .

gh-106118 : Fix compilation for platforms without O_CLOEXEC . The issue was introduced with Python 3.12b1 in gh-103295 . Patch by Erlend Aasland.

gh-105875 : SQLite 3.15.2 or newer is required to build the sqlite3 extension module. Patch by Erlend Aasland.

gh-90005 : Fix a regression in configure where we could end up unintentionally linking with libbsd .

gh-102404 : Document how to perform a WASI build on Linux. Also add Tools/wasm/build_wasi.sh as a reference implementation of the docs.

gh-89886 : Autoconf 2.71 and aclocal 1.16.4 is now required to regenerate !configure .

gh-104692 : Include commoninstall as a prerequisite for bininstall

This ensures that commoninstall is completed before bininstall is started when parallel builds are used ( make -j install ), and so the python3 symlink is only installed after all standard library modules are installed.

gh-101538 : Add experimental wasi-threads support. Patch by Takashi Yamamoto.

gh-110437 : Allows overriding the source of VC redistributables so that releases can be guaranteed to never downgrade between updates.

gh-109286 : Update Windows installer to use SQLite 3.43.1.

gh-82367 : os.path.realpath() now resolves MS-DOS style file names even if the file is not accessible. Patch by Moonsik Park.

gh-109991 : Update Windows build to use OpenSSL 3.0.11.

gh-106242 : Fixes realpath() to behave consistently when passed a path containing an embedded null character on Windows. In strict mode, it now raises OSError instead of the unexpected ValueError , and in non-strict mode will make the path absolute.

gh-83180 : Changes the Python Launcher for Windows to prefer an active virtual environment when the launched script has a shebang line using a Unix-like virtual command, even if the command requests a specific version of Python.

gh-106844 : Fix integer overflow and truncating by the null character in _winapi.LCMapStringEx() which affects ntpath.normcase() .

gh-105436 : Ensure that an empty environment block is terminated by two null characters, as is required by Windows.

gh-105146 : Updated the links at the end of the installer to point to Discourse rather than the mailing lists.

gh-103646 : When installed from the Microsoft Store, pip no longer defaults to per-user installs. However, as the install directory is unwritable, it should automatically decide to do a per-user install anyway. This should resolve issues when pip is passed an option that conflicts with --user .

gh-88745 : Improve performance of shutil.copy2() by using the operating system’s CopyFile2 function. This may result in subtle changes to metadata copied along with some files, bringing them in line with normal OS behavior.

gh-104820 : Fixes stat() and related functions on file systems that do not support file ID requests. This includes FAT32 and exFAT.

gh-104803 : Add os.path.isdevdrive() to detect whether a path is on a Windows Dev Drive. Returns False on platforms that do not support Dev Drive, and is absent on non-Windows platforms.

gh-109286 : Update macOS installer to use SQLite 3.43.1.

gh-109991 : Update macOS installer to use OpenSSL 3.0.11.

gh-99079 : Update macOS installer to use OpenSSL 3.0.9.

gh-104719 : Remove IDLE’s modification of tokenize.tabsize and test other uses of tokenize data and methods.

gh-109991 : Update GitHub CI workflows to use OpenSSL 3.0.11 and multissltests to use 1.1.1w, 3.0.11, and 3.1.3.

gh-108494 : Argument Clinic now has a partial support of the Limited API : see documentation in the Python Developer’s Guide Patch by Victor Stinner.

gh-107704 : It is now possible to deprecate passing keyword arguments for keyword-or-positional parameters with Argument Clinic, using the new / [from X.Y] syntax. (To be read as “positional-only from Python version X.Y” .) See documentation in the Python Developer’s Guide for more information.

gh-107880 : Argument Clinic can now clone __init__() and __new__() methods.

gh-104683 : Add --exclude option to Argument Clinic CLI.

gh-95065 : Argument Clinic now supports overriding automatically generated signature by using directive @text_signature . See documentation in the Python Developer’s Guide

gh-107609 : Fix duplicate module check in Argument Clinic. Previously, a duplicate definition would incorrectly be silently accepted. Patch by Erlend E. Aasland.

gh-107467 : The Argument Clinic command-line tool now prints to stderr instead of stdout on failure.

gh-106970 : Fix bugs in the Argument Clinic destination <name> clear command; the destination buffers would never be cleared, and the destination directive parser would simply continue to the fault handler after processing the command. Patch by Erlend E. Aasland.

gh-106706 : Change bytecode syntax for families to remove redundant name matching pseudo syntax.

gh-106359 : Argument Clinic now explicitly forbids “kwarg splats” in function calls used as annotations.

gh-103186 : freeze now fetches CONFIG_ARGS from the original CPython instance the Makefile uses to call utility scripts. Patch by Ijtaba Hussain.

gh-95065 : It is now possible to deprecate passing parameters positionally with Argument Clinic, using the new * [from X.Y] syntax. (To be read as “keyword-only from Python version X.Y” .) See documentation in the Python Developer’s Guide for more information. Patch by Erlend E. Aasland with help from Alex Waygood, Nikita Sobolev, and Serhiy Storchaka.

gh-85283 : If the Py_LIMITED_API macro is defined, Py_BUILD_CORE , Py_BUILD_CORE_BUILTIN and Py_BUILD_CORE_MODULE macros are now undefined by <Python.h> . Patch by Victor Stinner.

gh-110289 : Add PyUnicode_EqualToUTF8AndSize() and PyUnicode_EqualToUTF8() functions.

gh-110235 : Raise TypeError for duplicate/unknown fields in PyStructSequence constructor. Patched by Xuehai Pan.

gh-110014 : Remove undocumented PY_TIMEOUT_MAX constant from the limited C API. Patch by Victor Stinner.

gh-109521 : PyImport_GetImporter() now sets RuntimeError if it fails to get sys.path_hooks or sys.path_importer_cache or they are not list and dict correspondingly. Previously it could return NULL without setting error in obscure cases, crash or raise SystemError if these attributes have wrong type.

gh-108724 : Add PyMutex internal-only lightweight locking API.

gh-85283 : Add PySys_AuditTuple() function: similar to PySys_Audit() , but pass event arguments as a Python tuple object. Patch by Victor Stinner.

gh-108867 : Add PyThreadState_GetUnchecked() function: similar to PyThreadState_Get() , but don’t kill the process with a fatal error if it is NULL. The caller is responsible to check if the result is NULL. Previously, the function was private and known as _PyThreadState_UncheckedGet() . Patch by Victor Stinner.

gh-108765 : Python.h no longer includes the <ctype.h> standard header file. If needed, it should now be included explicitly. For example, it provides isalpha() and tolower() functions which are locale dependent. Python provides locale independent functions, like Py_ISALPHA() and Py_TOLOWER() . Patch by Victor Stinner.

gh-108765 : Python.h no longer includes the <unistd.h> standard header file. If needed, it should now be included explicitly. For example, it provides the functions: close() , getpagesize() , getpid() and sysconf() . Patch by Victor Stinner.

gh-108765 : Python.h no longer includes the <ieeefp.h> standard header. It was included for the finite() function which is now provided by the <math.h> header. It should now be included explicitly if needed. Remove also the HAVE_IEEEFP_H macro. Patch by Victor Stinner.

gh-108765 : Python.h no longer includes these standard header files: <time.h> , <sys/select.h> and <sys/time.h> . If needed, they should now be included explicitly. For example, <time.h> provides the clock() and gmtime() functions, <sys/select.h> provides the select() function, and <sys/time.h> provides the futimes() , gettimeofday() and setitimer() functions. Patch by Victor Stinner.

gh-108511 : Add functions PyObject_HasAttrWithError() , PyObject_HasAttrStringWithError() , PyMapping_HasKeyWithError() and PyMapping_HasKeyStringWithError() .

gh-107073 : Add PyObject_VisitManagedDict() and PyObject_ClearManagedDict() functions which must be called by the traverse and clear functions of a type using Py_TPFLAGS_MANAGED_DICT flag. Patch by Victor Stinner.

gh-108634 : Python built with configure --with-trace-refs (tracing references) now supports the Limited API . Patch by Victor Stinner.

gh-108014 : Add PyLong_AsInt() function: similar to PyLong_AsLong() , but store the result in a C int instead of a C long . Previously, it was known as the private function _PyLong_AsInt() (with an underscore prefix). Patch by Victor Stinner.

gh-108314 : Add PyDict_ContainsString() function: same as PyDict_Contains() , but key is specified as a const char * UTF-8 encoded bytes string, rather than a PyObject * . Patch by Victor Stinner.

gh-108337 : Add atomic operations on additional data types in pyatomic.h.

gh-108014 : Add Py_IsFinalizing() function: check if the main Python interpreter is shutting down . Patch by Victor Stinner.

gh-107916 : C API functions PyErr_SetFromErrnoWithFilename() , PyErr_SetExcFromWindowsErrWithFilename() and PyErr_SetFromWindowsErrWithFilename() save now the error code before calling PyUnicode_DecodeFSDefault() .

gh-107915 : Such C API functions as PyErr_SetString() , PyErr_Format() , PyErr_SetFromErrnoWithFilename() and many others no longer crash or ignore errors if it failed to format the error message or decode the filename. Instead, they keep a corresponding error.

gh-107810 : Improve DeprecationWarning for uses of PyType_Spec with metaclasses that have custom tp_new .

gh-107249 : Implement the Py_UNUSED macro for Windows MSVC compiler. Patch by Victor Stinner.

gh-107226 : PyModule_AddObjectRef() is now only available in the limited API version 3.10 or later.

gh-106320 : Remove private _PyUnicode_AsString() alias to PyUnicode_AsUTF8() . It was kept for backward compatibility with Python 3.0 - 3.2. The PyUnicode_AsUTF8() is available since Python 3.3. The PyUnicode_AsUTF8String() function can be used to keep compatibility with Python 3.2 and older. Patch by Victor Stinner.

gh-106572 : Convert PyObject_DelAttr() and PyObject_DelAttrString() macros to functions. Patch by Victor Stinner.

gh-106307 : Add PyMapping_GetOptionalItem() function.

gh-106521 : Add PyObject_GetOptionalAttr() and PyObject_GetOptionalAttrString() functions.

gh-106320 : Remove _PyInterpreterState_Get() alias to PyInterpreterState_Get() which was kept for backward compatibility with Python 3.8. Patch by Victor Stinner.

gh-106316 : Remove cpython/pytime.h header file: it only contained private functions. Patch by Victor Stinner.

gh-106023 : Remove private _PyObject_FastCall() function: use PyObject_Vectorcall() which is available since Python 3.8 ( PEP 590 ). Patch by Victor Stinner.

gh-106168 : If Python is built in debug mode or with assertions , PyTuple_SET_ITEM() and PyList_SET_ITEM() now check the index argument with an assertion. If the assertion fails, make sure that the size is set before. Patch by Victor Stinner.

gh-106084 : Remove the old aliases to functions calling functions which were kept for backward compatibility with Python 3.8 provisional API:

_PyObject_CallMethodNoArgs() : use PyObject_CallMethodNoArgs()

_PyObject_CallMethodOneArg() : use PyObject_CallMethodOneArg()

_PyObject_CallOneArg() : use PyObject_CallOneArg()

_PyObject_FastCallDict() : use PyObject_VectorcallDict()

_PyObject_Vectorcall() : use PyObject_Vectorcall()

_PyObject_VectorcallMethod() : use PyObject_VectorcallMethod()

_PyVectorcall_Function() : use PyVectorcall_Function()

Just remove the underscore prefix to update your code. Patch by Victor Stinner.

gh-106004 : Adds PyDict_GetItemRef() and PyDict_GetItemStringRef() functions: similar to PyDict_GetItemWithError() but returning a strong reference instead of a borrowed reference . Patch by Victor Stinner.

gh-105927 : Deprecate the PyWeakref_GetObject() and PyWeakref_GET_OBJECT() functions: use the new PyWeakref_GetRef() function instead. Patch by Victor Stinner.

gh-105927 : Add PyWeakref_GetRef() function: similar to PyWeakref_GetObject() but returns a strong reference , or NULL if the referent is no longer live. Patch by Victor Stinner.

gh-105922 : Add PyImport_AddModuleRef() : similar to PyImport_AddModule() , but return a strong reference instead of a borrowed reference . Patch by Victor Stinner.

gh-105227 : The new PyType_GetDict() provides the dictionary for the given type object that is normally exposed by cls.__dict__ . Normally it’s sufficient to use tp_dict , but for the static builtin types tp_dict is now always NULL . PyType_GetDict() provides the correct dict object instead.

gh-105375 : Fix a bug in PyErr_WarnExplicit() where an exception could end up being overwritten if the API failed internally.

gh-105603 : We’ve renamed the new (in 3.12) PyInterpreterConfig.own_gil to PyInterpreterConfig.gil and changed the meaning of the value from “bool” to an integer with supported values of PyInterpreterConfig_DEFAULT_GIL , PyInterpreterConfig_SHARED_GIL , and PyInterpreterConfig_OWN_GIL . The default is “shared”.

gh-105387 : In the limited C API version 3.12, Py_INCREF() and Py_DECREF() functions are now implemented as opaque function calls to hide implementation details. Patch by Victor Stinner.

gh-105396 : Deprecate the PyImport_ImportModuleNoBlock() function which is just an alias to PyImport_ImportModule() since Python 3.3. Patch by Victor Stinner.

gh-103968 : PyType_FromMetaclass() now allows metaclasses with tp_new set to NULL .

gh-105268 : Remove the old private, undocumented and untested _PyGC_FINALIZED() macro which was kept for backward compatibility with Python 3.8 and older. Patch by Victor Stinner.

gh-105182 : Remove PyEval_AcquireLock() and PyEval_ReleaseLock() functions, deprecated in Python 3.2. Patch by Victor Stinner.

gh-105182 : Remove PyEval_InitThreads() and PyEval_ThreadsInitialized() functions, deprecated in Python 3.9. Patch by Victor Stinner.

gh-105145 : Deprecate old Python initialization functions:

PySys_ResetWarnOptions()

Py_GetExecPrefix()

Py_GetPath()

Py_GetPrefix()

Py_GetProgramFullPath()

Py_GetProgramName()

Py_GetPythonHome()

gh-85275 : PyObject_AsCharBuffer() , PyObject_AsReadBuffer() , PyObject_CheckReadBuffer() , and PyObject_AsWriteBuffer() are removed. Please migrate to new buffer protocol; PyObject_GetBuffer() and PyBuffer_Release() .

gh-105156 : Deprecate the old Py_UNICODE and PY_UNICODE_TYPE types: use directly the wchar_t type instead. Since Python 3.3, Py_UNICODE and PY_UNICODE_TYPE are just aliases to wchar_t . Patch by Victor Stinner.

gh-105145 : Remove the following old functions to configure the Python initialization, deprecated in Python 3.11:

PySys_AddWarnOptionUnicode()

PySys_AddWarnOption()

PySys_AddXOption()

PySys_HasWarnOptions()

PySys_SetPath()

Py_SetPath()

Py_SetStandardStreamEncoding()

_Py_SetProgramFullPath()

gh-105107 : Remove functions deprecated in Python 3.9.

PyEval_CallObject() , PyEval_CallObjectWithKeywords() : use PyObject_CallNoArgs() and PyObject_Call() (positional arguments must not be NULL ) instead.

PyEval_CallFunction() : use PyObject_CallFunction() instead.

PyEval_CallMethod() : use PyObject_CallMethod() instead.

PyCFunction_Call() : use PyObject_Call() instead.

gh-105115 : PyTypeObject.tp_bases (and tp_mro ) for builtin static types are now shared by all interpreters, whereas in 3.12-beta1 they were stored on PyInterpreterState . Also note that now the tuples are immortal objects.

gh-105071 : Add PyUnstable_Exc_PrepReraiseStar to the unstable C api to expose the implementation of except* .

gh-104922 : PY_SSIZE_T_CLEAN is no longer required to use '#' formats in APIs like PyArg_ParseTuple() and Py_BuildValue() . They uses Py_ssize_t for '#' regardless PY_SSIZE_T_CLEAN .

gh-104584 : Add an unstable C API for hooking in an optimizer. This is mainly internal, but marked “unstable” to allow third-party experimentation.

gh-104668 : Don’t call PyOS_InputHook or PyOS_ReadlineFunctionPointer in subinterpreters, since it’s generally difficult to avoid using global state in their registered callbacks. This also avoids situations where extensions may find themselves running in a subinterpreter they don’t support (or haven’t yet been loaded in).

bpo-42327 : Add PyModule_Add() function: similar to PyModule_AddObjectRef() and PyModule_AddObject() , but always steals a reference to the value.

bpo-40309 : Properly handle trailing spaces before closing parenthesis in Py_BuildValue() format strings.

Python 3.12.0 beta 1 ¶

Release date: 2023-05-22

gh-99889 : Fixed a security in flaw in uu.decode() that could allow for directory traversal based on the input if no out_file was specified.

gh-104049 : Do not expose the local on-disk location in directory indexes produced by http.client.SimpleHTTPRequestHandler .

gh-99108 : Upgrade built-in hashlib SHA3 implementation to a verified implementation from the HACL* project. Used when OpenSSL is not present or lacks SHA3.

gh-102153 : urllib.parse.urlsplit() now strips leading C0 control and space characters following the specification for URLs defined by WHATWG in response to CVE-2023-24329 . Patch by Illia Volochii.

gh-102856 : Implement PEP 701 changes in the tokenize module. Patch by Marta Gómez Macías and Pablo Galindo Salgado

gh-104615 : Fix wrong ordering of assignments in code like a, a = x, y . Contributed by Carl Meyer.

gh-104572 : Improve syntax error message for invalid constructs in PEP 695 contexts and in annotations when from __future__ import annotations is active.

gh-104482 : Fix three error handling bugs in ast.c’s validation of pattern matching statements.

gh-102818 : Do not add a frame to the traceback in the sys.setprofile and sys.settrace trampoline functions. This ensures that frames are not duplicated if an exception is raised in the callback function, and ensures that frames are not omitted if a C callback is used and that does not add the frame.

gh-104405 : Fix an issue where some bytecode instructions could ignore PEP 523 when “inlining” calls.

gh-103082 : Change behavior of sys.monitoring.events.LINE events in sys.monitoring : Line events now occur when a new line is reached dynamically, instead of using a static approximation, as before. This makes the behavior very similar to that of “line” events in sys.settrace . This should ease porting of tools from 3.11 to 3.12.

gh-104263 : Fix float("nan") to produce a quiet NaN on platforms (like MIPS) where the meaning of the signalling / quiet bit is inverted from its usual meaning. Also introduce a new macro Py_INFINITY matching C99’s INFINITY , and refactor internals to rely on C99’s NAN and INFINITY macros instead of hard-coding bit patterns for infinities and NaNs. Thanks Sebastian Berg.

gh-99113 : Multi-phase init extension modules may now indicate that they support running in subinterpreters that have their own GIL. This is done by using Py_MOD_PER_INTERPRETER_GIL_SUPPORTED as the value for the Py_mod_multiple_interpreters module def slot. Otherwise the module, by default, cannot be imported in such subinterpreters. (This does not affect the main interpreter or subinterpreters that do not have their own GIL.) In addition to the isolation that multi-phase init already normally requires, support for per-interpreter GIL involves one additional constraint: thread-safety. If the module has external (linked) dependencies and those libraries have any state that isn’t thread-safe then the module must do the additional work to add thread-safety. This should be an uncommon case.

gh-99113 : The GIL is now (optionally) per-interpreter. This is the fundamental change for PEP 684. This is all made possible by virtue of the isolated state of each interpreter in the process. The behavior of the main interpreter remains unchanged. Likewise, interpreters created using Py_NewInterpreter() are not affected. To get an interpreter with its own GIL, call Py_NewInterpreterFromConfig() .

gh-104108 : Multi-phase init extension modules may now indicate whether or not they actually support multiple interpreters. By default such modules are expected to support use in multiple interpreters. In the uncommon case that one does not, it may use the new Py_mod_multiple_interpreters module def slot. A value of 0 means the module does not support them. 1 means it does. The default is 1 .

gh-104142 : Fix an issue where list or tuple repetition could fail to respect PEP 683 .

gh-104078 : Improve the performance of PyObject_HasAttrString()

gh-104066 : Improve the performance of hasattr() for module objects with a missing attribute.

gh-104028 : Reduce object creation while calling callback function from gc. Patch by Donghee Na.

gh-104018 : Disallow the “z” format specifier in %-format of bytes objects.

gh-102213 : Fix performance loss when accessing an object’s attributes with __getattr__ defined.

gh-103895 : Improve handling of edge cases in showing Exception.__notes__ . Ensures that the messages always end with a newline and that string/bytes are not exploded over multiple lines. Patch by Carey Metcalfe.

gh-103907 : Don’t modify the refcounts of known immortal objects ( True , False , and None ) in the main interpreter loop.

gh-103899 : Provide a helpful hint in the TypeError message when accidentally calling a module object that has a callable attribute of the same name (such as dis.dis() or datetime.datetime ).

gh-103845 : Remove both line and instruction instrumentation before adding new ones for monitoring, to avoid newly added instrumentation being removed immediately.

gh-103763 : Implement PEP 695 , adding syntactic support for generic classes, generic functions, and type aliases.

A new type X = ... syntax is added for type aliases, which resolves at runtime to an instance of the new class typing.TypeAliasType . The value is lazily evaluated and is accessible through the .__value__ attribute. This is implemented as a new AST node ast.TypeAlias .

New syntax ( class X[T]: ... , def func[T](): ... ) is added for defining generic functions and classes. This is implemented as a new type_params attribute on the AST nodes for classes and functions. This node holds instances of the new AST classes ast.TypeVar , ast.ParamSpec , and ast.TypeVarTuple .

typing.TypeVar , typing.ParamSpec , typing.ParamSpecArgs , typing.ParamSpecKwargs , typing.TypeVarTuple , and typing.Generic are now implemented in C rather than Python.

There are new bytecode instructions LOAD_LOCALS , LOAD_CLASSDICT_OR_GLOBAL , and LOAD_CLASSDICT_OR_DEREF to support correct resolution of names in class namespaces.

Patch by Eric Traut, Larry Hastings, and Jelle Zijlstra.

gh-103801 : Adds three minor linting fixes to the wasm module caught that were caught by ruff.

gh-103793 : Optimized asyncio Task creation by deferring expensive string formatting (task name generation) from Task creation to the first time get_name is called. This makes asyncio benchmarks up to 5% faster.

gh-102310 : Change the error range for invalid bytes literals.

gh-103590 : Do not wrap a single exception raised from a try-except* construct in an ExceptionGroup .

gh-103650 : Change the perf map format to remove the ‘0x’ prefix from the addresses

gh-102856 : Implement the required C tokenizer changes for PEP 701. Patch by Pablo Galindo Salgado, Lysandros Nikolaou, Batuhan Taskaya, Marta Gómez Macías and sunmy2019.

gh-100530 : Clarify the error message raised when the called part of a class pattern isn’t actually a class.

gh-101517 : Fix bug in line numbers of instructions emitted for except* .

gh-103492 : Clarify SyntaxWarning with literal is comparison by specifying which literal is problematic, since comparisons using is with e.g. None and bool literals are idiomatic.

gh-87729 : Add LOAD_SUPER_ATTR (and a specialization for super().method() ) to speed up super().method() and super().attr . This makes super().method() roughly 2.3x faster and brings it within 20% of the performance of a simple method call. Patch by Vladimir Matveev and Carl Meyer.

gh-103488 : Change the internal offset distinguishing yield and return target addresses, so that the instruction pointer is correct for exception handling and other stack unwinding.

gh-82012 : The bitwise inversion operator ( ~ ) on bool is deprecated. It returns the bitwise inversion of the underlying int representation such that bool(~True) == True , which can be confusing. Use not for logical negation of bools. In the rare case that you really need the bitwise inversion of the underlying int , convert to int explicitly ~int(x) .

gh-77757 : Exceptions raised in a typeobject’s __set_name__ method are no longer wrapped by a RuntimeError . Context information is added to the exception as a PEP 678 note.

gh-103333 : AttributeError now retains the name attribute when pickled and unpickled.

gh-103242 : Migrate set_ecdh_curve() method not to use deprecated OpenSSL APIs. Patch by Donghee Na.

gh-103323 : We’ve replaced our use of _PyRuntime.tstate_current with a thread-local variable. This is a fairly low-level implementation detail, and there should be no change in behavior.

gh-84436 : The implementation of PEP-683 which adds Immortal Objects by using a fixed reference count that skips reference counting to make objects truly immutable.

gh-102700 : Allow built-in modules to be submodules. This allows submodules to be statically linked into a CPython binary.

gh-103082 : Implement PEP 669 Low Impact Monitoring for CPython.

gh-88691 : Reduce the number of inline CACHE entries for CALL .

gh-102500 : Make the buffer protocol accessible in Python code using the new __buffer__ and __release_buffer__ magic methods. See PEP 688 for details. Patch by Jelle Zijlstra.

gh-97933 : PEP 709 : inline list, dict and set comprehensions to improve performance and reduce bytecode size.

gh-99184 : Bypass instance attribute access of __name__ in repr of weakref.ref .

gh-98003 : Complex function calls are now faster and consume no C stack space.

bpo-39610 : len() for 0-dimensional memoryview objects (such as memoryview(ctypes.c_uint8(42)) ) now raises a TypeError . Previously this returned 1 , which was not consistent with mem_0d[0] raising an IndexError .

bpo-31821 : Fix pause_reading() to work when called from connection_made() in asyncio .

gh-104600 : functools.update_wrapper() now sets the __type_params__ attribute (added by PEP 695 ).

gh-104340 : When an asyncio pipe protocol loses its connection due to an error, and the caller doesn’t await wait_closed() on the corresponding StreamWriter , don’t log a warning about an exception that was never retrieved. After all, according to the StreamWriter.close() docs, the wait_closed() call is optional (“not mandatory”).

gh-104555 : Fix issue where an issubclass() check comparing a class X against a runtime-checkable protocol Y with non-callable members would not cause TypeError to be raised if an isinstance() call had previously been made comparing an instance of X to Y . This issue was present in edge cases on Python 3.11, but became more prominent in 3.12 due to some unrelated changes that were made to runtime-checkable protocols. Patch by Alex Waygood.

gh-104372 : Refactored the _posixsubprocess internals to avoid Python C API usage between fork and exec when marking pass_fds= file descriptors inheritable.

gh-104484 : Added case_sensitive argument to pathlib.PurePath.match()

gh-75367 : Fix data descriptor detection in inspect.getattr_static() .

gh-104536 : Fix a race condition in the internal multiprocessing.process cleanup logic that could manifest as an unintended AttributeError when calling process.close() .

gh-103857 : Update datetime deprecations’ stracktrace to point to the calling line

gh-101520 : Move the core functionality of the tracemalloc module in the Python/ folder, leaving just the module wrapper in Modules/ .

gh-104392 : Remove undocumented and unused _paramspec_tvars attribute from some classes in typing .

gh-102613 : Fix issue where pathlib.Path.glob() raised RecursionError when walking deep directory trees.

gh-103000 : Improve performance of dataclasses.asdict() for the common case where dict_factory is dict . Patch by David C Ellis.

gh-104301 : Allow leading whitespace in disambiguated statements in pdb .

gh-104139 : Teach urllib.parse.unsplit() to retain the "//" when assembling itms-services://?action=generate-bugs style Apple Platform Deployment URLs.

gh-104307 : socket.getnameinfo() now releases the GIL while contacting the DNS server

gh-104310 : Users may now use importlib.util.allowing_all_extensions() (a context manager) to temporarily disable the strict compatibility checks for importing extension modules in subinterpreters.

gh-87695 : Fix issue where pathlib.Path.glob() raised OSError when it encountered a symlink to an overly long path.

gh-104265 : Prevent possible crash by disallowing instantiation of the _csv.Reader and _csv.Writer types. The regression was introduced in 3.10.0a4 with PR 23224 ( bpo-14935 ). Patch by Radislav Chugunov.

gh-102613 : Improve performance of pathlib.Path.glob() when expanding recursive wildcards (” ** ”) by merging adjacent wildcards and de-duplicating results only when necessary.

gh-65772 : Remove unneeded comments and code in turtle.py.

gh-90208 : Fixed issue where pathlib.Path.glob() returned incomplete results when it encountered a PermissionError . This method now suppresses all OSError exceptions, except those raised from calling is_dir() on the top-level path.

gh-104144 : Optimize asyncio.TaskGroup when using asyncio.eager_task_factory() . Skip scheduling a done callback if a TaskGroup task completes eagerly.

gh-104144 : Optimize asyncio.gather() when using asyncio.eager_task_factory() to complete eagerly if all fututres completed eagerly. Avoid scheduling done callbacks for futures that complete eagerly.

gh-104114 : Fix issue where pathlib.Path.glob() returns paths using the case of non-wildcard segments for corresponding path segments, rather than the real filesystem case.

gh-104104 : Improve performance of pathlib.Path.glob() by using re.IGNORECASE to implement case-insensitive matching.

gh-104102 : Improve performance of pathlib.Path.glob() when evaluating patterns that contain '../' segments.

gh-103822 : Update the return type of weekday to the newly added Day attribute

gh-103629 : Update the repr of typing.Unpack according to PEP 692 .

gh-103963 : Make dis display the names of the args for CALL_INTRINSIC_* .

gh-104035 : Do not ignore user-defined __getstate__ and __setstate__ methods for slotted frozen dataclasses.

gh-103987 : In mmap , fix several bugs that could lead to access to memory-mapped files after they have been invalidated.

gh-103977 : Improve import time of platform module.

gh-88773 : Added turtle.teleport() to the turtle module to move a turtle to a new point without tracing a line, visible or invisible. Patch by Liam Gersten.

gh-103935 : Use io.open_code() for files to be executed instead of raw open()

gh-68968 : Fixed garbled output of assertEqual() when an input lacks final newline.

gh-100370 : Fix potential OverflowError in sqlite3.Connection.blobopen() for 32-bit builds. Patch by Erlend E. Aasland.

gh-102628 : Substitute CTRL-D with CTRL-Z in sqlite3 CLI banner when running on Windows.

gh-103636 : Module-level attributes January and February are deprecated from calendar .

gh-103583 : Isolate _multibytecodec and codecs extension modules. Patches by Erlend E. Aasland.

gh-103848 : Add checks to ensure that [ bracketed ] hosts found by urllib.parse.urlsplit() are of IPv6 or IPvFuture format.

gh-103872 : Update the bundled copy of pip to version 23.1.2.

gh-99944 : Make dis display the value of oparg of KW_NAMES .

gh-74940 : The C.UTF-8 locale is no longer converted to en_US.UTF-8, enabling the use of UTF-8 encoding on systems which have no locales installed.

gh-103861 : Fix zipfile.Zipfile creating invalid zip files when force_zip64 was used to add files to them. Patch by Carey Metcalfe.

gh-103857 : Deprecated datetime.datetime.utcnow() and datetime.datetime.utcfromtimestamp() . (Patch by Paul Ganssle)

gh-103839 : Avoid compilation error due to tommath.h not being found when building Tkinter against Tcl 8.7 built with bundled libtommath.

gh-103791 : contextlib.suppress now supports suppressing exceptions raised as part of an ExceptionGroup . If other exceptions exist on the group, they are re-raised in a group that does not contain the suppressed exceptions.

gh-90750 : Use datetime.datetime.fromisocalendar() in the implementation of datetime.datetime.strptime() , which should now accept only valid ISO dates. (Patch by Paul Ganssle)

gh-103685 : Prepare tkinter.Menu.index() for Tk 8.7 so that it does not raise TclError: expected integer but got "" when it should return None .

gh-81403 : urllib.request.CacheFTPHandler no longer raises URLError if a cached FTP instance is reused. ftplib’s endtransfer method calls voidresp to drain the connection to handle FTP instance reuse properly.

gh-103699 : Add __orig_bases__ to non-generic TypedDicts, call-based TypedDicts, and call-based NamedTuples. Other TypedDicts and NamedTuples already had the attribute.

gh-103693 : Add convenience variable feature to pdb

gh-92248 : Deprecate type , choices , and metavar parameters of argparse.BooleanOptionalAction .

gh-89415 : Add socket constants for source-specific multicast. Patch by Reese Hyde.

gh-103673 : socketserver gains ForkingUnixStreamServer and ForkingUnixDatagramServer classes. Patch by Jay Berry.

gh-103636 : Added Enum for months and days in the calendar module.

gh-84976 : Create a new Lib/_pydatetime.py file that defines the Python version of the datetime module, and make datetime import the contents of the new library only if the C implementation is missing. Currently, the full Python implementation is defined and then deleted if the C implementation is not available, slowing down import datetime unnecessarily.

gh-103596 : Attributes/methods are no longer shadowed by same-named enum members, although they may be shadowed by enum.property’s.

gh-103584 : Updated importlib.metadata with changes from importlib_metadata 5.2 through 6.5.0, including: Support installed-files.txt for Distribution.files when present. PackageMetadata now stipulates an additional get method allowing for easy querying of metadata keys that may not be present. packages_distributions now honors packages and modules with Python modules that not .py sources (e.g. .pyc , .so ). Expand protocol for PackageMetadata.get_all to match the upstream implementation of email.message.Message.get_all in python/typeshed#9620. Deprecated use of Distribution without defining abstract methods. Deprecated expectation that PackageMetadata.__getitem__ will return None for missing keys. In the future, it will raise a KeyError .

gh-103578 : Fixed a bug where pdb crashes when reading source file with different encoding by replacing io.open() with io.open_code() . The new method would also call into the hook set by PyFile_SetOpenCodeHook() .

gh-103556 : Now creating inspect.Signature objects with positional-only parameter with a default followed by a positional-or-keyword parameter without one is impossible.

gh-103559 : Update the bundled copy of pip to version 23.1.1.

gh-103548 : Improve performance of pathlib.Path.absolute() and cwd() by joining paths only when necessary. Also improve performance of pathlib.PurePath.is_absolute() on Posix by skipping path parsing and normalization.

gh-103538 : Remove _tkinter module code guarded by definition of the TK_AQUA macro which was only needed for Tk 8.4.7 or earlier and was never actually defined by any build system or documented for manual use.

gh-103533 : Update cProfile to use PEP 669 API

gh-103525 : Fix misleading exception message when mixed str and bytes arguments are supplied to pathlib.PurePath and Path .

gh-103489 : Add getconfig() and setconfig() to Connection to make configuration changes to a database connection. Patch by Erlend E. Aasland.

gh-103365 : Set default Flag boundary to STRICT and fix bitwise operations.

gh-103472 : Avoid a potential ResourceWarning in http.client.HTTPConnection by closing the proxy / tunnel’s CONNECT response explicitly.

gh-103462 : Fixed an issue with using writelines() in asyncio to send very large payloads that exceed the amount of data that can be written in one call to socket.socket.send() or socket.socket.sendmsg() , resulting in the remaining buffer being left unwritten.

gh-103449 : Fix a bug in doc string generation in dataclasses.dataclass() .

gh-103092 : Isolate _collections (apply PEP 687 ). Patch by Erlend E. Aasland.

gh-103357 : Added support for logging.Formatter defaults parameter to logging.config.dictConfig() and logging.config.fileConfig() . Patch by Bar Harel.

gh-103092 : Adapt the winreg extension module to PEP 687 .

gh-74690 : The performance of isinstance() checks against runtime-checkable protocols has been considerably improved for protocols that only have a few members. To achieve this improvement, several internal implementation details of the typing module have been refactored, including typing._ProtocolMeta.__instancecheck__ , typing._is_callable_members_only , and typing._get_protocol_attrs . Patches by Alex Waygood.

gh-74690 : The members of a runtime-checkable protocol are now considered “frozen” at runtime as soon as the class has been created. See “What’s new in Python 3.12” for more details.

gh-103256 : Fixed a bug that caused hmac to raise an exception when the requested hash algorithm was not available in OpenSSL despite being available separately as part of hashlib itself. It now falls back properly to the built-in. This could happen when, for example, your OpenSSL does not include SHA3 support and you want to compute hmac.digest(b'K', b'M', 'sha3_256') .

gh-102778 : Support sys.last_exc in idlelib .

gh-103285 : Improve performance of ast.get_source_segment() .

gh-103225 : Fix a bug in pdb when displaying line numbers of module-level source code.

gh-103092 : Adapt the msvcrt extension module to PEP 687 .

gh-103092 : Adapt the winsound extension module to PEP 687 .

gh-93910 : Remove deprecation of enum member.member access.

gh-102978 : Fixes unittest.mock.patch() not enforcing function signatures for methods decorated with @classmethod or @staticmethod when patch is called with autospec=True .

gh-103092 : Isolate _socket (apply PEP 687 ). Patch by Erlend E. Aasland.

gh-100479 : Add pathlib.PurePath.with_segments() , which creates a path object from arguments. This method is called whenever a derivative path is created, such as from pathlib.PurePath.parent . Subclasses may override this method to share information between path objects.

gh-103220 : Fix issue where os.path.join() added a slash when joining onto an incomplete UNC drive with a trailing slash on Windows.

gh-103204 : Fixes http.server accepting HTTP requests with HTTP version numbers preceded by ‘+’, or ‘-’, or with digit-separating ‘_’ characters. The length of the version numbers is also constrained.

gh-75586 : Fix various Windows-specific issues with shutil.which .

gh-103193 : Improve performance of inspect.getattr_static() . Patch by Alex Waygood.

gh-103176 : sys._current_exceptions() now returns a mapping from thread-id to an exception instance, rather than to a (typ, exc, tb) tuple.

gh-103143 : Polish the help messages and docstrings of pdb .

gh-103015 : Add entrypoint keyword-only parameter to sqlite3.Connection.load_extension() , for overriding the SQLite extension entry point. Patch by Erlend E. Aasland.

gh-103000 : Improve performance of dataclasses.astuple() and dataclasses.asdict() in cases where the contents are common Python types.

gh-102953 : The extraction methods in tarfile , and shutil.unpack_archive() , have a new a filter argument that allows limiting tar features than may be surprising or dangerous, such as creating files outside the destination directory. See Extraction filters for details.

gh-97696 : Implemented an eager task factory in asyncio. When used as a task factory on an event loop, it performs eager execution of coroutines. Coroutines that are able to complete synchronously (e.g. return or raise without blocking) are returned immediately as a finished task, and the task is never scheduled to the event loop. If the coroutine blocks, the (pending) task is scheduled and returned.

gh-81079 : Add case_sensitive keyword-only argument to pathlib.Path.glob() and rglob() .

gh-101819 : Isolate the io extension module by applying PEP 687 . Patch by Kumar Aditya, Victor Stinner, and Erlend E. Aasland.

gh-91896 : Deprecate collections.abc.ByteString

gh-101362 : Speed up pathlib.Path construction by omitting the path anchor from the internal list of path parts.

gh-102114 : Functions in the dis module that accept a source code string as argument now print a more concise traceback when the string contains a syntax or indentation error.

gh-62432 : The unittest runner will now exit with status code 5 if no tests were run. It is common for test runner misconfiguration to fail to find any tests, this should be an error.

gh-78079 : Fix incorrect normalization of UNC device path roots, and partial UNC share path roots, in pathlib.PurePath . Pathlib no longer appends a trailing slash to such paths.

gh-85984 : Add tty.cfmakeraw() and tty.cfmakecbreak() to tty and modernize, the behavior of tty.setraw() and tty.setcbreak() to use POSIX.1-2017 Chapter 11 “General Terminal Interface” flag masks by default.

gh-101688 : Implement types.get_original_bases() to provide further introspection for types.

gh-101640 : argparse.ArgumentParser now catches errors when writing messages, such as when sys.stderr is None . Patch by Oleg Iarygin.

gh-83861 : Fix datetime.astimezone method return value when invoked on a naive datetime instance that represents local time falling in a timezone transition gap. PEP 495 requires that instances with fold=1 produce earlier times than those with fold=0 in this case.

gh-89550 : Decrease execution time of some gzip file writes by 15% by adding more appropriate buffering.

gh-95299 : Remove the bundled setuptools wheel from ensurepip , and stop installing setuptools in environments created by venv .

gh-99353 : Respect the http.client.HTTPConnection .debuglevel flag in urllib.request.AbstractHTTPHandler when its constructor parameter debuglevel is not set. And do the same for *HTTPS* .

gh-98040 : Remove the long-deprecated imp module.

gh-97850 : Deprecate pkgutil.find_loader() and pkgutil.get_loader() in favor of importlib.util.find_spec() .

gh-94473 : Flatten arguments in tkinter.Canvas.coords() . It now accepts not only x1, y1, x2, y2, ... and [x1, y1, x2, y2, ...] , but also (x1, y1), (x2, y2), ... and [(x1, y1), (x2, y2), ...] .

gh-98040 : Remove more deprecated importlib APIs: find_loader() , find_module() , importlib.abc.Finder , pkgutil.ImpImporter , pkgutil.ImpLoader .

gh-96522 : Fix potential deadlock in pty.spawn()

gh-96534 : Support divert(4) added in FreeBSD 14.

gh-87474 : Fix potential file descriptor leaks in subprocess.Popen .

gh-94906 : Support multiple steps in math.nextafter() . Patch by Shantanu Jain and Matthias Gorgens.

gh-51574 : Make tempfile.mkdtemp() return absolute paths when its dir parameter is relative.

gh-94518 : Convert private _posixsubprocess.fork_exec() to use Argument Clinic.

gh-92184 : When creating zip files using zipfile , os.altsep , if not None , will always be treated as a path separator even when it is not / . Patch by Carey Metcalfe.

bpo-46797 : Deprecation warnings are now emitted for ast.Num , ast.Bytes , ast.Str , ast.NameConstant and ast.Ellipsis . These have been documented as deprecated since Python 3.8, and will be removed in Python 3.14.

bpo-44844 : Enables webbrowser to detect and launch Microsoft Edge browser.

bpo-45606 : Fixed the bug in pathlib.Path.glob() – previously a dangling symlink would not be found by this method when the pattern is an exact match, but would be found when the pattern contains a wildcard or the recursive wildcard ( ** ). With this change, a dangling symlink will be found in both cases.

bpo-23041 : Add QUOTE_STRINGS and QUOTE_NOTNULL to the suite of csv module quoting styles.

bpo-24964 : Added http.client.HTTPConnection.get_proxy_response_headers() that provides access to the HTTP headers on a proxy server response to the CONNECT request.

bpo-17258 : multiprocessing now supports stronger HMAC algorithms for inter-process connection authentication rather than only HMAC-MD5.

bpo-39744 : Make asyncio.subprocess.Process.communicate() close the subprocess’s stdin even when called with input=None .

bpo-22708 : http.client CONNECT method tunnel improvements: Use HTTP 1.1 protocol; send a matching Host: header with CONNECT, if one is not provided; convert IDN domain names to Punycode. Patch by Michael Handler.

gh-67056 : Document that the effect of registering or unregistering an atexit cleanup function from within a registered cleanup function is undefined.

gh-103629 : Mention the new way of typing **kwargs with Unpack and TypedDict introduced in PEP 692 .

gh-48241 : Clarifying documentation about the url parameter to urllib.request.urlopen and urllib.request.Request needing to be encoded properly.

gh-86094 : Add support for Unicode Path Extra Field in ZipFile. Patch by Yeojin Kim and Andrea Giudiceandrea

gh-99202 : Fix extension type from documentation for compiling in C++20 mode

gh-104494 : Update test_pack_configure_in and test_place_configure_in for changes to error message formatting in Tk 8.7.

gh-104461 : Run test_configure_screen on X11 only, since the DISPLAY environment variable and -screen option for toplevels are not useful on Tk for Win32 or Aqua.

gh-86275 : Added property-based tests to the zoneinfo tests, along with stubs for the hypothesis interface. (Patch by Paul Ganssle)

gh-103329 : Regression tests for the behaviour of unittest.mock.PropertyMock were added.

gh-102795 : fix use of poll in test_epoll’s test_control_and_wait

gh-75729 : Fix the os.spawn* tests failing on Windows when the working directory or interpreter path contains spaces.

gh-101282 : BOLT optimization is now applied to the libpython shared library if building a shared library. BOLT instrumentation and application settings can now be influenced via the BOLT_INSTRUMENT_FLAGS and BOLT_APPLY_FLAGS configure variables.

gh-99017 : PYTHON_FOR_REGEN now require Python 3.10 or newer.

gh-104490 : Define .PHONY / virtual make targets consistently and properly.

gh-104106 : Add gcc fallback of mkfifoat/mknodat for macOS. Patch by Donghee Na.

gh-103532 : The TKINTER_PROTECT_LOADTK macro is no longer defined or used in the _tkinter module. It was previously only defined when building against Tk 8.4.13 and older, but Tk older than 8.5.12 has been unsupported since gh-91152 .

gh-99069 : Extended workaround defining static_assert when missing from the libc headers to all clang and gcc builds. In particular, this fixes building on macOS <= 10.10.

gh-100220 : Changed the default value of the SHELL Makefile variable from /bin/sh to /bin/sh -e to ensure that complex recipes correctly fail after an error. Previously, make install could fail to install some files and yet return a successful result.

gh-90656 : Add platform triplets for 64-bit LoongArch:

loongarch64-linux-gnusf

loongarch64-linux-gnuf32

loongarch64-linux-gnu

Patch by Zhang Na.

gh-104623 : Update Windows installer to use SQLite 3.42.0.

gh-82814 : Fix a potential [Errno 13] Permission denied when using shutil.copystat() within Windows Subsystem for Linux (WSL) on a mounted filesystem by adding errno.EACCES to the list of ignored errors within the internal implementation.

gh-103088 : Fix virtual environment activate script having incorrect line endings for Cygwin.

gh-103088 : Fixes venvs not working in bash on Windows across different disks

gh-102997 : Update Windows installer to use SQLite 3.41.2.

gh-88013 : Fixed a bug where TypeError was raised when calling ntpath.realpath() with a bytes parameter in some cases.

gh-99834 : Update macOS installer to Tcl/Tk 8.6.13.

gh-104623 : Update macOS installer to SQLite 3.42.0.

gh-103545 : Add os.PRIO_DARWIN_THREAD , os.PRIO_DARWIN_PROCESS , os.PRIO_DARWIN_BG and os.PRIO_DARWIN_NONUI . These can be used with os.setpriority to run the process at a lower priority and make use of the efficiency cores on Apple Silicon systems.

gh-104180 : Support reading SOCKS proxy configuration from macOS System Configuration. Patch by Sam Schott.

gh-60436 : update curses textbox to additionally handle backspace using the curses.ascii.DEL key press.

gh-102997 : Update macOS installer to SQLite 3.41.2.

gh-104499 : Fix completions for Tk Aqua 8.7 (currently blank).

gh-104496 : About prints both tcl and tk versions if different (expected someday).

gh-88496 : Fix IDLE test hang on macOS.

gh-104389 : Argument Clinic C converters now accept the unused keyword, for wrapping a parameter with Py_UNUSED . Patch by Erlend E. Aasland.

gh-101291 : Added unstable C API for extracting the value of “compact” integers: PyUnstable_Long_IsCompact() and PyUnstable_Long_CompactValue() .

gh-104109 : We’ve added Py_NewInterpreterFromConfig() and PyInterpreterConfig to the public C-API (but not the stable ABI; not yet at least). The new function may be used to create a new interpreter with various features configured. The function was added to support PEP 684 (per-interpreter GIL).

gh-103968 : PyType_FromSpec() and its variants now allow creating classes whose metaclass overrides tp_new . The tp_new is ignored. This behavior is deprecated and will be disallowed in 3.14+. The new PyType_FromMetaclass() already disallows it.

gh-103743 : Add PyUnstable_Object_GC_NewWithExtraData() function that can be used to allocate additional memory after an object for data not managed by Python.

gh-103295 : Introduced PyUnstable_WritePerfMapEntry() , PyUnstable_PerfMapState_Init() and PyUnstable_PerfMapState_Fini() . These allow extension modules (JIT compilers in particular) to write to perf-map files in a thread safe manner. The Python support for the Linux perf profiler also uses these APIs to write entries in the perf-map file.

gh-103509 : Added C API for extending types whose instance memory layout is opaque: PyType_Spec.basicsize can now be zero or negative, PyObject_GetTypeData() can be used to get subclass-specific data, and Py_TPFLAGS_ITEMS_AT_END can be used to safely extend variable-size objects. See PEP 697 for details.

gh-103091 : Add a new C-API function to eagerly assign a version tag to a PyTypeObject: PyUnstable_Type_AssignVersionTag() .

gh-101408 : PyObject_GC_Resize should calculate preheader size if needed. Patch by Donghee Na.

gh-98836 : Add support of more formatting options (left aligning, octals, uppercase hexadecimals, intmax_t , ptrdiff_t , wchar_t C strings, variable width and precision) in PyUnicode_FromFormat() and PyUnicode_FromFormatV() .

gh-96803 : Add unstable C-API functions to get the code object, lasti and line number from the internal _PyInterpreterFrame in the limited API. The functions are:

PyCodeObject * PyUnstable_InterpreterFrame_GetCode(struct _PyInterpreterFrame *frame)

int PyUnstable_InterpreterFrame_GetLasti(struct _PyInterpreterFrame *frame)

int PyUnstable_InterpreterFrame_GetLine(struct _PyInterpreterFrame *frame)

Python 3.12.0 alpha 7 ¶

Release date: 2023-04-04

gh-102192 : Deprecated _PyErr_ChainExceptions in favour of _PyErr_ChainExceptions1 .

gh-89987 : Reduce the number of inline CACHE entries for BINARY_SUBSCR .

gh-102859 : Removed JUMP_IF_FALSE_OR_POP and JUMP_IF_TRUE_OR_POP instructions.

gh-101975 : Fixed stacktop value on tracing entries to avoid corruption on garbage collection.

gh-102778 : Add sys.last_exc and deprecate sys.last_type , sys.last_value and sys.last_traceback , which hold the same information in its legacy form.

gh-100982 : Replace all occurrences of COMPARE_AND_BRANCH with COMPARE_OP .

gh-102701 : Fix overflow when creating very large dict.

gh-102755 : Add PyErr_DisplayException() which takes just an exception instance, to replace the legacy PyErr_Display() which takes the (typ, exc, tb) triplet.

gh-102594 : Add note to exception raised in PyErr_SetObject when normalization fails.

gh-90997 : Shrink the number of inline CACHE entries used by LOAD_GLOBAL .

gh-102491 : Improve import time of platform by removing IronPython version parsing. The IronPython version parsing was not functional (see https://github.com/IronLanguages/ironpython3/issues/1667 ).

gh-101291 : Rearrage bits in first field (after header) of PyLongObject. * Bits 0 and 1: 1 - sign. I.e. 0 for positive numbers, 1 for zero and 2 for negative numbers. * Bit 2 reserved (probably for the immortal bit) * Bits 3+ the unsigned size.

This makes a few operations slightly more efficient, and will enable a more compact and faster 2s-complement representation of most ints in future.

gh-102397 : Fix segfault from race condition in signal handling during garbage collection. Patch by Kumar Aditya.

gh-102406 : codecs encoding/decoding errors now get the context information (which operation and which codecs) attached as PEP 678 notes instead of through chaining a new instance of the exception.

gh-102281 : Fix potential nullptr dereference and use of uninitialized memory in fileutils. Patch by Max Bachmann.

gh-102300 : Reuse operands with refcount of 1 in float specializations of BINARY_OP.

gh-102255 : Improve build support for the Xbox. Patch by Max Bachmann.

gh-102027 : Fix SSE2 and SSE3 detection in _blake2 internal module. Patch by Max Bachmann.

gh-101865 : Deprecate co_lnotab in code objects, schedule it for removal in Python 3.14

bpo-1635741 : Adapt _pickle to PEP 687 . Patch by Mohamed Koubaa and Erlend Aasland.

gh-103085 : Pure python locale.getencoding() will not warn deprecation.

gh-103068 : It’s no longer possible to register conditional breakpoints in Pdb that raise SyntaxError . Patch by Tian Gao.

gh-102549 : Don’t ignore exceptions in member type creation.

gh-103056 : Ensure final _generate_next_value_ is a staticmethod .

gh-103046 : Display current line label correctly in dis when show_caches is False and lasti points to a CACHE entry.

gh-102433 : isinstance() checks against runtime-checkable protocols now use inspect.getattr_static() rather than hasattr() to lookup whether attributes exist. This means that descriptors and __getattr__() methods are no longer unexpectedly evaluated during isinstance() checks against runtime-checkable protocols. However, it may also mean that some objects which used to be considered instances of a runtime-checkable protocol may no longer be considered instances of that protocol on Python 3.12+, and vice versa. Most users are unlikely to be affected by this change. Patch by Alex Waygood.

gh-103023 : It’s no longer possible to register expressions to display in Pdb that raise SyntaxError . Patch by Tian Gao.

gh-102947 : Improve traceback when dataclasses.fields() is called on a non-dataclass. Patch by Alex Waygood

gh-102780 : The asyncio.Timeout context manager now works reliably even when performing cleanup due to task cancellation. Previously it could raise a CancelledError instead of an TimeoutError in such cases.

gh-102871 : Remove support for obsolete browsers from webbrowser . Removed browsers include Grail, Mosaic, Netscape, Galeon, Skipstone, Iceape, Firebird, and Firefox versions 35 and below.

gh-102839 : Improve performance of math.log() arguments handling by removing the argument clinic.

gh-102828 : Add the onexc arg to shutil.rmtree() , which is like onerror but expects an exception instance rather than an exc_info tuple. Deprecate onerror .

gh-88965 : typing: Fix a bug relating to substitution in custom classes generic over a ParamSpec . Previously, if the ParamSpec was substituted with a parameters list that itself contained a TypeVar , the TypeVar in the parameters list could not be subsequently substituted. This is now fixed.

Patch by Nikita Sobolev.

gh-76846 : Fix issue where __new__() and __init__() methods of pathlib.PurePath and Path subclasses were not called in some circumstances.

gh-78530 : asyncio.wait() now accepts generators yielding tasks. Patch by Kumar Aditya.

gh-102748 : asyncio.iscoroutine() now returns False for generators as asyncio does not support legacy generator-based coroutines. Patch by Kumar Aditya.

gh-102670 : Optimized fmean(), correlation(), covariance(), and linear_regression() using the new math.sumprod() function.

gh-102615 : Typing: Improve the repr of generic aliases for classes generic over a ParamSpec . (Use square brackets to represent a parameter list.)

gh-100112 : asyncio.Task.get_coro() now always returns a coroutine when wrapping an awaitable object. Patch by Kumar Aditya.

gh-102578 : Speed up setting or deleting mutable attributes on non-dataclass subclasses of frozen dataclasses. Due to the implementation of __setattr__ and __delattr__ for frozen dataclasses, this previously had a time complexity of O ( n ). It now has a time complexity of O (1).

gh-102519 : Add os.listdrives() , os.listvolumes() and os.listmounts() functions on Windows for enumerating drives, volumes and mount points

gh-74468 : Attribute name of the extracted tarfile file object now holds filename of itself rather than of the archive it is contained in. Patch by Oleg Iarygin.

gh-102378 : Private helper method inspect._signature_strip_non_python_syntax will no longer strip / from the input string.

gh-79940 : Add inspect.getasyncgenstate() and inspect.getasyncgenlocals() . Patch by Thomas Krennwallner.

gh-102103 : Add module argument to dataclasses.make_dataclass() and make classes produced by it pickleable.

gh-102069 : Fix __weakref__ descriptor generation for custom dataclasses.

gh-102038 : Skip a stat in site if we have already found a pyvenv.cfg

gh-98886 : Fix issues when defining dataclasses that have fields with specific underscore names that aren’t clearly reserved by dataclasses .

gh-101673 : Fix a pdb bug where ll clears the changes to local variables.

gh-101313 : Added -h and –help arguments to the webbrowser CLI

gh-100372 : ssl.SSLContext.load_verify_locations() no longer incorrectly accepts some cases of trailing data when parsing DER.

gh-89727 : Fix pathlib.Path.walk RecursionError on deep directory trees by rewriting it using iteration instead of recursion.

gh-100131 : Added an optional delete keyword argument to tempfile.TemporaryDirectory .

gh-48330 : Added --durations command line option, showing the N slowest test cases. unittest.TextTestRunner and unittest.TextTestResult constructors accept a new durations keyword argument. Subclasses should take this into account or accept **kwargs . Added unittest.TestResult.addDuration() method and unittest.TestResult.collectedDurations attribute.

(Contributed by Giampaolo Rodola)

gh-98169 : Fix dataclasses.astuple() crash when collections.defaultdict is present in the attributes.

gh-96931 : Fix incorrect results from ssl.SSLSocket.shared_ciphers()

gh-95495 : When built against OpenSSL 3.0, the ssl module had a bug where it reported unauthenticated EOFs (i.e. without close_notify) as a clean TLS-level EOF. It now raises SSLEOFError , matching the behavior in previous versions of OpenSSL. The options attribute on SSLContext also no longer includes OP_IGNORE_UNEXPECTED_EOF by default. This option may be set to specify the previous OpenSSL 3.0 behavior.

gh-94684 : Now uuid.uuid3() and uuid.uuid5() functions support bytes objects as their name argument.

gh-94440 : Fix a concurrent.futures.process bug where ProcessPoolExecutor shutdown could hang after a future has been quickly submitted and canceled.

gh-72346 : Added deprecation warning to isdst parameter of email.utils.localtime() .

bpo-36305 : Fix handling of Windows filenames that resemble drives, such as ./a:b , in pathlib .

gh-103112 : Add docstring to http.client.HTTPResponse.read() to fix pydoc output.

gh-102980 : Improve test coverage on pdb .

gh-102537 : Adjust the error handling strategy in test_zoneinfo.TzPathTest.python_tzpath_context . Patch by Paul Ganssle.

gh-101377 : Improved test_locale_calendar_formatweekday of calendar.

gh-102973 : Add a dev container (along with accompanying Dockerfile) for development purposes.

gh-102711 : Fix -Wstrict-prototypes compiler warnings.

gh-102690 : Update webbrowser to fall back to Microsoft Edge instead of Internet Explorer.

gh-99726 : Improves correctness of stat results for Windows, and uses faster API when available

gh-102809 : Misc/gdbinit was removed.

gh-102013 : Add a new (unstable) C-API function for iterating over GC’able objects using a callback: PyUnstable_VisitObjects .

Python 3.12.0 alpha 6 ¶

Release date: 2023-03-07

gh-99108 : Replace builtin hashlib implementations of MD5 and SHA1 with verified ones from the HACL* project.

gh-101727 : Updated the OpenSSL version used in Windows and macOS binary release builds to 1.1.1t to address CVE-2023-0286 , CVE-2022-4303 , and CVE-2022-4303 per the OpenSSL 2023-02-07 security advisory .

gh-99108 : Replace the builtin hashlib implementations of SHA2-384 and SHA2-512 originally from LibTomCrypt with formally verified, side-channel resistant code from the HACL* project. The builtins remain a fallback only used when OpenSSL does not provide them.

gh-101283 : subprocess.Popen now uses a safer approach to find cmd.exe when launching with shell=True . Patch by Eryk Sun, based on a patch by Oleg Iarygin.

gh-102493 : Fix regression in semantics of normalisation in PyErr_SetObject .

gh-102416 : Do not memoize incorrectly automatically generated loop rules in the parser. Patch by Pablo Galindo.

gh-102356 : Fix a bug that caused a crash when deallocating deeply nested filter objects. Patch by Marta Gómez Macías.

gh-102336 : Cleanup Windows 7 specific special handling. Patch by Max Bachmann.

gh-102250 : Fixed a segfault occurring when the interpreter calls a __bool__ method that raises.

gh-102126 : Fix deadlock at shutdown when clearing thread states if any finalizer tries to acquire the runtime head lock. Patch by Kumar Aditya.

gh-102027 : Use GetCurrentProcessId on Windows when getpid is unavailable. Patch by Max Bachmann.

gh-102056 : Fix error handling bugs in interpreter’s exception printing code, which could cause a crash on infinite recursion.

gh-100982 : Restrict the scope of the FOR_ITER_RANGE instruction to the scope of the original FOR_ITER instruction, to allow instrumentation.

gh-101967 : Fix possible segfault in positional_only_passed_as_keyword function, when new list created.

gh-101952 : Fix possible segfault in BUILD_SET opcode, when new set created.

gh-74895 : socket.getaddrinfo no longer raises OverflowError for int port values outside of the C long range. Out of range values are left up to the underlying string based C library API to report. A socket.gaierror SAI_SERVICE may occur instead, or no error at all as not all platform C libraries generate an error.

gh-101799 : Add CALL_INTRINSIC_2 and use it instead of PREP_RERAISE_STAR .

gh-101857 : Fix xattr support detection on Linux systems by widening the check to linux, not just glibc. This fixes support for musl.

gh-84783 : Make the slice object hashable. Patch by Will Bradshaw and Furkan Onder.

gh-87849 : Change the SEND instruction to leave the receiver on the stack. This allows the specialized form of SEND to skip the chain of C calls and jump directly to the RESUME in the generator or coroutine.

gh-101765 : Fix SystemError / segmentation fault in iter __reduce__ when internal access of builtins.__dict__ keys mutates the iter object.

gh-101430 : Update tracemalloc to handle presize of object properly. Patch by Donghee Na.

gh-101696 : Invalidate type version tag in _PyStaticType_Dealloc for static types, avoiding bug where a false cache hit could crash the interpreter. Patch by Kumar Aditya.

gh-101632 : Adds a new RETURN_CONST instruction.

gh-100719 : Remove gi_code field from generator (and coroutine and async generator) objects as it is redundant. The frame already includes a reference to the code object.

gh-98627 : When an interpreter is configured to check (and only then), importing an extension module will now fail when the extension does not support multiple interpreters (i.e. doesn’t implement PEP 489 multi-phase init). This does not apply to the main interpreter, nor to subinterpreters created with Py_NewInterpreter() .

gh-102302 : Micro-optimise hashing of inspect.Parameter , reducing the time it takes to hash an instance by around 40%.

gh-101979 : Fix a bug where parentheses in the metavar argument to argparse.ArgumentParser.add_argument() were dropped. Patch by Yeojin Kim.

gh-91038 : platform.platform() now has boolean default arguments.

gh-81652 : Add mmap.MAP_ALIGNED_SUPER FreeBSD and mmap.MAP_CONCEAL OpenBSD constants to mmap . Patch by Yeojin Kim.

gh-102179 : Fix os.dup2() error message for negative fds.

gh-101961 : For the binary mode, fileinput.hookcompressed() doesn’t set the encoding value even if the value is None . Patch by Gihwan Kim.

gh-101936 : The default value of fp becomes io.BytesIO if HTTPError is initialized without a designated fp parameter. Patch by Long Vo.

gh-101566 : In zipfile, sync Path with zipp 3.14 , including fix for extractall on the underlying zipfile after being wrapped in Path .

gh-97930 : Apply changes from importlib_resources 5.12 , including fix for MultiplexedPath to support directories in multiple namespaces (python/importlib_resources#265).

gh-101997 : Upgrade pip wheel bundled with ensurepip (pip 23.0.1)

gh-99108 : The built-in extension modules for hashlib SHA2 algorithms, used when OpenSSL does not provide them, now live in a single internal _sha2 module instead of separate _sha256 and _sha512 modules.

gh-101892 : Callable iterators no longer raise SystemError when the callable object exhausts the iterator but forgets to either return a sentinel value or raise StopIteration .

gh-87634 : Remove locking behavior from functools.cached_property() .

gh-97786 : Fix potential undefined behaviour in corner cases of floating-point-to-time conversions.

gh-101517 : Fixed bug where bdb looks up the source line with linecache with a lineno=None , which causes it to fail with an unhandled exception.

gh-101773 : Optimize fractions.Fraction for small components. The private argument _normalize of the fractions.Fraction constructor has been removed.

gh-101693 : In sqlite3.Cursor.execute() , DeprecationWarning is now emitted when named placeholders are used together with parameters supplied as a sequence instead of as a dict . Starting from Python 3.14, using named placeholders with parameters supplied as a sequence will raise a ProgrammingError . Patch by Erlend E. Aasland.

gh-101446 : Change repr of collections.OrderedDict to use regular dictionary formatting instead of pairs of keys and values.

gh-101362 : Speed up pathlib.PurePath construction by handling arguments more uniformly. When a pathlib.Path argument is supplied, we use its string representation rather than joining its parts with os.path.join() .

gh-101362 : Speed up pathlib.PurePath construction by calling os.path.join() only when two or more arguments are given.

gh-101362 : Speed up pathlib.Path construction by running the path flavour compatibility check only when pathlib is imported.

gh-85984 : Refactored the implementation of pty.fork() to use os.login_tty() .

A DeprecationWarning is now raised by pty.master_open() and pty.slave_open() . They were undocumented and deprecated long long ago in the docstring in favor of pty.openpty() .

gh-101561 : Add a new decorator typing.override() . See PEP 698 for details. Patch by Steven Troxler.

gh-63301 : Set exit code when tabnanny CLI exits on error.

gh-101360 : Fix anchor matching in pathlib.PureWindowsPath.match() . Path and pattern anchors are now matched with fnmatch , just like other path parts. This allows patterns such as "*:/Users/*" to be matched.

gh-101277 : Remove global state from itertools module ( PEP 687 ). Patches by Erlend E. Aasland.

gh-100809 : Fix handling of drive-relative paths (like ‘C:’ and ‘C:foo’) in pathlib.Path.absolute() . This method now uses the OS API to retrieve the correct current working directory for the drive.

gh-99138 : Apply PEP 687 to zoneinfo . Patch by Erlend E. Aasland.

gh-96764 : asyncio.wait_for() now uses asyncio.timeout() as its underlying implementation. Patch by Kumar Aditya.

gh-88233 : Correctly preserve “extra” fields in zipfile regardless of their ordering relative to a zip64 “extra.”

bpo-23224 : Fix segfaults when creating lzma.LZMADecompressor and bz2.BZ2Decompressor objects without calling __init__() , and fix leakage of locks and internal buffers when calling the __init__() methods of lzma.LZMADecompressor , lzma.LZMACompressor , bz2.BZ2Compressor , and bz2.BZ2Decompressor objects multiple times.

gh-85417 : Update cmath documentation to clarify behaviour on branch cuts.

gh-97725 : Fix asyncio.Task.print_stack() description for file=None . Patch by Oleg Iarygin.

gh-102019 : Fix deadlock on shutdown if test_current_{exception,frames} fails. Patch by Jacob Bower.

gh-85984 : Utilize new “winsize” functions from termios in pty tests.

gh-89792 : test_tools now copies up to 10x less source data to a temporary directory during the freeze test by ignoring git metadata and other artifacts. It also limits its python build parallelism based on os.cpu_count instead of hard coding it as 8 cores.

gh-99942 : On Android, in a static build, python-config in embed mode no longer incorrectly reports a library to link to.

gh-99942 : On Android, python.pc now correctly reports the library to link to, the same as python-config.sh.

gh-100221 : Fix creating install directories in make sharedinstall if they exist outside DESTDIR already.

gh-96821 : Explicitly mark C extension modules that need defined signed integer overflow, and add a configure option --with-strict-overflow . Patch by Matthias Görgens and Shantanu Jain.

gh-102344 : Implement winreg.QueryValue using QueryValueEx and winreg.SetValue using SetValueEx . Patch by Max Bachmann.

gh-101881 : Handle read and write operations on non-blocking pipes properly on Windows.

gh-101881 : Add support for the os.get_blocking() and os.set_blocking() functions on Windows.

gh-101849 : Ensures installer will correctly upgrade existing py.exe launcher installs.

gh-101763 : Updates copy of libffi bundled with Windows installs to 3.4.4.

gh-101759 : Update Windows installer to SQLite 3.40.1.

gh-101614 : Correctly handle extensions built against debug binaries that reference python3_d.dll .

gh-101196 : The functions os.path.isdir , os.path.isfile , os.path.islink and os.path.exists are now 13% to 28% faster on Windows, by making fewer Win32 API calls.

gh-101759 : Update macOS installer to SQLite 3.40.1.

gh-101907 : Removes use of non-standard C++ extension in public header files.

gh-99293 : Document that the Py_TPFLAGS_VALID_VERSION_TAG is an internal feature, should not be used, and will be removed.

gh-101578 : Add PyErr_GetRaisedException() and PyErr_SetRaisedException() for saving and restoring the current exception. These functions return and accept a single exception object, rather than the triple arguments of the now-deprecated PyErr_Fetch() and PyErr_Restore() . This is less error prone and a bit more efficient.

Add PyException_GetArgs() and PyException_SetArgs() as convenience functions for retrieving and modifying the args passed to the exception’s constructor.

gh-91744 : Introduced the Unstable C API tier , marking APi that is allowed to change in minor releases without a deprecation period. See PEP 689 for details.

Python 3.12.0 alpha 5 ¶

Release date: 2023-02-07

gh-99108 : Replace the builtin hashlib implementations of SHA2-224 and SHA2-256 originally from LibTomCrypt with formally verified, side-channel resistant code from the HACL* project. The builtins remain a fallback only used when OpenSSL does not provide them.

gh-92173 : Fix the defs and kwdefs arguments to PyEval_EvalCodeEx() and a reference leak in that function.

gh-59956 : The GILState API is now partially compatible with subinterpreters. Previously, PyThreadState_GET() and PyGILState_GetThisThreadState() would get out of sync, causing inconsistent behavior and crashes.

gh-101400 : Fix wrong lineno in exception message on continue or break which are not in a loop. Patch by Donghee Na.

gh-101372 : Fix is_normalized() to properly handle the UCD 3.2.0 cases. Patch by Donghee Na.

gh-101266 : Fix sys.getsizeof() reporting for int subclasses.

gh-101291 : Refactor the PyLongObject struct into a normal Python object header and a PyLongValue struct.

gh-101046 : Fix a possible memory leak in the parser when raising MemoryError . Patch by Pablo Galindo

gh-101037 : Fix potential memory underallocation issue for instances of int subclasses with value zero.

gh-100762 : Record the (virtual) exception block depth in the oparg of YIELD_VALUE . Use this to avoid the expensive throw() when closing generators (and coroutines) that can be closed trivially.

gh-100982 : Adds a new COMPARE_AND_BRANCH instruction. This is a bit more efficient when performing a comparison immediately followed by a branch, and restores the design intent of PEP 659 that specializations are local to a single instruction.

gh-100942 : Fixed segfault in property.getter/setter/deleter that occurred when a property subclass overrode the __new__ method to return a non-property instance.

gh-100923 : Remove the mask cache entry for the COMPARE_OP instruction and embed the mask into the oparg.

gh-100892 : Fix race while iterating over thread states in clearing threading.local . Patch by Kumar Aditya.

gh-91351 : Fix a case where re-entrant imports could corrupt the import deadlock detection code and cause a KeyError to be raised out of importlib/_bootstrap . In addition to the straightforward cases, this could also happen when garbage collection leads to a warning being emitted – as happens when it collects an open socket or file)

gh-100726 : Optimize construction of range object for medium size integers.

gh-100712 : Added option to build cpython with specialization disabled, by setting ENABLE_SPECIALIZATION=False in opcode , followed by make regen-all .

bpo-32780 : Inter-field padding is now inserted into the PEP3118 format strings obtained from ctypes.Structure objects, reflecting their true representation in memory.

gh-101541 : [Enum] - fix psuedo-flag creation

gh-101570 : Upgrade pip wheel bundled with ensurepip (pip 23.0)

gh-101323 : Fix a bug where errors where not thrown by zlib._ZlibDecompressor if encountered during decompressing.

gh-101317 : Add ssl_shutdown_timeout parameter for asyncio.StreamWriter.start_tls() .

gh-101326 : Fix regression when passing None as second or third argument to FutureIter.throw .

gh-92123 : Adapt the _elementtree extension module to multi-phase init ( PEP 489 ). Patches by Erlend E. Aasland.

gh-100795 : Avoid potential unexpected freeaddrinfo call (double free) in socket when when a libc getaddrinfo() implementation leaves garbage in an output pointer when returning an error. Original patch by Sergey G. Brester.

gh-101143 : Remove unused references to TimerHandle in asyncio.base_events.BaseEventLoop._add_callback .

gh-101144 : Make zipfile.Path.open() and zipfile.Path.read_text() also accept encoding as a positional argument. This was the behavior in Python 3.9 and earlier. 3.10 introduced a regression where supplying it as a positional argument would lead to a TypeError .

gh-94518 : Group-related variables of _posixsubprocess module are renamed to stress that supplementary group affinity is added to a fork, not replace the inherited ones. Patch by Oleg Iarygin.

gh-101015 : Fix typing.get_type_hints() on '*tuple[...]' and *tuple[...] . It must not drop the Unpack part.

gh-101000 : Add os.path.splitroot() , which splits a path into a 3-item tuple (drive, root, tail) . This new function is used by pathlib to improve the performance of path construction by up to a third.

gh-100573 : Fix a Windows asyncio bug with named pipes where a client doing os.stat() on the pipe would cause an error in the server that disabled serving future requests.

gh-39615 : warnings.warn() now has the ability to skip stack frames based on code filename prefix rather than only a numeric stacklevel via the new skip_file_prefixes keyword argument.

gh-100750 : pass encoding kwarg to subprocess in platform

gh-100160 : Emit a deprecation warning in asyncio.DefaultEventLoopPolicy.get_event_loop() if there is no current event loop set and it decides to create one.

gh-96290 : Fix handling of partial and invalid UNC drives in ntpath.splitdrive() , and in ntpath.normpath() on non-Windows systems. Paths such as ‘\server’ and ‘\’ are now considered by splitdrive() to contain only a drive, and consequently are not modified by normpath() on non-Windows systems. The behaviour of normpath() on Windows systems is unaffected, as native OS APIs are used. Patch by Eryk Sun, with contributions by Barney Gale.

gh-99952 : Fix a reference undercounting issue in ctypes.Structure with from_param() results larger than a C pointer.

gh-67790 : Add float-style formatting support for fractions.Fraction instances.

gh-99266 : Preserve more detailed error messages in ctypes .

gh-86682 : Ensure runtime-created collections have the correct module name using the newly added (internal) sys._getframemodulename() .

gh-88597 : uuid now has a command line interface. Try python -m uuid -h .

gh-60580 : ctypes.wintypes.BYTE definition changed from c_byte to c_ubyte to match Windows SDK. Patch by Anatoly Techtonik and Oleg Iarygin.

gh-94518 : _posixsubprocess now initializes all UID and GID variables using a reserved -1 value instead of a separate flag. Patch by Oleg Iarygin.

bpo-38941 : The xml.etree.ElementTree module now emits DeprecationWarning when testing the truth value of an xml.etree.ElementTree.Element . Before, the Python implementation emitted FutureWarning , and the C implementation emitted nothing.

bpo-40077 : Convert elementtree types to heap types. Patch by Erlend E. Aasland.

bpo-29847 : Fix a bug where pathlib.Path accepted and ignored keyword arguments. Patch provided by Yurii Karabas.

gh-77772 : ctypes.CDLL , ctypes.OleDLL , ctypes.WinDLL , and ctypes.PyDLL now accept path-like objects as their name argument. Patch by Robert Hoelzl.

gh-88324 : Reword subprocess to emphasize default behavior of stdin , stdout , and stderr arguments. Remove inaccurate statement about child file handle inheritance.

gh-101334 : test_tarfile has been updated to pass when run as a high UID.

gh-101282 : Update BOLT configuration not to use deprecated usage of --split functions . Patch by Donghee Na.

gh-101522 : Allow overriding Windows dependencies versions and paths using MSBuild properties.

gh-77532 : Minor fixes to allow building with PlatformToolset=ClangCL on Windows.

gh-101152 : In accordance with PEP 699 , the ma_version_tag field in PyDictObject is deprecated for extension modules. Accessing this field will generate a compiler warning at compile time. This field will be removed in Python 3.14.

gh-100340 : Allows -Wno-int-conversion for wasm-sdk 17 and onwards, thus enables building WASI builds once against the latest sdk.

gh-101060 : Conditionally add -fno-reorder-blocks-and-partition in configure. Effectively fixes --enable-bolt when using Clang, as this appears to be a GCC-only flag.

gh-98705 : __bool__ is defined in AIX system header files which breaks the build in AIX, so undefine it.

gh-98636 : Fix a regression in detecting gdbm_compat library for the _gdbm module build.

gh-96305 : _aix_support now uses a simple code to get platform details rather than the now non-existent _bootsubprocess during bootstrap.

gh-101543 : Ensure the install path in the registry is only used when the standard library hasn’t been located in any other way.

gh-101467 : The py.exe launcher now correctly filters when only a single runtime is installed. It also correctly handles prefix matches on tags so that -3.1 does not match 3.11 , but would still match 3.1-32 .

gh-99834 : Updates bundled copy of Tcl/Tk to 8.6.13.0

gh-101135 : Restore ability to launch older 32-bit versions from the py.exe launcher when both 32-bit and 64-bit installs of the same version are available.

gh-82052 : Fixed an issue where writing more than 32K of Unicode output to the console screen in one go can result in mojibake.

gh-100320 : Ensures the PythonPath registry key from an install is used when launching from a different copy of Python that relies on an existing install to provide a copy of its modules and standard library.

gh-100247 : Restores support for the py.exe launcher finding shebang commands in its configuration file using the full command name.

Python 3.12.0 alpha 4 ¶

Release date: 2023-01-10

gh-100776 : Fix misleading default value in input() ’s __text_signature__ .

gh-99005 : Remove UNARY_POSITIVE , ASYNC_GEN_WRAP and LIST_TO_TUPLE , replacing them with intrinsics.

gh-99005 : Add new CALL_INTRINSIC_1 instruction. Remove IMPORT_STAR , PRINT_EXPR and STOPITERATION_ERROR , replacing them with the CALL_INTRINSIC_1 instruction.

gh-100288 : Remove the LOAD_ATTR_METHOD_WITH_DICT specialized instruction. Stats show it is not useful.

gh-100720 : Added _PyFrame_NumSlotsForCodeObject , which returns the number of slots needed in a frame for a given code object.

gh-100719 : Removed the co_nplaincellvars field from the code object, as it is redundant.

gh-100637 : Fix int.__sizeof__() calculation to include the 1-element ob_digit array for 0 and False .

gh-100649 : Update the native_thread_id field of PyThreadState after fork.

gh-100126 : Fix an issue where “incomplete” frames could be briefly visible to C code while other frames are being torn down, possibly resulting in corruption or hard crashes of the interpreter while running finalizers.

gh-87447 : Fix SyntaxError on comprehension rebind checking with names that are not actually redefined.

Now reassigning b in [(b := 1) for a, b.prop in some_iter] is allowed. Reassigning a is still disallowed as per PEP 572 .

gh-100268 : Add int.is_integer() to improve duck type compatibility between int and float .

gh-100425 : Improve the accuracy of sum() with compensated summation.

gh-100374 : Fix incorrect result and delay in socket.getfqdn() . Patch by Dominic Socular.

gh-100357 : Convert vars , dir , next , getattr , and iter to argument clinic.

gh-100117 : Improve the output of codeobject.co_lines() by emitting only one entry for each line range.

gh-90043 : Handle NaNs when specializing COMPARE_OP for float values.

gh-100222 : Redefine the _Py_CODEUNIT typedef as a union to describe its layout to the C compiler, avoiding type punning and improving clarity.

gh-99955 : Internal compiler functions (in compile.c) now consistently return -1 on error and 0 on success.

gh-100188 : The BINARY_SUBSCR_LIST_INT and BINARY_SUBSCR_TUPLE_INT instructions are no longer used for negative integers because those instructions always miss when encountering negative integers.

gh-99110 : Initialize frame->previous in frameobject.c to fix a segmentation fault when accessing frames created by PyFrame_New() .

gh-94155 : Improved the hashing algorithm for code objects, mitigating some hash collisions.

gh-99540 : None now hashes to a constant value. This is not a requirements change.

gh-100143 : When built with --enable-pystats , stats collection is now off by default. To enable it early at startup, pass the -Xpystats flag. Stats are now always dumped, even if switched off.

gh-100146 : Improve BUILD_LIST opcode so that it works similarly to the BUILD_TUPLE opcode, by stealing references from the stack rather than repeatedly using stack operations to set list elements. Implementation details are in a new private API _PyList_FromArraySteal() .

gh-100110 : Specialize FOR_ITER for tuples.

gh-100050 : Honor existing errors obtained when searching for mismatching parentheses in the tokenizer. Patch by Pablo Galindo

gh-92216 : Improve the performance of hasattr() for type objects with a missing attribute.

gh-99582 : Freeze zipimport module into _bootstrap_python .

gh-99554 : Pack debugging location tables more efficiently during bytecode compilation.

gh-98522 : Add an internal version number to code objects, to give better versioning of inner functions and comprehensions, and thus better specialization of those functions. This change is invisible to both Python and C extensions.

gh-94603 : Improve performance of list.pop for small lists.

gh-89051 : Add ssl.OP_LEGACY_SERVER_CONNECT

bpo-32782 : ctypes arrays of length 0 now report a correct itemsize when a memoryview is constructed from them, rather than always giving a value of 0.

gh-100833 : Speed up math.fsum() by removing defensive volatile qualifiers.

gh-100805 : Modify random.choice() implementation to once again work with NumPy arrays.

gh-100813 : Add socket.IP_PKTINFO constant.

gh-100792 : Make email.message.Message.__contains__() twice as fast.

gh-91851 : Microoptimizations for fractions.Fraction.__round__() , fractions.Fraction.__ceil__() and fractions.Fraction.__floor__() .

gh-90104 : Avoid RecursionError on repr if a dataclass field definition has a cyclic reference.

gh-100689 : Fix crash in pyexpat by statically allocating PyExpat_CAPI capsule.

gh-100740 : Fix unittest.mock.Mock not respecting the spec for attribute names prefixed with assert .

gh-91219 : Change SimpleHTTPRequestHandler to support subclassing to provide a different set of index file names instead of using __init__ parameters.

gh-100690 : Mock objects which are not unsafe will now raise an AttributeError when accessing an attribute that matches the name of an assertion but without the prefix assert_ , e.g. accessing called_once instead of assert_called_once . This is in addition to this already happening for accessing attributes with prefixes assert , assret , asert , aseert , and assrt .

gh-89727 : Simplify and optimize os.walk() by using isinstance() checks to check the top of the stack.

gh-100485 : Add math.sumprod() to compute the sum of products.

gh-86508 : Fix asyncio.open_connection() to skip binding to local addresses of different family. Patch by Kumar Aditya.

gh-97930 : importlib.resources.files now accepts a module as an anchor instead of only accepting packages. If a module is passed, resources are resolved adjacent to that module (in the same package or at the package root). The parameter was renamed from package to anchor with a compatibility shim for those passing by keyword. Additionally, the new anchor parameter is now optional and will default to the caller’s module.

gh-100585 : Fixed a bug where importlib.resources.as_file was leaving file pointers open

gh-100562 : Improve performance of pathlib.Path.absolute() by nearly 2x. This comes at the cost of a performance regression in pathlib.Path.cwd() , which is generally used less frequently in user code.

gh-100519 : Small simplification of http.cookiejar.eff_request_host() that improves readability and better matches the RFC wording.

gh-100287 : Fix the interaction of unittest.mock.seal() with unittest.mock.AsyncMock .

gh-100488 : Add Fraction.is_integer() to check whether a fractions.Fraction is an integer. This improves duck type compatibility with float and int .

gh-100474 : http.server now checks that an index page is actually a regular file before trying to serve it. This avoids issues with directories named index.html .

gh-100363 : Speed up asyncio.get_running_loop() by removing redundant getpid checks. Patch by Kumar Aditya.

gh-78878 : Fix crash when creating an instance of _ctypes.CField .

gh-100348 : Fix ref cycle in asyncio._SelectorSocketTransport by removing _read_ready_cb in close .

gh-100344 : Provide C implementation for asyncio.current_task() for a 4x-6x speedup.

gh-100272 : Fix JSON serialization of OrderedDict. It now preserves the order of keys.

gh-83076 : Instantiation of Mock() and AsyncMock() is now 3.8x faster.

gh-100234 : Set a default value of 1.0 for the lambd parameter in random.expovariate().

gh-100228 : A DeprecationWarning may be raised when os.fork() or os.forkpty() is called from multi-threaded processes. Forking with threads is unsafe and can cause deadlocks, crashes and subtle problems. Lack of a warning does not indicate that the fork call was actually safe, as Python may not be aware of all threads.

gh-100039 : Improve signatures for enums and flags.

gh-100133 : Fix regression in asyncio where a subprocess would sometimes lose data received from pipe.

bpo-44592 : Fixes inconsistent handling of case sensitivity of extrasaction arg in csv.DictWriter .

gh-100098 : Fix tuple subclasses being cast to tuple when used as enum values.

gh-85432 : Rename the fmt parameter of the pure-Python implementation of datetime.time.strftime() to format . Rename the t parameter of datetime.datetime.fromtimestamp() to timestamp . These changes mean the parameter names in the pure-Python implementation now match the parameter names in the C implementation. Patch by Alex Waygood.

gh-98778 : Update HTTPError to be initialized properly, even if the fp is None . Patch by Donghee Na.

gh-99925 : Unify error messages in JSON serialization between json.dumps(float('nan'), allow_nan=False) and json.dumps(float('nan'), allow_nan=False, indent=<SOMETHING>) . Now both include the representation of the value that could not be serialized.

gh-89727 : Fix issue with os.walk() where a RecursionError would occur on deep directory structures by adjusting the implementation of os.walk() to be iterative instead of recursive.

gh-94943 : Add Dataclass support to the Enum __repr__() . When inheriting from a dataclass , only show the field names in the value section of the member repr() , and not the dataclass’ class name.

gh-83035 : Fix inspect.getsource() handling of decorator calls with nested parentheses.

gh-99576 : Fix .save() method for LWPCookieJar and MozillaCookieJar : saved file was not truncated on repeated save.

gh-94912 : Add inspect.markcoroutinefunction() decorator which manually marks a function as a coroutine for the benefit of iscoroutinefunction() .

gh-99509 : Add PEP 585 support for multiprocessing.queues.Queue .

gh-99482 : Remove Jython partial compatibility code from several stdlib modules.

gh-99433 : Fix doctest failure on types.MethodWrapperType in modules.

gh-85267 : Several improvements to inspect.signature() ’s handling of __text_signature . - Fixes a case where inspect.signature() dropped parameters - Fixes a case where inspect.signature() raised tokenize.TokenError - Allows inspect.signature() to understand defaults involving binary operations of constants - inspect.signature() is documented as only raising TypeError or ValueError , but sometimes raised RuntimeError . These cases now raise ValueError - Removed a dead code path

gh-91166 : asyncio is optimized to avoid excessive copying when writing to socket and use sendmsg() if the platform supports it. Patch by Kumar Aditya.

gh-98030 : Add missing TCP socket options from Linux: TCP_MD5SIG , TCP_THIN_LINEAR_TIMEOUTS , TCP_THIN_DUPACK , TCP_REPAIR , TCP_REPAIR_QUEUE , TCP_QUEUE_SEQ , TCP_REPAIR_OPTIONS , TCP_TIMESTAMP , TCP_CC_INFO , TCP_SAVE_SYN , TCP_SAVED_SYN , TCP_REPAIR_WINDOW , TCP_FASTOPEN_CONNECT , TCP_ULP , TCP_MD5SIG_EXT , TCP_FASTOPEN_KEY , TCP_FASTOPEN_NO_COOKIE , TCP_ZEROCOPY_RECEIVE , TCP_INQ , TCP_TX_DELAY .

gh-88500 : Reduced the memory usage of urllib.parse.unquote() and urllib.parse.unquote_to_bytes() on large values.

gh-96127 : inspect.signature was raising TypeError on call with mock objects. Now it correctly returns (*args, **kwargs) as inferred signature.

gh-95882 : Fix a 3.11 regression in asynccontextmanager() , which caused it to propagate exceptions with incorrect tracebacks and fix a 3.11 regression in contextmanager() , which caused it to propagate exceptions with incorrect tracebacks for StopIteration .

gh-78707 : Deprecate passing more than one positional argument to pathlib.PurePath.relative_to() and is_relative_to() .

gh-92122 : Fix reStructuredText syntax errors in docstrings in the enum module.

gh-91851 : Optimize the Fraction arithmetics for small components.

bpo-24132 : Make pathlib.PurePath and Path subclassable (private to start). Previously, attempting to instantiate a subclass resulted in an AttributeError being raised. Patch by Barney Gale.

bpo-40447 : Accept os.PathLike (such as pathlib.Path ) in the stripdir arguments of compileall.compile_file() and compileall.compile_dir() .

bpo-36880 : Fix a reference counting issue when a ctypes callback with return type py_object returns None , which could cause crashes.

gh-100616 : Document existing attr parameter to curses.window.vline() function in curses .

gh-100472 : Remove claim in documentation that the stripdir , prependdir and limit_sl_dest parameters of compileall.compile_dir() and compileall.compile_file() could be bytes .

bpo-25377 : Clarify use of octal format of mode argument in help(os.chmod) as well as help(os.fchmod)

gh-100454 : Start running SSL tests with OpenSSL 3.1.0-beta1.

gh-100086 : The Python test runner (libregrtest) now logs Python build information like “debug” vs “release” build, or LTO and PGO optimizations. Patch by Victor Stinner.

gh-93018 : Make two tests forgiving towards host system libexpat with backported security fixes applied.

gh-100540 : Removed the --with-system-ffi configure option; libffi must now always be supplied by the system on all non-Windows platforms. The option has had no effect on non-Darwin platforms for several releases, and in 3.11 only had the non-obvious effect of invoking pkg-config to find libffi and never setting -DUSING_APPLE_OS_LIBFFI . Now on Darwin platforms configure will first check for the OS libffi and then fall back to the same processing as other platforms if it is not found.

gh-88267 : Avoid exporting Python symbols in linked Windows applications when the core is built as static.

bpo-41916 : Allow override of ac_cv_cxx_thread so that cross compiled python can set -pthread for CXX.

gh-100180 : Update Windows installer to OpenSSL 1.1.1s

gh-99191 : Use _MSVC_LANG >= 202002L instead of less-precise _MSC_VER >=1929 to more accurately test for C++20 support in PC/_wmimodule.cpp .

gh-79218 : Define MS_WIN64 for Mingw-w64 64bit, fix cython compilation failure.

gh-99941 : Ensure that asyncio.Protocol.data_received() receives an immutable bytes object (as documented), instead of bytearray .

bpo-43984 : winreg.SetValueEx() now leaves the target value untouched in the case of conversion errors. Previously, -1 would be written in case of such errors.

bpo-34816 : hasattr(ctypes.windll, 'nonexistant') now returns False instead of raising OSError .

gh-100180 : Update macOS installer to OpenSSL 1.1.1s

gh-100540 : Removed obsolete dlfcn.h shim from the _ctypes extension module, which has not been necessary since Mac OS X 10.2.

bpo-45256 : Fix a bug that caused an AttributeError to be raised in python-gdb.py when py-locals is used without a frame.

gh-100342 : Add missing NULL check for possible allocation failure in *args parsing in Argument Clinic.

gh-99947 : Raising SystemError on import will now have its cause be set to the original unexpected exception.

gh-99240 : In argument parsing, after deallocating newly allocated memory, reset its pointer to NULL.

gh-98724 : The Py_CLEAR , Py_SETREF and Py_XSETREF macros now only evaluate their arguments once. If an argument has side effects, these side effects are no longer duplicated. Patch by Victor Stinner.

Python 3.12.0 alpha 3 ¶

Release date: 2022-12-06

gh-100001 : python -m http.server no longer allows terminal control characters sent within a garbage request to be printed to the stderr server log.

This is done by changing the http.server BaseHTTPRequestHandler .log_message method to replace control characters with a \x HH hex escape before printing.

gh-87604 : Avoid publishing list of active per-interpreter audit hooks via the gc module

gh-99891 : Fix a bug in the tokenizer that could cause infinite recursion when showing syntax warnings that happen in the first line of the source. Patch by Pablo Galindo

gh-91054 : Add PyCode_AddWatcher() and PyCode_ClearWatcher() APIs to register callbacks to receive notification on creation and destruction of code objects.

gh-99729 : Fix an issue that could cause frames to be visible to Python code as they are being torn down, possibly leading to memory corruption or hard crashes of the interpreter.

gh-99708 : Fix bug where compiler crashes on an if expression with an empty body block.

gh-99578 : Fix a reference bug in _imp.create_builtin() after the creation of the first sub-interpreter for modules builtins and sys . Patch by Victor Stinner.

gh-99581 : Fixed a bug that was causing a buffer overflow if the tokenizer copies a line missing the newline character from a file that is as long as the available tokenizer buffer. Patch by Pablo galindo

gh-99553 : Fix bug where an ExceptionGroup subclass can wrap a BaseException .

gh-99547 : Add a function to os.path to check if a path is a junction: isjunction. Add similar functionality to pathlib.Path as is_junction.

gh-99370 : Fix zip path for venv created from a non-installed python on POSIX platforms.

gh-99377 : Add audit events for thread creation and clear operations.

gh-98686 : Remove the BINARY_OP_GENERIC and COMPARE_OP_GENERIC “specializations”.

gh-99298 : Remove the remaining error paths for attribute specializations, and refuse to specialize attribute accesses on types that haven’t had PyType_Ready() called on them yet.

gh-99127 : Allow some features of syslog to the main interpreter only. Patch by Donghee Na.

gh-91053 : Optimizing interpreters and JIT compilers may need to invalidate internal metadata when functions are modified. This change adds the ability to provide a callback that will be invoked each time a function is created, modified, or destroyed.

gh-90994 : Improve error messages when there’s a syntax error with call arguments. The following three cases are covered: - No value is assigned to a named argument, eg foo(a=) . - A value is assigned to a star argument, eg foo(*args=[0]) . - A value is assigned to a double-star keyword argument, eg foo(**kwarg={'a': 0}) .

bpo-45026 : Optimize the range object iterator. It is now smaller, faster iteration of ranges containing large numbers. Smaller pickles, faster unpickling.

bpo-31718 : Raise ValueError instead of SystemError when methods of uninitialized io.IncrementalNewlineDecoder objects are called. Patch by Oren Milman.

bpo-38031 : Fix a possible assertion failure in io.FileIO when the opener returns an invalid file descriptor.

gh-100001 : Also escape s in the http.server BaseHTTPRequestHandler.log_message so that it is technically possible to parse the line and reconstruct what the original data was. Without this a xHH is ambiguous as to if it is a hex replacement we put in or the characters r”x” came through in the original request line.

gh-99957 : Add frozen_default parameter to typing.dataclass_transform() .

gh-79033 : Fix asyncio.Server.wait_closed() to actually do what the docs promise – wait for all existing connections to complete, after closing the server.

gh-51524 : Fix bug when calling trace.CoverageResults with valid infile.

gh-99645 : Fix a bug in handling class cleanups in unittest.TestCase . Now addClassCleanup() uses separate lists for different TestCase subclasses, and doClassCleanups() only cleans up the particular class.

gh-99508 : Fix TypeError in Lib/importlib/_bootstrap_external.py while calling _imp.source_hash() .

gh-66285 : Fix asyncio to not share event loop and signal wakeupfd in forked processes. Patch by Kumar Aditya.

gh-97001 : Release the GIL when calling termios APIs to avoid blocking threads.

gh-92647 : Use final status of an enum to determine lookup or creation branch of functional API.

gh-99388 : Add loop_factory parameter to asyncio.run() to allow specifying a custom event loop factory. Patch by Kumar Aditya.

gh-99341 : Fix ast.increment_lineno() to also cover ast.TypeIgnore when changing line numbers.

gh-99382 : Check the number of arguments in substitution in user generics containing a TypeVarTuple and one or more TypeVar .

gh-99379 : Fix substitution of ParamSpec followed by TypeVarTuple in generic aliases.

gh-99344 : Fix substitution of TypeVarTuple and ParamSpec together in user generics.

gh-99284 : Remove _use_broken_old_ctypes_structure_semantics_ old untested and undocumented hack from ctypes .

gh-99201 : Fix IndexError when initializing the config variables on Windows if HAVE_DYNAMIC_LOADING is not set.

gh-99240 : Fix double-free bug in Argument Clinic str_converter by extracting memory clean up to a new post_parsing section.

gh-64490 : Fix refcount error when arguments are packed to tuple in Argument Clinic.

gh-99029 : pathlib.PurePath.relative_to() now treats naked Windows drive paths as relative. This brings its behaviour in line with other parts of pathlib.

gh-98253 : The implementation of the typing module is now more resilient to reference leaks in binary extension modules.

Previously, a reference leak in a typed C API-based extension module could leak internals of the typing module, which could in turn introduce leaks in essentially any other package with typed function signatures. Although the typing package is not the original source of the problem, such non-local dependences exacerbate debugging of large-scale projects, and the implementation was therefore changed to reduce harm by providing better isolation.

gh-98458 : Fix infinite loop in unittest when a self-referencing chained exception is raised

gh-93453 : asyncio.get_event_loop() and many other asyncio functions like asyncio.ensure_future() , asyncio.shield() or asyncio.gather() , and also the get_event_loop() method of asyncio.BaseDefaultEventLoopPolicy now raise a RuntimeError if called when there is no running event loop and the current event loop was not set. Previously they implicitly created and set a new current event loop. DeprecationWarning is no longer emitted if there is no running event loop but the current event loop was set.

gh-97966 : On uname_result , restored expectation that _fields and _asdict would include all six properties including processor .

gh-98248 : Provide informative error messages in struct.pack() when its integral arguments are not in range.

gh-98108 : zipfile.Path is now pickleable if its initialization parameters were pickleable (e.g. for file system paths).

gh-98098 : Created packages from zipfile and test_zipfile modules, separating zipfile.Path functionality.

gh-82836 : Fix is_private properties in the ipaddress module. Previously non-private networks (0.0.0.0/0) would return True from this method; now they correctly return False .

gh-96828 : Add an OP_ENABLE_KTLS option for enabling the use of the kernel TLS (kTLS). Patch by Illia Volochii.

gh-88863 : To avoid apparent memory leaks when asyncio.open_connection() raises, break reference cycles generated by local exception and future instances (which has exception instance as its member var). Patch by Dong Uk, Kang.

gh-91078 : TarFile.next() now returns None when called on an empty tarfile.

bpo-47220 : Document the optional callback parameter of WeakMethod . Patch by Géry Ogam.

bpo-44817 : Ignore WinError 53 (ERROR_BAD_NETPATH), 65 (ERROR_NETWORK_ACCESS_DENIED) and 161 (ERROR_BAD_PATHNAME) when using ntpath.realpath().

bpo-41260 : Rename the fmt parameter of the pure Python implementation of datetime.date.strftime() to format .

bpo-15999 : All built-in functions now accept arguments of any type instead of just bool and int for boolean parameters.

gh-99931 : Use sphinxext-opengraph to generate OpenGraph metadata .

gh-89682 : Reworded docstring of the default __contains__ to clarify that it returns a bool .

gh-88330 : Improved the description of what a resource is in importlib.resources docs.

gh-92892 : Document that calling variadic functions with ctypes requires special care on macOS/arm64 (and possibly other platforms).

bpo-41825 : Restructured the documentation for the os.wait* family of functions, and improved the docs for os.waitid() with more explanation of the possible argument constants.

gh-99892 : Skip test_normalization() of test_unicodedata if it fails to download NormalizationTest.txt file from pythontest.net. Patch by Victor Stinner.

gh-99934 : Correct test_marsh on (32 bit) x86: test_deterministic sets was failing.

gh-99741 : We’ve implemented multi-phase init (PEP 489/630/687) for the internal (for testing) _xxsubinterpreters module.

gh-99659 : Optional big memory tests in test_sqlite3 now catch the correct sqlite.DataError exception type in case of too large strings and/or blobs passed.

gh-99593 : Cover the Unicode C API with tests.

gh-96002 : Add functional test for Argument Clinic.

gh-99086 : Fix -Wimplicit-int , -Wstrict-prototypes , and -Wimplicit-function-declaration compiler warnings in configure checks.

gh-99337 : Fix a compilation issue with GCC 12 on macOS.

gh-99289 : Add a COMPILEALL_OPTS variable in Makefile to override compileall options (default: -j0 ) in make install . Also merged the compileall commands into a single command building .pyc files for the all optimization levels (0, 1, 2) at once. Patch by Victor Stinner.

gh-98872 : Fix a possible fd leak in Programs/_freeze_module.c introduced in Python 3.11.

gh-88226 : Always define TARGET_* labels in Python/ceval.c , even if USE_COMPUTED_GOTOS is disabled. This allows breakpoints to be set at those labels in (for instance) gdb .

gh-99345 : Use faster initialization functions to detect install location for Windows Store package

gh-98629 : Fix initialization of sys.version and sys._git on Windows

gh-99442 : Fix handling in Python Launcher for Windows when argv[0] does not include a file extension.

bpo-40882 : Fix a memory leak in multiprocessing.shared_memory.SharedMemory on Windows.

gh-87235 : On macOS python3 /dev/fd/9 9</path/to/script.py failed for any script longer than a couple of bytes.

gh-98940 : Fix Mac/Extras.install.py file filter bug.

gh-64490 : Argument Clinic varargs bugfixes

Fix out-of-bounds error in _PyArg_UnpackKeywordsWithVararg() .

Fix incorrect check which allowed more than one varargs in clinic.py.

Fix miscalculation of noptargs in generated code.

Do not generate noptargs when there is a vararg argument and no optional argument.

gh-98680 : PyBUF_* constants were marked as part of Limited API of Python 3.11+. These were available in 3.11.0 with Py_LIMITED_API defined for 3.11, and are necessary to use the buffer API.

gh-99612 : Fix PyUnicode_DecodeUTF8Stateful() for ASCII-only data: *consumed was not set.

gh-47146 : The structmember.h header is deprecated. Its non-deprecated contents are now available just by including Python.h , with a Py_ prefix added if it was missing. (Deprecated contents are T_OBJECT , T_NONE , and no-op flags.) Patch by Petr Viktorin, based on earlier work by Alexander Belopolsky and Matthias Braun.

Python 3.12.0 alpha 2 ¶

Release date: 2022-11-14

gh-98433 : The IDNA codec decoder used on DNS hostnames by socket or asyncio related name resolution functions no longer involves a quadratic algorithm. This prevents a potential CPU denial of service if an out-of-spec excessive length hostname involving bidirectional characters were decoded. Some protocols such as urllib http 3 xx redirects potentially allow for an attacker to supply such a name.

Individual labels within an IDNA encoded DNS name will now raise an error early during IDNA decoding if they are longer than 1024 unicode characters given that each decoded DNS label must be 63 or fewer characters and the entire decoded DNS name is limited to 255. Only an application presenting a hostname or label consisting primarily of RFC 3454 section 3.1 “Nothing” characters to be removed would run into of this new limit. See also RFC 5894 section 6 and RFC 3491 .

gh-98739 : Update bundled libexpat to 2.5.0

gh-81057 : The docs clearly say that PyImport_Inittab , PyImport_AppendInittab() , and PyImport_ExtendInittab() should not be used after Py_Initialize() has been called. We now enforce this for the two functions. Additionally, the runtime now uses an internal copy of PyImport_Inittab , to guard against modification.

gh-99298 : Fix an issue that could potentially cause incorrect error handling for some bytecode instructions.

gh-99254 : The compiler now removes all unused constants from code objects (except the first one, which may be a docstring).

gh-99205 : Fix an issue that prevented PyThreadState and PyInterpreterState memory from being freed properly.

gh-81057 : The 18 global C variables holding the state of the allocators have been moved to _PyRuntimeState . This is a strictly internal change with no change in behavior.

gh-99181 : Fix failure in except* with unhashable exceptions.

gh-99204 : Fix calculation of sys._base_executable when inside a POSIX virtual environment using copies of the python binary when the base installation does not provide the executable name used by the venv. Calculation will fall back to alternative names (“python<MAJOR>”, “python<MAJOR>.<MINOR>”).

gh-96055 : Update faulthandler to emit an error message with the proper unexpected signal number. Patch by Donghee Na.

gh-99153 : Fix location of SyntaxError for a try block with both except and except* .

gh-98686 : Merge the adaptive opcode logic into each instruction’s unquickened variant, and merge the logic in EXTENDED_ARG_QUICK into EXTENDED_ARG . With these changes, the quickening that happens at code object creation is now only responsible for initializing warmup counters and inserting superinstructions.

gh-99103 : Fix the error reporting positions of specialized traceback anchors when the source line contains Unicode characters.

gh-99139 : Improve the error suggestion for NameError exceptions for instances. Now if a NameError is raised in a method and the instance has an attribute that’s exactly equal to the name in the exception, the suggestion will include self.<NAME> instead of the closest match in the method scope. Patch by Pablo Galindo

gh-98401 : Octal escapes with value larger than 0o377 (ex: "\477" ), deprecated in Python 3.11, now produce a SyntaxWarning , instead of DeprecationWarning . In a future Python version they will be eventually a SyntaxError . Patch by Victor Stinner.

gh-98401 : A backslash-character pair that is not a valid escape sequence now generates a SyntaxWarning , instead of DeprecationWarning . For example, re.compile("\d+\.\d+") now emits a SyntaxWarning ( "\d" is an invalid escape sequence), use raw strings for regular expression: re.compile(r"\d+\.\d+") . In a future Python version, SyntaxError will eventually be raised, instead of SyntaxWarning . Patch by Victor Stinner.

gh-96793 : Handle StopIteration and StopAsyncIteration raised in generator or coroutines in the bytecode, rather than in wrapping C code.

gh-98931 : Improve the SyntaxError error message when the user types import x from y instead of from y import x . Patch by Pablo Galindo

gh-98852 : Fix subscription of type aliases containing bare generic types or types like TypeVar : for example tuple[A, T][int] and tuple[TypeVar, T][int] , where A is a generic type, and T is a type variable.

gh-98925 : Lower the recursion depth for marshal on WASI to support (in-development) wasmtime 2.0.

gh-98783 : Fix multiple crashes in debug mode when str subclasses are used instead of str itself.

gh-98811 : Use complete source locations to simplify detection of __future__ imports which are not at the beginning of the file. Also corrects the offset in the exception raised in one case, which was off by one and impeded highlighting.

gh-96793 : Add specialization of FOR_ITER for generators. Saves multiple layers of dispatch and checking to get from the FOR_ITER instruction in the caller to the RESUME in the generator.

gh-98762 : Fix source locations of match sub-patterns.

gh-98586 : Added the methods PyObject_Vectorcall() and PyObject_VectorcallMethod() to the Limited API along with the auxiliary macro constant PY_VECTORCALL_ARGUMENTS_OFFSET .

The availability of these functions enables more efficient PEP 590 vector calls from binary extension modules that avoid argument boxing/unboxing overheads.

gh-99257 : Fix an issue where member descriptors (such as those for __slots__ ) could behave incorrectly or crash instead of raising a TypeError when accessed via an instance of an invalid type.

gh-93143 : Rather than changing co_code , the interpreter will now display a RuntimeWarning and assign None to any fast locals that are left unbound after jumps or del statements executed while tracing.

gh-96421 : When calling into Python code from C code, through PyEval_EvalFrameEx() or a related C-API function, a shim frame in inserted into the call stack. This occurs in the _PyEval_EvalFrameDefault() function. The extra frame should be invisible to all Python and most C extensions, but out-of-process profilers and debuggers need to be aware of it. These shim frames can be detected by checking frame->owner == FRAME_OWNED_BY_CSTACK .

Extensions implementing their own interpreters using PEP 523 need to be aware of this shim frame and the changes to the semantics of RETURN_VALUE , YIELD_VALUE , and RETURN_GENERATOR , which now clear the frame.

gh-98415 : Fix detection of MAC addresses for uuid on certain OSs. Patch by Chaim Sanders

gh-98686 : Quicken all code objects, and specialize adaptive bytecode instructions more aggressively.

gh-92119 : Print exception class name instead of its string representation when raising errors from ctypes calls.

gh-91058 : ImportError raised from failed from <module> import <name> now include suggestions for the value of <name> based on the available names in <module> . Patch by Pablo Galindo

gh-96793 : The FOR_ITER now leaves the iterator on the stack on termination of the loop. This is to assist specialization of loops for generators.

gh-90716 : Add _pylong.py module. It includes asymptotically faster algorithms that can be used for operations on integers with many digits. It is used by longobject.c to speed up some operations.

gh-95389 : Expose ETH_P_ALL and some of the ETHERTYPE_* constants in socket . Patch by Noam Cohen.

gh-93696 : Allow pdb to locate source for frozen modules in the standard library.

gh-99418 : Fix bug in urllib.parse.urlparse() that causes URL schemes that begin with a digit, a plus sign, or a minus sign to be parsed incorrectly.

gh-94597 : Deprecate asyncio.AbstractChildWatcher to be removed in Python 3.14. Patch by Kumar Aditya.

gh-99305 : Improve performance of secrets.token_hex() .

gh-74044 : Fixed bug where inspect.signature() reported incorrect arguments for decorated methods.

gh-99275 : Fix SystemError in ctypes when exception was not set during __initsubclass__ .

gh-99277 : Remove older version of _SSLProtocolTransport.get_write_buffer_limits in asyncio.sslproto

gh-99248 : fix negative numbers failing in verify()

gh-99155 : Fix statistics.NormalDist pickle with 0 and 1 protocols.

gh-93464 : enum.auto() is now correctly activated when combined with other assignment values. E.g. ONE = auto(), 'some text' will now evaluate as (1, 'some text') .

gh-99134 : Update the bundled copy of pip to version 22.3.1.

gh-92584 : Remove the distutils package. It was deprecated in Python 3.10 by PEP 632 “Deprecate distutils module”. For projects still using distutils and cannot be updated to something else, the setuptools project can be installed: it still provides distutils . Patch by Victor Stinner.

gh-98999 : Now _pyio is consistent with _io in raising ValueError when executing methods over closed buffers.

gh-83004 : Clean up refleak on failed module initialisation in _zoneinfo

gh-83004 : Clean up refleaks on failed module initialisation in _pickle

gh-83004 : Clean up refleak on failed module initialisation in _io .

gh-98897 : Fix memory leak in math.dist() when both points don’t have the same dimension. Patch by Kumar Aditya.

gh-98878 : Use the frame bound builtins when offering a name suggestion in traceback to prevent crashing when __builtins__ is not a dict.

gh-98139 : In importlib._bootstrap , enhance namespace package repr to <module 'x' (namespace) from ['path']> .

gh-90352 : Fix _SelectorDatagramTransport to inherit from DatagramTransport in asyncio . Patch by Kumar Aditya.

gh-98793 : Fix argument typechecks in _overlapped.WSAConnect() and _overlapped.Overlapped.WSASendTo() functions.

gh-98744 : Prevent crashing in traceback when retrieving the byte-offset for some source files that contain certain unicode characters.

gh-98740 : Fix internal error in the re module which in very rare circumstances prevented compilation of a regular expression containing a conditional expression without the “else” branch.

gh-98703 : Fix asyncio.StreamWriter.drain() to call protocol.connection_lost callback only once on Windows.

gh-98624 : Add a mutex to unittest.mock.NonCallableMock to protect concurrent access to mock attributes.

gh-98658 : The array.array class now supports subscripting, making it a generic type .

gh-98284 : Improved TypeError message for undefined abstract methods of a abc.ABC instance. The names of the missing methods are surrounded by single-quotes to highlight them.

gh-96151 : Allow BUILTINS to be a valid field name for frozen dataclasses.

gh-98086 : Make sure patch.dict() can be applied on async functions.

gh-72719 : Remove modules asyncore and asynchat , which were deprecated by PEP 594 .

gh-96192 : Fix handling of bytes path-like objects in os.ismount() .

gh-94172 : ftplib : Remove the FTP_TLS.ssl_version class attribute: use the context parameter instead. Patch by Victor Stinner

gh-94172 : Remove the keyfile and certfile parameters from the ftplib , imaplib , poplib and smtplib modules, and the key_file , cert_file and check_hostname parameters from the http.client module, all deprecated since Python 3.6. Use the context parameter ( ssl_context in imaplib ) instead. Patch by Victor Stinner.

gh-83638 : Add the autocommit attribute to sqlite3.Connection and the autocommit parameter to sqlite3.connect() to control PEP 249 -compliant transaction handling . Patch by Erlend E. Aasland.

gh-92452 : Fixed a race condition that could cause sysconfig.get_config_var() to incorrectly return None in multi-threaded programs.

gh-91803 : Fix an error when using a method of objects mocked with unittest.mock.create_autospec() after it was sealed with unittest.mock.seal() function.

bpo-38523 : shutil.copytree() now applies the ignore_dangling_symlinks argument recursively.

bpo-40358 : Add walk_up argument in pathlib.PurePath.relative_to() .

bpo-36267 : Fix IndexError in argparse.ArgumentParser when a store_true action is given an explicit argument.

gh-98832 : Changes wording of docstring for pathlib.Path.iterdir() .

gh-97966 : Update uname docs to clarify the special nature of the platform attribute and to indicate when it became late-bound.

gh-98903 : The Python test suite now fails with exit code 4 if no tests ran. It should help detecting typos in test names and test methods.

gh-98713 : Fix a bug in the typing tests where a test relying on CPython-specific implementation details was not decorated with @cpython_only and was not skipped on other implementations.

gh-87390 : Add tests for star-unpacking with PEP 646, and some other miscellaneous PEP 646 tests.

gh-96853 : Added explicit coverage of Py_Initialize (and hence Py_InitializeEx ) back to the embedding tests (all other embedding tests migrated to Py_InitializeFromConfig in Python 3.11)

bpo-34272 : Some C API tests were moved into the new Lib/test/test_capi/ directory.

gh-99086 : Fix -Wimplicit-int compiler warning in configure check for PTHREAD_SCOPE_SYSTEM .

gh-99016 : Fix build with PYTHON_FOR_REGEN=python3.8 .

gh-97731 : Specify the full path to the source location for make docclean (needed for cross-builds).

gh-98949 : Drop unused build dependency on readelf .

gh-98989 : Use python3.11 , if available, for regeneration and freezing.

gh-98831 : Add new tooling, in Tools/cases_generator , to generate the interpreter switch statement from a list of opcode definitions. This only affects adding, modifying or removing instruction definitions. The instruction definitions now live in Python/bytecodes.c , in the form of a custom DSL (under development) . The tooling reads this file and writes Python/generated_cases.c.h , which is then included by Python/ceval.c to provide most of the cases of the main interpreter switch.

gh-98817 : Remove PCbuild/lib.pyproj: it’s not used for anything, is only a minor convenience for Visual Studio users (who probably mostly don’t even know about it), and it takes a lot of maintenance effort to keep updated.

gh-98776 : Fix make regen-test-levenshtein for out-of-tree builds.

gh-98707 : Don’t use vendored libmpdec headers if --with-system-libmpdec is passed to configure . Don’t use vendored libexpat headers if --with-system-expat is passed to configure .

gh-98689 : Update Windows builds to zlib v1.2.13. v1.2.12 has CVE-2022-37434 , but the vulnerable inflateGetHeader API is not used by Python.

gh-98790 : Assumes that a missing DLLs directory means that standard extension modules are in the executable’s directory.

gh-98745 : Update py.exe launcher to install 3.11 by default and 3.12 on request.

gh-98692 : Fix the Python Launcher for Windows ignoring unrecognized shebang lines instead of treating them as local paths

gh-94328 : Update Windows installer to use SQLite 3.39.4.

gh-94328 : Update macOS installer to SQLite 3.39.4.

gh-98724 : The Py_CLEAR , Py_SETREF and Py_XSETREF macros now only evaluate their argument once. If the argument has side effects, these side effects are no longer duplicated. Patch by Victor Stinner.

gh-98978 : Fix use-after-free in Py_SetPythonHome(NULL) , Py_SetProgramName(NULL) and _Py_SetProgramFullPath(NULL) function calls. Issue reported by Benedikt Reinartz. Patch by Victor Stinner.

gh-98410 : Add getbufferproc and releasebufferproc to the stable API.

gh-98610 : Some configurable capabilities of sub-interpreters have changed. They always allow subprocesses ( subprocess ) now, whereas before subprocesses could be optionally disallowed for a sub-interpreter. Instead os.exec() can now be disallowed. Disallowing daemon threads is now supported. Disallowing all threads is still allowed, but is never done by default. Note that the optional restrictions are only available through _Py_NewInterpreterFromConfig() , which isn’t a public API. They do not affect the main interpreter, nor Py_NewInterpreter() .

gh-98608 : A _PyInterpreterConfig has been added and _Py_NewInterpreter() has been renamed to _Py_NewInterpreterFromConfig() . The “isolated_subinterpreters” argument is now a granular config that captures the previous behavior. Note that this is all “private” API.

gh-96853 : Py_InitializeEx now correctly calls PyConfig_Clear after initializing the interpreter (the omission didn’t cause a memory leak only because none of the dynamically allocated config fields are populated by the wrapper function)

gh-91248 : Add PyFrame_GetVar() and PyFrame_GetVarString() functions to get a frame variable by its name. Patch by Victor Stinner.

Python 3.12.0 alpha 1 ¶

Release date: 2022-10-25

gh-97616 : Fix multiplying a list by an integer ( list *= int ): detect the integer overflow when the new allocated length is close to the maximum size. Issue reported by Jordan Limor. Patch by Victor Stinner.

gh-97514 : On Linux the multiprocessing module returns to using filesystem backed unix domain sockets for communication with the forkserver process instead of the Linux abstract socket namespace. Only code that chooses to use the “forkserver” start method is affected.

Abstract sockets have no permissions and could allow any user on the system in the same network namespace (often the whole system) to inject code into the multiprocessing forkserver process. This was a potential privilege escalation. Filesystem based socket permissions restrict this to the forkserver process user as was the default in Python 3.8 and earlier.

This prevents Linux CVE-2022-42919 .

gh-87389 : http.server : Fix an open redirection vulnerability in the HTTP server when an URI path starts with // . Vulnerability discovered, and initial fix proposed, by Hamza Avvan.

gh-79096 : LWPCookieJar and MozillaCookieJar create files with file mode 600 instead of 644 (Microsoft Windows is not affected)

gh-92888 : Fix memoryview use after free when accessing the backing buffer in certain cases.

gh-68966 : The deprecated mailcap module now refuses to inject unsafe text (filenames, MIME types, parameters) into shell commands. Instead of using such text, it will warn and act as if a match was not found (or for test commands, as if the test failed).

gh-98374 : Suppress ImportError for invalid query for help() command. Patch by Donghee Na.

gh-98461 : Fix source location in bytecode for list, set and dict comprehensions as well as generator expressions.

gh-98354 : Added unicode check for name attribute of spec argument passed in _imp.create_builtin() function.

gh-98398 : Fix source location of ‘assert’ bytecodes.

gh-98390 : Fix location of sub-expressions of boolean expressions, by reducing their scope to that of the sub-expression.

gh-98254 : Modules from the standard library are now potentially suggested as part of the error messages displayed by the interpreter when an NameError is raised to the top level. Patch by Pablo Galindo

gh-97997 : Add running column offset to the tokenizer state to avoid calculating AST column information with pointer arithmetic.

gh-97973 : Modify the tokenizer to return all necessary information the parser needs to set location information in the AST nodes, so that the parser does not have to calculate those doing pointer arithmetic.

gh-96078 : os.sched_yield() now release the GIL while calling sched_yield(2). Patch by Donghee Na.

gh-97955 : Migrate zoneinfo to Argument Clinic.

gh-97912 : The compiler now avoids quadratic behavior when finding which instructions should use the LOAD_FAST_CHECK opcode.

gh-97002 : Fix an issue where several frame objects could be backed by the same interpreter frame, possibly leading to corrupted memory and hard crashes of the interpreter.

gh-97943 : Bugfix: PyFunction_GetAnnotations() should return a borrowed reference. It was returning a new reference.

gh-97922 : The Garbage Collector now runs only on the eval breaker mechanism of the Python bytecode evaluation loop instead on object allocations. The GC can also run when PyErr_CheckSignals() is called so C extensions that need to run for a long time without executing any Python code also have a chance to execute the GC periodically.

gh-65961 : When __package__ is different than __spec__.parent , raise a DeprecationWarning instead of ImportWarning .

Also remove importlib.util.set_package() which was scheduled for removal.

gh-97850 : Long deprecated, module_repr() should now be completely eradicated.

gh-86298 : In cases where warnings.warn_explicit() consults the module’s loader, an DeprecationWarning is issued when m.__loader__ differs from m.__spec__.loader .

gh-97779 : Ensure that all Python frame objects are backed by “complete” frames.

gh-91052 : Add API for subscribing to modification events on selected dictionaries.

gh-97752 : Fix possible data corruption or crashes when accessing the f_back member of newly-created generator or coroutine frames.

gh-97591 : Fixed a missing incref/decref pair in Exception.__setstate__() . Patch by Ofey Chan.

gh-97670 : Remove the sys.getdxp() function and the Tools/scripts/analyze_dxp.py script. DXP stands for “dynamic execution pairs”. They were related to DYNAMIC_EXECUTION_PROFILE and DXPAIRS macros which have been removed in Python 3.11. Python can now be built with ./configure --enable-pystats to gather statistics on Python opcodes. Patch by Victor Stinner.

gh-94526 : Fix the Python path configuration used to initialized sys.path at Python startup. Paths are no longer encoded to UTF-8/strict to avoid encoding errors if it contains surrogate characters (bytes paths are decoded with the surrogateescape error handler). Patch by Victor Stinner.

gh-96670 : The parser now raises SyntaxError when parsing source code containing null bytes. Patch by Pablo Galindo

gh-96975 : Fix a crash occurring when PyEval_GetFrame() is called while the topmost Python frame is in a partially-initialized state.

gh-96848 : Fix command line parsing: reject -X int_max_str_digits option with no value (invalid) when the PYTHONINTMAXSTRDIGITS environment variable is set to a valid limit. Patch by Victor Stinner.

gh-95921 : Fix overly-broad source position information for chained comparisons used as branching conditions.

gh-96821 : Fix undefined behaviour in audioop.c .

gh-96821 : Fix undefined behaviour in _testcapimodule.c .

gh-95778 : When ValueError is raised if an integer is larger than the limit, mention the sys.set_int_max_str_digits() function in the error message. Patch by Victor Stinner.

gh-96387 : At Python exit, sometimes a thread holding the GIL can wait forever for a thread (usually a daemon thread) which requested to drop the GIL, whereas the thread already exited. To fix the race condition, the thread which requested the GIL drop now resets its request before exiting. Issue discovered and analyzed by Mingliang ZHAO. Patch by Victor Stinner.

gh-96864 : Fix a possible assertion failure, fatal error, or SystemError if a line tracing event raises an exception while opcode tracing is enabled.

gh-95778 : The PyLong_FromString function was refactored to make it more maintainable and extensible.

gh-96678 : Fix undefined behaviour in C code of null pointer arithmetic.

gh-96754 : Make sure that all frame objects created are created from valid interpreter frames. Prevents the possibility of invalid frames in backtraces and signal handlers.

gh-90997 : Improve the performance of reading and writing inline bytecode caches on some platforms.

gh-96751 : Remove dead code from CALL_FUNCTION_EX opcode.

gh-90751 : memoryview now supports half-floats. Patch by Donghee Na and Antoine Pitrou.

gh-96678 : Fix case of undefined behavior in ceval.c

gh-64373 : Convert _functools to argument clinic.

gh-96641 : Do not expose KeyWrapper in _functools .

gh-96636 : Ensure that tracing, sys.setrace() , is turned on immediately. In pre-release versions of 3.11, some tracing events might have been lost when turning on tracing in a __del__ method or interrupt.

gh-96572 : Fix use after free in trace refs build mode. Patch by Kumar Aditya.

gh-96611 : When loading a file with invalid UTF-8 inside a multi-line string, a correct SyntaxError is emitted.

gh-96612 : Make sure that incomplete frames do not show up in tracemalloc traces.

gh-90230 : Fix compiler warnings and test failures when building with --enable-pystats .

gh-96587 : Correctly raise SyntaxError on exception groups ( PEP 654 ) on python versions prior to 3.11

gh-96569 : Remove two cases of undefined behavior, by adding NULL checks.

gh-96582 : Fix possible NULL pointer dereference in _PyThread_CurrentFrames . Patch by Kumar Aditya.

gh-91079 : Separate Python recursion checking from C recursion checking which reduces the chance of C stack overflow and allows the recursion limit to be increased safely.

gh-93911 : Fix an issue that could prevent LOAD_ATTR from specializing properly when accessing properties.

gh-96348 : Emit a DeprecationWarning when throw() , throw() or athrow() are called with more than one argument.

gh-95196 : Disable incorrect pickling of the C implemented classmethod descriptors.

gh-96364 : Fix text signatures of list.__getitem__ and dict.__getitem__ .

gh-96352 : Fix AttributeError missing name and obj attributes in object.__getattribute__() . Patch by Philip Georgi.

gh-93554 : Change the jump opcodes so that all conditional jumps are forward jumps. Backward jumps are converted by the assembler into a conditional forward jump whose target is the fallthrough block (and with a reversed condition), followed by an unconditional backward jump. For example:

POP_JUMP_IF_TRUE BACKWARD_TARGET becomes POP_JUMP_IF_FALSE NEXT_BLOCK; JUMP BACKWARD_TARGET .

All the directed conditional jump opcodes were removed: POP_JUMP_FORWARD_IF_TRUE , POP_JUMP_BACKWARD_IF_TRUE , POP_JUMP_FORWARD_IF_FALSE , POP_JUMP_BACKWARD_IF_FALSE , POP_JUMP_FORWARD_IF_NONE , POP_JUMP_BACKWARD_IF_NONE , POP_JUMP_FORWARD_IF_NOT_NONE , POP_JUMP_BACKWARD_IF_NOT_NONE .

The corresponding opcodes without direction are no longer pseudo-instructions, and they implement the forward conditional jumps.

gh-96268 : Loading a file with invalid UTF-8 will now report the broken character at the correct location.

gh-96237 : The internal field _PyInterpreterFrame.f_func is renamed to _PyInterpreterFrame.f_funcobj and may be any object. The f_globals and f_builtin fields may hold junk values.

It is safest to treat the _PyInterpreterFrame struct as opaque.

gh-96187 : Fixed a bug that caused _PyCode_GetExtra to return garbage for negative indexes. Patch by Pablo Galindo

gh-96143 : Add a new -X perf Python command line option as well as sys.activate_stack_trampoline() and sys.deactivate_stack_trampoline() function in the sys module that allows to set/unset the interpreter in a way that the Linux perf profiler can detect Python calls. The new sys.is_stack_trampoline_active() function allows to query the state of the perf trampoline. Design by Pablo Galindo. Patch by Pablo Galindo and Christian Heimes with contributions from Gregory P. Smith [Google] and Mark Shannon.

gh-96071 : Fix a deadlock in PyGILState_Ensure() when allocating new thread state. Patch by Kumar Aditya.

gh-96046 : PyType_Ready() now initializes ht_cached_keys and performs additional checks to ensure that type objects are properly configured. This avoids crashes in 3rd party packages that don’t use regular API to create new types.

gh-96005 : On WASI ENOTCAPABLE is now mapped to PermissionError . The errno modules exposes the new error number. getpath.py now ignores PermissionError when it cannot open landmark files pybuilddir.txt and pyenv.cfg .

gh-93678 : Added test a harness for direct unit tests of the compiler’s optimization stage. The _testinternalcapi.optimize_cfg() function runs the optimiser on a sequence of instructions. The CfgOptimizationTestCase class in test.support has utilities for invoking the optimizer and checking the output.

gh-95245 : Reduces the size of a “simple” Python object from 8 to 6 words by moving the weakreflist pointer into the pre-header directly before the object’s dict/values pointer.

gh-90997 : Compile virtual try / except blocks to handle exceptions raised during close() or throw() calls through a suspended frame.

gh-95977 : Optimized calling __get__() with vectorcall. Patch by Kumar Aditya.

gh-91210 : Improve error message when a parameter without a default value follows one with a default value, and show the same message, even when the non-default/default sequence is preceded by positional-only parameters.

gh-95922 : Fixed bug where the compiler’s eliminate_empty_basic_blocks function ignores the last block of the code unit.

gh-95818 : Skip over incomplete frames in PyThreadState_GetFrame() .

gh-95876 : Fix format string in _PyPegen_raise_error_known_location that can lead to memory corruption on some 64bit systems. The function was building a tuple with i (int) instead of n (Py_ssize_t) for Py_ssize_t arguments.

gh-95605 : Fix misleading contents of error message when converting an all-whitespace string to float .

gh-95150 : Update code object hashing and equality to consider all debugging and exception handling tables. This fixes an issue where certain non-identical code objects could be “deduplicated” during compilation.

gh-91146 : Reduce allocation size of list from str.split() and str.rsplit() . Patch by Donghee Na and Inada Naoki.

gh-87092 : Create a ‘jump target label’ abstraction in the compiler so that the compiler’s codegen stage does not work directly with basic blocks. This prepares the code for changes to the underlying CFG generation mechanism.

gh-95355 : _PyPegen_Parser_New now properly detects token memory allocation errors. Patch by Honglin Zhu.

gh-90081 : Run Python code in tracer/profiler function at full speed. Fixes slowdown in earlier versions of 3.11.

gh-95324 : Emit a warning in debug mode if an object does not call PyObject_GC_UnTrack() before deallocation. Patch by Pablo Galindo.

gh-95245 : Merge managed dict and values pointer into a single tagged pointer to save one word in the pre-header.

gh-93678 : Add cfg_builder struct and refactor the relevant code so that a cfg can be constructed without an instance of the compiler struct.

gh-95185 : Prevented crashes in the AST constructor when compiling some absurdly long expressions like "+0"*1000000 . RecursionError is now raised instead. Patch by Pablo Galindo

gh-93351 : ast.AST node positions are now validated when provided to compile() and other related functions. If invalid positions are detected, a ValueError will be raised.

gh-94438 : Fix an issue that caused extended opcode arguments and some conditional pops to be ignored when calculating valid jump targets for assignments to the f_lineno attribute of frame objects. In some cases, this could cause inconsistent internal state, resulting in a hard crash of the interpreter.

gh-95060 : Undocumented PyCode_Addr2Location function now properly returns when addrq argument is less than zero.

gh-95113 : Replace all EXTENDED_ARG_QUICK instructions with basic EXTENDED_ARG instructions in unquickened code. Consumers of non-adaptive bytecode should be able to handle extended arguments the same way they were handled in CPython 3.10 and older.

gh-91409 : Fix incorrect source location info caused by certain optimizations in the bytecode compiler.

gh-95023 : Implement os.setns() and os.unshare() for Linux. Patch by Noam Cohen.

gh-94036 : Fix incorrect source location info for some multi-line attribute accesses and method calls.

gh-94938 : Fix error detection in some builtin functions when keyword argument name is an instance of a str subclass with overloaded __eq__ and __hash__ . Previously it could cause SystemError or other undesired behavior.

gh-94996 : ast.parse() will no longer parse function definitions with positional-only params when passed feature_version less than (3, 8) . Patch by Shantanu Jain.

gh-94739 : Allow jumping within, out of, and across exception handlers in the debugger.

gh-94949 : ast.parse() will no longer parse parenthesized context managers when passed feature_version less than (3, 9) . Patch by Shantanu Jain.

gh-94947 : ast.parse() will no longer parse assignment expressions when passed feature_version less than (3, 8) . Patch by Shantanu Jain.

gh-91256 : Ensures the program name is known for help text during interpreter startup.

gh-94869 : Fix the column offsets for some expressions in multi-line f-strings ast nodes. Patch by Pablo Galindo.

gh-94893 : Fix an issue where frame object manipulations could corrupt inline bytecode caches.

gh-94822 : Fix an issue where lookups of metaclass descriptors may be ignored when an identically-named attribute also exists on the class itself.

gh-91153 : Fix an issue where a bytearray item assignment could crash if it’s resized by the new value’s __index__() method.

gh-90699 : Fix reference counting bug in bool.__repr__() . Patch by Kumar Aditya.

gh-94694 : Fix an issue that could cause code with multi-line method lookups to have misleading or incorrect column offset information. In some cases (when compiling a hand-built AST) this could have resulted in a hard crash of the interpreter.

gh-93252 : Fix an issue that caused internal frames to outlive failed Python function calls, possibly resulting in memory leaks or hard interpreter crashes.

gh-94215 : Fix an issue where exceptions raised by line-tracing events would cause frames to be left in an invalid state, possibly resulting in a hard crash of the interpreter.

gh-92228 : Disable the compiler’s inline-small-exit-blocks optimization for exit blocks that are associated with source code lines. This fixes a bug where the debugger cannot tell where an exception handler ends and the following code block begins.

gh-94485 : Line number of a module’s RESUME instruction is set to 0 as specified in PEP 626 .

gh-94438 : Account for instructions that can push NULL to the stack when setting line number in a frame. Prevents some (unlikely) crashes.

gh-91719 : Reload opcode when raising unknown opcode error in the interpreter main loop, for C compilers to generate dispatching code independently.

gh-94329 : Compile and run code with unpacking of extremely large sequences (1000s of elements). Such code failed to compile. It now compiles and runs correctly.

gh-94360 : Fixed a tokenizer crash when reading encoded files with syntax errors from stdin with non utf-8 encoded text. Patch by Pablo Galindo

gh-88116 : Fix an issue when reading line numbers from code objects if the encoded line numbers are close to INT_MIN . Patch by Pablo Galindo

gh-94262 : Don’t create frame objects for incomplete frames. Prevents the creation of generators and closures from being observable to Python and C extensions, restoring the behavior of 3.10 and earlier.

gh-94192 : Fix error for dictionary literals with invalid expression as value.

gh-87995 : types.MappingProxyType instances are now hashable if the underlying mapping is hashable.

gh-93883 : Revise the display strategy of traceback enhanced error locations. The indicators are only shown when the location doesn’t span the whole line.

gh-94163 : Add BINARY_SLICE and STORE_SLICE instructions for more efficient handling and better specialization of slicing operations, where the slice is explicit in the source code.

gh-94021 : Fix unreachable code warning in Python/specialize.c .

gh-93911 : Specialize LOAD_ATTR for objects with custom __getattribute__ .

gh-93955 : Improve performance of attribute lookups on objects with custom __getattribute__ and __getattr__ . Patch by Ken Jin.

gh-93911 : Specialize LOAD_ATTR for property() attributes.

gh-93678 : Refactor compiler optimisation code so that it no longer needs the struct assembler and struct compiler passed around. Instead, each function takes the CFG and other data that it actually needs. This will make it possible to test this code directly.

gh-93841 : When built with -enable-pystats , sys._stats_on() , sys._stats_off() , sys._stats_clear() and sys._stats_dump() functions have been added to enable gathering stats for parts of programs.

gh-93516 : Store offset of first traceable instruction in code object to avoid having to recompute it for each instruction when tracing.

gh-93516 : Lazily create a table mapping bytecode offsets to line numbers to speed up calculation of line numbers when tracing.

gh-89828 : types.GenericAlias no longer relays the __class__ attribute. For example, isinstance(list[int], type) no longer returns True .

gh-93678 : Refactor the compiler to reduce boilerplate and repetition.

gh-93671 : Fix some exponential backtrace case happening with deeply nested sequence patterns in match statements. Patch by Pablo Galindo

gh-93662 : Make sure that the end column offsets are correct in multi-line method calls. Previously, the end column could precede the column offset.

gh-93461 : importlib.invalidate_caches() now drops entries from sys.path_importer_cache with a relative path as name. This solves a caching issue when a process changes its current working directory.

FileFinder no longer inserts a dot in the path, e.g. /egg/./spam is now /egg/spam .

gh-93621 : Change order of bytecode instructions emitted for with and async with to reduce the number of entries in the exception table.

gh-93533 : Reduce the size of the inline cache for LOAD_METHOD by 2 bytes.

gh-93444 : Removed redundant fields from the compiler’s basicblock struct: b_nofallthrough , b_exit , b_return . They can be easily calculated from the opcode of the last instruction of the block.

gh-93429 : LOAD_METHOD instruction has been removed. It was merged back into LOAD_ATTR .

gh-93418 : Fixed an assert where an f-string has an equal sign ‘=’ following an expression, but there’s no trailing brace. For example, f”{i=”.

gh-93382 : Cache the result of PyCode_GetCode() function to restore the O (1) lookup of the co_code attribute.

gh-93359 : Ensure that custom ast nodes without explicit end positions can be compiled. Patch by Pablo Galindo.

gh-93356 : Code for exception handlers is emitted at the end of the code unit’s bytecode. This avoids one jump when no exception is raised.

gh-93354 : Use exponential backoff for specialization counters in the interpreter. Can reduce the number of failed specializations significantly and avoid slowdown for those parts of a program that are not suitable for specialization.

gh-93283 : Improve error message for invalid syntax of conversion character in f-string expressions.

gh-93345 : Fix a crash in substitution of a TypeVar in nested generic alias after TypeVarTuple .

gh-93223 : When a bytecode instruction jumps to an unconditional jump instruction, the first instruction can often be optimized to target the unconditional jump’s target directly. For tracing reasons, this would previously only occur if both instructions have the same line number. This also now occurs if the unconditional jump is artificial, i.e., if it has no associated line number.

gh-84694 : The --experimental-isolated-subinterpreters configure option and EXPERIMENTAL_ISOLATED_SUBINTERPRETERS macro have been removed.

gh-91924 : Fix __lltrace__ debug feature if the stdout encoding is not UTF-8. Patch by Victor Stinner.

gh-93040 : Wraps unused parameters in Objects/obmalloc.c with Py_UNUSED .

gh-93143 : Avoid NULL checks for uninitialized local variables by determining at compile time which variables must be initialized.

gh-93061 : Backward jumps after async for loops are no longer given dubious line numbers.

gh-93065 : Fix contextvars HAMT implementation to handle iteration over deep trees.

The bug was discovered and fixed by Eli Libman. See MagicStack/immutables#84 for more details.

gh-93012 : Added the new function PyType_FromMetaclass() , which generalizes the existing PyType_FromModuleAndSpec() using an additional metaclass argument. This is useful for language binding tools, where it can be used to intercept type-related operations like subclassing or static attribute access by specifying a metaclass with custom slots.

Importantly, PyType_FromMetaclass() is available in the Limited API, which provides a path towards migrating more binding tools onto the Stable ABI.

gh-93021 : Fix the __text_signature__ for __get__() methods implemented in C. Patch by Jelle Zijlstra.

gh-89914 : The operand of the YIELD_VALUE instruction is set to the stack depth. This is done to help frame handling on yield and may assist debuggers.

gh-92955 : Fix memory leak in code object’s lines and positions iterators as they were not finalized at exit. Patch by Kumar Aditya.

gh-92930 : Fixed a crash in _pickle.c from mutating collections during __reduce__ or persistent_id .

gh-90690 : The PRECALL instruction has been removed. It offered only a small advantage for specialization and is not needed in the vast majority of cases.

gh-92914 : Always round the allocated size for lists up to the nearest even number.

gh-92858 : Improve error message for some suites with syntax error before ‘:’

gh-90473 : Decrease default recursion limit on WASI to address limited call stack size.

gh-92804 : Fix memory leak in memoryview iterator as it was not finalized at exit. Patch by Kumar Aditya.

gh-92777 : Specialize LOAD_METHOD for objects with lazy dictionaries. Patch by Ken Jin.

gh-92658 : Add support for connecting and binding to Hyper-V sockets on Windows Hyper-V hosts and guests.

gh-92236 : Remove spurious “LINE” event when starting a generator or coroutine, visible tracing functions implemented in C.

gh-91102 : _warnings.warn_explicit() is ported to Argument Clinic.

gh-92619 : Make the compiler duplicate an exit block only if none of its instructions have a lineno (previously only the first instruction in the block was checked, leading to unnecessarily duplicated blocks).

gh-88750 : The deprecated debug build only PYTHONTHREADDEBUG environment variable no longer does anything.

gh-92261 : Fix hang when trying to iterate over a typing.Union .

gh-91432 : Specialized the FOR_ITER opcode using the PEP 659 machinery

gh-91399 : Removed duplicate ‘{0, 0, 0, 0, 0, 0}’ entry in ‘Objects/unicodetype_db.h’.

gh-91578 : Updates the error message for abstract class.

bpo-47091 : Improve performance of repetition of list and tuple by using memcpy to copy data and performing the reference increments in one step.

bpo-46142 : Make --help output shorter by moving some info to the new --help-env and --help-xoptions command-line options. Also add --help-all option to print complete usage.

bpo-42316 : Document some places where an assignment expression needs parentheses.

gh-89237 : Fix hang on Windows in subprocess.wait_closed() in asyncio with ProactorEventLoop . Patch by Kumar Aditya.

gh-97928 : tkinter.Text.count() raises now an exception for options starting with “-” instead of silently ignoring them.

gh-98393 : The os module no longer accepts bytes-like paths, like bytearray and memoryview types: only the exact bytes type is accepted for bytes strings. Patch by Victor Stinner.

gh-98363 : Added itertools.batched() to batch data into lists of a given length with the last list possibly being shorter than the others.

gh-98331 : Update the bundled copies of pip and setuptools to versions 22.3 and 65.5.0 respectively.

gh-98307 : A createSocket() method was added to SysLogHandler .

gh-96035 : Fix bug in urllib.parse.urlparse() that causes certain port numbers containing whitespace, underscores, plus and minus signs, or non-ASCII digits to be incorrectly accepted.

gh-98257 : Make sys.setprofile() and sys.settrace() functions reentrant. They can no long fail with: RuntimeError("Cannot install a trace function while another trace function is being installed") . Patch by Victor Stinner.

gh-98251 : Allow venv to pass along PYTHON* variables to ensurepip and pip when they do not impact path resolution

gh-94597 : Deprecated asyncio.AbstractEventLoopPolicy.get_child_watcher() and asyncio.AbstractEventLoopPolicy.set_child_watcher() methods to be removed in Python 3.14. Patch by Kumar Aditya.

gh-98178 : On macOS, fix a crash in syslog.syslog() in multi-threaded applications. On macOS, the libc syslog() function is not thread-safe, so syslog.syslog() no longer releases the GIL to call it. Patch by Victor Stinner.

gh-44098 : Release the GIL when creating mmap.mmap objects on Unix.

gh-87730 : Wrap network errors consistently in urllib FTP support, so the test suite doesn’t fail when a network is available but the public internet is not reachable.

gh-94597 : The child watcher classes MultiLoopChildWatcher , FastChildWatcher and SafeChildWatcher are deprecated and will be removed in Python 3.14. Patch by Kumar Aditya.

gh-98023 : Change default child watcher to PidfdChildWatcher on Linux systems which supports it. Patch by Kumar Aditya.

gh-90985 : Earlier in 3.11 we deprecated asyncio.Task.cancel("message") . We realized we were too harsh, and have undeprecated it.

gh-65961 : Do not rely solely on __cached__ on modules; code will also support __spec__.cached .

gh-97646 : Replace deprecated application/javascript with text/javascript in mimetypes . See RFC 9239 . Patch by Noam Cohen.

gh-97930 : Apply changes from importlib_resources 5.8 and 5.9: Traversable.joinpath provides a concrete implementation. as_file now supports directories of resources.

gh-97850 : Remove deprecated importlib.util.set_loader() and importlib.util.module_for_loader() from importlib.util .

gh-97837 : Change deprecate warning message in unittest from

It is deprecated to return a value!=None

It is deprecated to return a value that is not None from a test case

gh-97825 : Fixes AttributeError when subprocess.check_output() is used with argument input=None and either of the arguments encoding or errors are used.

gh-97008 : NameError and AttributeError spelling suggestions provided since gh-82711 are now also emitted by the pure Python traceback module. Tests for those suggestions now exercise both implementations to ensure they are equivalent. Patch by Carl Friedrich Bolz-Tereick and Łukasz Langa.

gh-97799 : dataclass now uses inspect.get_annotations() to examine the annotations on class objects.

gh-97781 : Removed deprecated interfaces in importlib.metadata (entry points accessed as dictionary, implicit dictionary construction of sequence of EntryPoint objects, mutablility of EntryPoints result, access of entry point by index). entry_points now has a simpler, more straightforward API (returning EntryPoints ).

gh-96827 : Avoid spurious tracebacks from asyncio when default executor cleanup is delayed until after the event loop is closed (e.g. as the result of a keyboard interrupt).

gh-95534 : gzip.GzipFile.read() reads 10% faster.

gh-97592 : Avoid a crash in the C version of asyncio.Future.remove_done_callback() when an evil argument is passed.

gh-97639 : Remove tokenize.NL check from tabnanny .

gh-97545 : Make Semaphore run faster.

gh-73588 : Fix generation of the default name of tkinter.Checkbutton . Previously, checkbuttons in different parent widgets could have the same short name and share the same state if arguments “name” and “variable” are not specified. Now they are globally unique.

gh-96865 : fix Flag to use boundary CONFORM

This restores previous Flag behavior of allowing flags with non-sequential values to be combined; e.g.

class Skip(Flag): TWO = 2 EIGHT = 8

Skip.TWO | Skip.EIGHT -> <Skip.TWO|EIGHT: 10>

gh-97005 : Update bundled libexpat to 2.4.9

gh-85760 : Fix race condition in asyncio where process_exited() called before the pipe_data_received() leading to inconsistent output. Patch by Kumar Aditya.

gh-96704 : Pass the correct contextvars.Context when a asyncio exception handler is called on behalf of a task or callback handle. This adds a new Task method, get_context , and also a new Handle method with the same name. If this method is not found on a task object (perhaps because it is a third-party library that does not yet provide this method), the context prevailing at the time the exception handler is called is used.

gh-96819 : Fixed check in multiprocessing.resource_tracker that guarantees that the length of a write to a pipe is not greater than PIPE_BUF .

gh-95865 : Reduce urllib.parse.quote_from_bytes() memory use on large values.

Contributed by Dennis Sweeney.

gh-96741 : Corrected type annotation for dataclass attribute pstats.FunctionProfile.ncalls to be str .

gh-96734 : Update unicodedata database to Unicode 15.0.0.

gh-96735 : Fix undefined behaviour in struct.unpack() .

gh-46412 : Improve performance of bool(db) for large ndb/gdb databases. Previously this would call len(db) which would iterate over all keys – the answer (empty or not) is known after the first key.

gh-96652 : Fix the faulthandler implementation of faulthandler.register(signal, chain=True) if the sigaction() function is not available: don’t call the previous signal handler if it’s NULL. Patch by Victor Stinner.

gh-68163 : Correct conversion of numbers.Rational ’s to float .

gh-96538 : Speed up bisect.bisect() functions by taking advantage of type-stability.

gh-96465 : Fraction hashes are now cached.

gh-96079 : In typing , fix missing field name and incorrect __module__ in _AnnotatedAlias.

gh-96415 : Remove types._cell_factory from module namespace.

gh-95987 : Fix repr of Any subclasses.

gh-96388 : Work around missing socket functions in socket ’s __repr__ .

gh-96385 : Fix TypeVarTuple.__typing_prepare_subst__ . TypeError was not raised when using more than one TypeVarTuple , like [*T, *V] in type alias substitutions.

gh-96142 : Add match_args , kw_only , slots , and weakref_slot to _DataclassParams .

gh-96073 : In inspect , fix overeager replacement of “ typing. ” in formatting annotations.

gh-89258 : Added a getChildren() method to logging.Logger , to get the immediate child loggers of a logger.

gh-96346 : Use double caching for compiled RE patterns.

gh-96349 : Fixed a minor performance regression in threading.Event.__init__()

gh-90467 : Fix asyncio.streams.StreamReaderProtocol to keep a strong reference to the created task, so that it’s not garbage collected

gh-96172 : Fix a bug in unicodedata : east_asian_width used to return the wrong value for unassigned characters; and for yet unassigned, but reserved characters.

gh-96159 : Fix a performance regression in logging TimedRotatingFileHandler. Only check for special files when the rollover time has passed.

gh-96175 : Fix unused localName parameter in the Attr class in xml.dom.minidom .

gh-96145 : Add AttrDict to JSON module for use with object_hook.

gh-96052 : Fix handling compiler warnings (SyntaxWarning and DeprecationWarning) in codeop.compile_command() when checking for incomplete input. Previously it emitted warnings and raised a SyntaxError. Now it always returns None for incomplete input without emitting any warnings.

gh-96125 : Fix incorrect condition that causes sys.thread_info.name to be wrong on pthread platforms.

gh-96019 : Fix a bug in the makeunicodedata.py script leading to about 13 KiB of space saving in the unicodedata module, specifically the character decomposition data.

gh-95463 : Remove an incompatible change from bpo-28080 that caused a regression that ignored the utf8 in ZipInfo.flag_bits . Patch by Pablo Galindo.

gh-69142 : Add %:z strftime format code (generates tzoffset with colons as separator), see strftime() and strptime() Behavior .

gh-95899 : Fix asyncio.Runner to call asyncio.set_event_loop() only once to avoid calling attach_loop() multiple times on child watchers. Patch by Kumar Aditya.

gh-95736 : Fix unittest.IsolatedAsyncioTestCase to set event loop before calling setup functions. Patch by Kumar Aditya.

gh-95865 : Speed up urllib.parse.quote_from_bytes() by replacing a list comprehension with map() .

gh-95861 : Add support for computing Spearman’s correlation coefficient to the existing statistics.correlation() function.

gh-95804 : Fix logging shutdown handler so it respects MemoryHandler.flushOnClose .

gh-95704 : When a task catches asyncio.CancelledError and raises some other error, the other error should generally not silently be suppressed.

gh-95149 : The HTTPStatus enum offers a couple of properties to indicate the HTTP status category e.g. HTTPStatus.OK.is_success .

gh-95609 : Update bundled pip to 22.2.2.

gh-95289 : Fix asyncio.TaskGroup to propagate exception when asyncio.CancelledError was replaced with another exception by a context manager. Patch by Kumar Aditya and Guido van Rossum.

gh-94909 : Fix incorrect joining of relative Windows paths with drives in pathlib.PurePath initializer.

gh-95385 : Faster json.dumps() when sorting of keys is not requested (default).

gh-83901 : Improve Signature.bind error message for missing keyword-only arguments.

gh-95339 : Update bundled pip to 22.2.1.

gh-95045 : Fix GC crash when deallocating _lsprof.Profiler by untracking it before calling any callbacks. Patch by Kumar Aditya.

gh-95231 : Fail gracefully if EPERM or ENOSYS is raised when loading crypt methods. This may happen when trying to load MD5 on a Linux kernel with FIPS enabled.

gh-95097 : Fix asyncio.run() for asyncio.Task implementations without uncancel() method. Patch by Kumar Aditya.

gh-95087 : Fix IndexError in parsing invalid date in the email module.

gh-95199 : Upgrade bundled setuptools to 63.2.0.

gh-95194 : Upgrade bundled pip to 22.2.

gh-93899 : Fix check for existence of os.EFD_CLOEXEC , os.EFD_NONBLOCK and os.EFD_SEMAPHORE flags on older kernel versions where these flags are not present. Patch by Kumar Aditya.

gh-95166 : Fix concurrent.futures.Executor.map() to cancel the currently waiting on future on an error - e.g. TimeoutError or KeyboardInterrupt.

gh-95132 : Fix a sqlite3 regression where *args and **kwds were incorrectly relayed from connect() to the Connection factory. The regression was introduced in 3.11a1 with PR 24421 ( gh-85128 ). Patch by Erlend E. Aasland.

gh-93157 : Fix fileinput module didn’t support errors option when inplace is true.

gh-91212 : Fixed flickering of the turtle window when the tracer is turned off. Patch by Shin-myoung-serp.

gh-95077 : Add deprecation warning for enum member.member access (e.g. Color.RED.BLUE ). Remove EnumMeta.__getattr__ .

gh-95109 : Ensure that timeouts scheduled with asyncio.Timeout that have already expired are delivered promptly.

gh-95105 : wsgiref.types.InputStream.__iter__() should return Iterator[bytes] , not Iterable[bytes] . Patch by Shantanu Jain.

gh-95066 : Replaced assert with exception in ast.parse() , when feature_version has an invalid major version. Patch by Shantanu Jain.

gh-77617 : Add sqlite3 command-line interface . Patch by Erlend Aasland.

gh-95005 : Replace _PyAccu with _PyUnicodeWriter in JSON encoder and StringIO and remove the _PyAccu implementation.

gh-90085 : Remove -c/--clock and -t/--time CLI options of timeit . The options had been deprecated since Python 3.3 and the functionality was removed in Python 3.7. Patch by Shantanu Jain.

gh-94857 : Fix refleak in _io.TextIOWrapper.reconfigure . Patch by Kumar Aditya.

gh-94821 : Fix binding of unix socket to empty address on Linux to use an available address from the abstract namespace, instead of “0”.

gh-94736 : Fix crash when deallocating an instance of a subclass of _multiprocessing.SemLock . Patch by Kumar Aditya.

gh-81620 : Add random.binomialvariate().

gh-74116 : Allow asyncio.StreamWriter.drain() to be awaited concurrently by multiple tasks. Patch by Kumar Aditya.

gh-87822 : When called with capture_locals=True , the traceback module functions swallow exceptions raised from calls to repr() on local variables of frames. This is in order to prioritize the original exception over rendering errors. An indication of the failure is printed in place of the missing value. (Patch by Simon-Martin Schroeder).

gh-88050 : Fix asyncio subprocess transport to kill process cleanly when process is blocked and avoid RuntimeError when loop is closed. Patch by Kumar Aditya.

gh-94637 : SSLContext.set_default_verify_paths() now releases the GIL around SSL_CTX_set_default_verify_paths call. The function call performs I/O and CPU intensive work.

gh-94309 : Deprecate aliases typing.Hashable and typing.Sized

gh-92546 : An undocumented python -m pprint benchmark is moved into pprint suite of pyperformance. Patch by Oleg Iarygin.

gh-94607 : Fix subclassing complex generics with type variables in typing . Previously an error message saying Some type variables ... are not listed in Generic[...] was shown. typing no longer populates __parameters__ with the __parameters__ of a Python class.

gh-94619 : Remove the long-deprecated module_repr() from importlib .

gh-93910 : The ability to access the other values of an enum on an enum (e.g. Color.RED.BLUE ) has been restored in order to fix a performance regression.

gh-93896 : Fix asyncio.run() and unittest.IsolatedAsyncioTestCase to always the set event loop as it was done in Python 3.10 and earlier. Patch by Kumar Aditya.

gh-94343 : Allow setting the attributes of reprlib.Repr during object initialization

gh-94382 : Port static types of _multiprocessing module to heap types. Patch by Kumar Aditya.

gh-78724 : Fix crash in struct.Struct when it was not completely initialized by initializing it in __new__() . Patch by Kumar Aditya.

gh-94510 : Re-entrant calls to sys.setprofile() and sys.settrace() now raise RuntimeError . Patch by Pablo Galindo.

gh-92336 : Fix bug where linecache.getline() fails on bad files with UnicodeDecodeError or SyntaxError . It now returns an empty string as per the documentation.

gh-94398 : Once a asyncio.TaskGroup has started shutting down (i.e., at least one task has failed and the task group has started cancelling the remaining tasks), it should not be possible to add new tasks to the task group.

gh-94383 : xml.etree : Remove the ElementTree.Element.copy() method of the pure Python implementation, deprecated in Python 3.10, use the copy.copy() function instead. The C implementation of xml.etree has no copy() method, only a __copy__() method. Patch by Victor Stinner.

gh-94379 : zipimport : Remove find_loader() and find_module() methods, deprecated in Python 3.10: use the find_spec() method instead. See PEP 451 for the rationale. Patch by Victor Stinner.

gh-94352 : shlex.split() : Passing None for s argument now raises an exception, rather than reading sys.stdin . The feature was deprecated in Python 3.9. Patch by Victor Stinner.

gh-94318 : Strip trailing spaces in pydoc text output.

gh-89988 : Fix memory leak in pickle.Pickler when looking up dispatch_table . Patch by Kumar Aditya.

gh-90016 : Deprecate sqlite3 default adapters and converters . Patch by Erlend E. Aasland.

gh-94254 : Fixed types of struct module to be immutable. Patch by Kumar Aditya.

gh-93259 : Now raise ValueError when None or an empty string are passed to Distribution.from_name (and other callers).

gh-74696 : shutil.make_archive() now passes the root_dir argument to custom archivers which support it.

gh-94216 : The dis module now has the opcodes for pseudo instructions (those which are used by the compiler during code generation but then removed or replaced by real opcodes before the final bytecode is emitted).

gh-93096 : Removed undocumented python -m codecs . Use python -m unittest test.test_codecs.EncodedFileTest instead.

gh-94207 : Made _struct.Struct GC-tracked in order to fix a reference leak in the _struct module.

gh-93096 : Removed undocumented -t argument of python -m base64 . Use python -m unittest test.test_base64.LegacyBase64TestCase.test_encodebytes instead.

gh-94226 : Remove the locale.format() function, deprecated in Python 3.7: use locale.format_string() instead. Patch by Victor Stinner.

gh-94199 : Remove the ssl.match_hostname() function. The ssl.match_hostname() was deprecated in Python 3.7. OpenSSL performs hostname matching since Python 3.7, Python no longer uses the ssl.match_hostname() function. Patch by Victor Stinner.

gh-94214 : Document the context object used in the venv.EnvBuilder class, and add the new environment’s library path to it.

gh-94199 : Remove the ssl.wrap_socket() function, deprecated in Python 3.7: instead, create a ssl.SSLContext object and call its ssl.SSLContext.wrap_socket method. Any package that still uses ssl.wrap_socket() is broken and insecure. The function neither sends a SNI TLS extension nor validates server hostname. Code is subject to CWE-295 Improper Certificate Validation. Patch by Victor Stinner.

gh-94199 : Remove the ssl.RAND_pseudo_bytes() function, deprecated in Python 3.6: use os.urandom() or ssl.RAND_bytes() instead. Patch by Victor Stinner.

gh-94199 : hashlib : Remove the pure Python implementation of hashlib.pbkdf2_hmac() , deprecated in Python 3.10. Python 3.10 and newer requires OpenSSL 1.1.1 ( PEP 644 ): this OpenSSL version provides a C implementation of pbkdf2_hmac() which is faster. Patch by Victor Stinner.

gh-94196 : gzip : Remove the filename attribute of gzip.GzipFile , deprecated since Python 2.6, use the name attribute instead. In write mode, the filename attribute added '.gz' file extension if it was not present. Patch by Victor Stinner.

gh-94182 : run the asyncio.PidfdChildWatcher on the running loop, this allows event loops to run subprocesses when there is no default event loop running on the main thread

gh-94169 : Remove io.OpenWrapper and _pyio.OpenWrapper , deprecated in Python 3.10: just use open() instead. The open() ( io.open() ) function is a built-in function. Since Python 3.10, _pyio.open() is also a static method. Patch by Victor Stinner.

gh-91742 : Fix pdb crash after jump caused by a null pointer dereference. Patch by Kumar Aditya.

gh-94101 : Manual instantiation of ssl.SSLSession objects is no longer allowed as it lead to misconfigured instances that crashed the interpreter when attributes where accessed on them.

gh-84753 : inspect.iscoroutinefunction() , inspect.isgeneratorfunction() , and inspect.isasyncgenfunction() now properly return True for duck-typed function-like objects like instances of unittest.mock.AsyncMock .

This makes inspect.iscoroutinefunction() consistent with the behavior of asyncio.iscoroutinefunction() . Patch by Mehdi ABAAKOUK.

gh-94028 : Fix a regression in the sqlite3 where statement objects were not properly cleared and reset after use in cursor iters. The regression was introduced by PR 27884 in Python 3.11a1. Patch by Erlend E. Aasland.

gh-93973 : Add keyword argument all_errors to asyncio.create_connection so that multiple connection errors can be raised as an ExceptionGroup .

gh-93963 : Officially deprecate from importlib.abc classes moved to importlib.resources.abc .

gh-93858 : Prevent error when activating venv in nested fish instances.

gh-93820 : Pickle enum.Flag by name.

gh-93847 : Fix repr of enum of generic aliases.

gh-91404 : Revert the re memory leak when a match is terminated by a signal or memory allocation failure as the implemented fix caused a major performance regression.

gh-83499 : Fix double closing of file description in tempfile .

gh-93820 : Fixed a regression when copy.copy() -ing enum.Flag with multiple flag members.

gh-79512 : Fixed names and __module__ value of weakref classes ReferenceType , ProxyType , CallableProxyType . It makes them pickleable.

gh-91389 : Fix an issue where dis utilities could report missing or incorrect position information in the presence of CACHE entries.

gh-93626 : Set __future__.annotations to have a None mandatoryRelease to indicate that it is currently ‘TBD’.

gh-90473 : Emscripten and WASI have no home directory and cannot provide PEP 370 user site directory.

gh-90494 : copy.copy() and copy.deepcopy() now always raise a TypeError if __reduce__() returns a tuple with length 6 instead of silently ignore the 6th item or produce incorrect result.

gh-90549 : Fix a multiprocessing bug where a global named resource (such as a semaphore) could leak when a child process is spawned (as opposed to forked).

gh-93521 : Fixed a case where dataclasses would try to add __weakref__ into the __slots__ for a dataclass that specified weakref_slot=True when it was already defined in one of its bases. This resulted in a TypeError upon the new class being created.

gh-79579 : sqlite3 now correctly detects DML queries with leading comments. Patch by Erlend E. Aasland.

gh-93421 : Update sqlite3.Cursor.rowcount when a DML statement has run to completion. This fixes the row count for SQL queries like UPDATE ... RETURNING . Patch by Erlend E. Aasland.

gh-93475 : Expose FICLONE and FICLONERANGE constants in fcntl . Patch by Illia Volochii.

gh-93370 : Deprecate version and version_info .

gh-91810 : Suppress writing an XML declaration in open files in ElementTree.write() with encoding='unicode' and xml_declaration=None .

gh-91162 : Support splitting of unpacked arbitrary-length tuple over TypeVar and TypeVarTuple parameters. For example:

A[T, *Ts][*tuple[int, ...]] -> A[int, *tuple[int, ...]]

A[*Ts, T][*tuple[int, ...]] -> A[*tuple[int, ...], int]

gh-93353 : Fix the importlib.resources.as_file() context manager to remove the temporary file if destroyed late during Python finalization: keep a local reference to the os.remove() function. Patch by Victor Stinner.

gh-83658 : Make multiprocessing.Pool raise an exception if maxtasksperchild is not None or a positive int.

gh-93312 : Add os.PIDFD_NONBLOCK flag to open a file descriptor for a process with os.pidfd_open() in non-blocking mode. Patch by Kumar Aditya.

gh-88123 : Implement Enum.__contains__ that returns True or False to replace the deprecated behaviour that would sometimes raise a TypeError .

gh-93297 : Make asyncio task groups prevent child tasks from being GCed

gh-85308 : Changed argparse.ArgumentParser to use filesystem encoding and error handler instead of default text encoding to read arguments from file (e.g. fromfile_prefix_chars option). This change affects Windows; argument file should be encoded with UTF-8 instead of ANSI Codepage.

gh-93156 : Accessing the pathlib.PurePath.parents sequence of an absolute path using negative index values produced incorrect results.

gh-93162 : Add the ability for logging.config.dictConfig() to usefully configure QueueHandler and QueueListener as a pair, and add logging.getHandlerByName() and logging.getHandlerNames() APIs to allow access to handlers by name.

gh-93243 : The smtpd module was removed per the schedule in PEP 594 .

gh-92886 : Replace assert statements with raise AssertionError() in BaseHandler so that the tested behaviour is maintained running with optimizations (-O) .

gh-90155 : Fix broken asyncio.Semaphore when acquire is cancelled.

gh-90817 : The locale.resetlocale() function is deprecated and will be removed in Python 3.13. Use locale.setlocale(locale.LC_ALL, "") instead. Patch by Victor Stinner.

gh-91513 : Added taskName attribute to logging module for use with asyncio tasks.

gh-74696 : shutil.make_archive() no longer temporarily changes the current working directory during creation of standard .zip or tar archives.

gh-92728 : The re.template() function and the corresponding re.TEMPLATE and re.T flags are restored after they were removed in 3.11.0b1, but they are now deprecated, so they might be removed from Python 3.13.

gh-93033 : Search in some strings (platform dependent i.e [U+0xFFFF, U+0x0100] on Windows or [U+0xFFFFFFFF, U+0x00010000] on Linux 64-bit) are now up to 10 times faster.

gh-89973 : Fix re.error raised in fnmatch if the pattern contains a character range with upper bound lower than lower bound (e.g. [c-a] ). Now such ranges are interpreted as empty ranges.

gh-93044 : No longer convert the database argument of sqlite3.connect() to bytes before passing it to the factory.

gh-93010 : In a very special case, the email package tried to append the nonexistent InvalidHeaderError to the defect list. It should have been InvalidHeaderDefect .

gh-92986 : Fix ast.unparse() when ImportFrom.level is None

gh-92932 : Now dis() and get_instructions() handle operand values for instructions prefixed by EXTENDED_ARG_QUICK . Patch by Sam Gross and Donghee Na.

gh-92675 : Fix venv.ensure_directories() to accept pathlib.Path arguments in addition to str paths. Patch by David Foster.

gh-87901 : Removed the encoding argument from os.popen() that was added in 3.11b1.

gh-91922 : Fix function sqlite.connect() and the sqlite.Connection constructor on non-UTF-8 locales. Also, they now support bytes paths non-decodable with the current FS encoding.

gh-92869 : Added c_time_t to ctypes , which has the same size as the time_t type in C.

gh-92839 : Fixed crash resulting from calling bisect.insort() or bisect.insort_left() with the key argument not equal to None .

gh-90473 : subprocess now fails early on Emscripten and WASI platforms to work around missing os.pipe() on WASI.

gh-89325 : Removed many old deprecated unittest features: TestCase method aliases, undocumented and broken TestCase method assertDictContainsSubset , undocumented TestLoader.loadTestsFromModule parameter use_load_tests , and an underscored alias of the TextTestResult class.

gh-92734 : Allow multi-element reprs emitted by reprlib to be pretty-printed using configurable indentation.

gh-92671 : Fixed ast.unparse() for empty tuples in the assignment target context.

gh-91581 : utcfromtimestamp() no longer attempts to resolve fold in the pure Python implementation, since the fold is never 1 in UTC. In addition to being slightly faster in the common case, this also prevents some errors when the timestamp is close to datetime.min . Patch by Paul Ganssle.

gh-86388 : Removed randrange() functionality deprecated since Python 3.10. Formerly, randrange(10.0) losslessly converted to randrange(10). Now, it raises a TypeError. Also, the exception raised for non-integral values such as randrange(10.5) or randrange(‘10’) has been changed from ValueError to TypeError.

gh-90385 : Add pathlib.Path.walk() as an alternative to os.walk() .

gh-92550 : Fix pathlib.Path.rglob() for empty pattern.

gh-92591 : Allow logging filters to return a logging.LogRecord instance so that filters attached to logging.Handler s can enrich records without side effects on other handlers.

gh-92445 : Fix a bug in argparse where nargs="*" would raise an error instead of returning an empty list when 0 arguments were supplied if choice was also defined in parser.add_argument .

gh-92547 : Remove undocumented sqlite3 features deprecated in Python 3.10:

sqlite3.enable_shared_cache()

sqlite3.OptimizedUnicode

gh-92530 : Fix an issue that occurred after interrupting threading.Condition.notify() .

gh-92531 : The statistics.median_grouped() function now always return a float. Formerly, it did not convert the input type when for sequences of length one.

gh-84131 : The pathlib.Path deprecated method link_to has been removed. Use 3.10’s hardlink_to() method instead as its semantics are consistent with that of symlink_to() .

gh-89336 : Removed configparser module APIs: the SafeConfigParser class alias, the ParsingError.filename property and parameter, and the ConfigParser.readfp method, all of which were deprecated since Python 3.2.

gh-92391 : Add __class_getitem__() to csv.DictReader and csv.DictWriter , allowing them to be parameterized at runtime. Patch by Marc Mueller.

gh-91968 : Add SO_RTABLE and SO_USER_COOKIE constants to socket .

gh-91810 : ElementTree method write() and function tostring() now use the text file’s encoding (“UTF-8” if not available) instead of locale encoding in XML declaration when encoding="unicode" is specified.

gh-81790 : os.path.splitdrive() now understands DOS device paths with UNC links (beginning \\?\UNC\ ). Contributed by Barney Gale.

gh-91760 : Apply more strict rules for numerical group references and group names in regular expressions. Only sequence of ASCII digits is now accepted as a numerical reference. The group name in bytes patterns and replacement strings can now only contain ASCII letters and digits and underscore.

gh-90622 : Worker processes for concurrent.futures.ProcessPoolExecutor are no longer spawned on demand (a feature added in 3.9) when the multiprocessing context start method is "fork" as that can lead to deadlocks in the child processes due to a fork happening while threads are running.

gh-91577 : Move imports in SharedMemory methods to module level so that they can be executed late in python finalization.

gh-91581 : Remove an unhandled error case in the C implementation of calls to datetime.fromtimestamp with no time zone (i.e. getting a local time from an epoch timestamp). This should have no user-facing effect other than giving a possibly more accurate error message when called with timestamps that fall on 10000-01-01 in the local time. Patch by Paul Ganssle.

gh-91539 : Improve performance of urllib.request.getproxies_environment when there are many environment variables

gh-91524 : Speed up the regular expression substitution (functions re.sub() and re.subn() and corresponding re.Pattern methods) for replacement strings containing group references by 2–3 times.

gh-91447 : Fix findtext in the xml module to only give an empty string when the text attribute is set to None .

gh-91456 : Deprecate current default auto() behavior: In 3.13 the default will be for for auto() to always return the largest member value incremented by 1, and to raise if incompatible value types are used.

bpo-47231 : Fixed an issue with inconsistent trailing slashes in tarfile longname directories.

bpo-39064 : zipfile.ZipFile now raises zipfile.BadZipFile instead of ValueError when reading a corrupt zip file in which the central directory offset is negative.

bpo-41287 : Fix handling of the doc argument in subclasses of property() .

gh-90005 : ctypes dependency libffi is now detected with pkg-config .

bpo-32547 : The constructors for DictWriter and DictReader now coerce the fieldnames argument to a list if it is an iterator.

bpo-35540 : Fix dataclasses.asdict() crash when collections.defaultdict is present in the attributes.

bpo-47063 : Add an index_pages parameter to support using non-default index page names.

bpo-47025 : Drop support for bytes on sys.path .

bpo-46951 : Order the contents of zipapp archives, to make builds more reproducible.

bpo-42777 : Implement pathlib.Path.is_mount() for Windows paths.

bpo-46755 : In QueueHandler , clear stack_info from LogRecord to prevent stack trace from being written twice.

bpo-45393 : Fix the formatting for await x and not x in the operator precedence table when using the help() system.

bpo-46642 : Improve error message when trying to subclass an instance of typing.TypeVar , typing.ParamSpec , typing.TypeVarTuple , etc. Based on patch by Gregory Beauregard.

bpo-46364 : Restrict use of sockets instead of pipes for stdin of subprocesses created by asyncio to AIX platform only.

bpo-28249 : Set doctest.DocTest.lineno to None when object does not have __doc__ .

bpo-46197 : Fix ensurepip environment isolation for subprocess running pip .

bpo-45924 : Fix asyncio incorrect traceback when future’s exception is raised multiple times. Patch by Kumar Aditya.

bpo-45046 : Add support of context managers in unittest : methods enterContext() and enterClassContext() of class TestCase , method enterAsyncContext() of class IsolatedAsyncioTestCase and function unittest.enterModuleContext() .

bpo-44173 : Enable fast seeking of uncompressed unencrypted zipfile.ZipExtFile

bpo-42627 : Fix incorrect parsing of Windows registry proxy settings

bpo-42047 : Add threading.get_native_id() support for DragonFly BSD. Patch by David Carlier.

bpo-14243 : The tempfile.NamedTemporaryFile function has a new optional parameter delete_on_close

bpo-41246 : Give the same callback function for when the overlapped operation is done to the functions recv , recv_into , recvfrom , sendto , send and sendfile inside IocpProactor .

bpo-39264 : Fixed collections.UserDict.get() to not call __missing__() when a value is not found. This matches the behavior of dict . Patch by Bar Harel.

bpo-38693 : importlib now uses f-strings internally instead of str.format .

bpo-38267 : Add timeout parameter to asyncio.loop.shutdown_default_executor() . The default value is None , which means the executor will be given an unlimited amount of time. When called from asyncio.Runner or asyncio.run() , the default timeout is 5 minutes.

bpo-34828 : sqlite3.Connection.iterdump() now handles databases that use AUTOINCREMENT in one or more tables.

bpo-32990 : Support reading wave files with the WAVE_FORMAT_EXTENSIBLE format in the wave module.

bpo-26253 : Allow adjustable compression level for tarfile streams in tarfile.open() .

gh-85525 : Remove extra row

gh-86404 : Deprecated tools make suspicious and rstlint.py are now removed. They have been replaced by sphinx-lint .

gh-97741 : Fix ! in c domain ref target syntax via a conf.py patch, so it works as intended to disable ref target resolution.

gh-96432 : Fraction literals now support whitespace around the forward slash, Fraction('2 / 3') .

gh-96098 : Improve discoverability of the higher level concurrent.futures module by providing clearer links from the lower level threading and multiprocessing modules.

gh-95957 : What’s New 3.11 now has instructions for how to provide compiler and linker flags for Tcl/Tk and OpenSSL on RHEL 7 and CentOS 7.

gh-95588 : Clarified the conflicting advice given in the ast documentation about ast.literal_eval() being “safe” for use on untrusted input while at the same time warning that it can crash the process. The latter statement is true and is deemed unfixable without a large amount of work unsuitable for a bugfix. So we keep the warning and no longer claim that literal_eval is safe.

gh-91207 : Fix stylesheet not working in Windows CHM htmlhelp docs and add warning that they are deprecated. Contributed by C.A.M. Gerlach.

gh-95454 : Replaced incorrectly written true/false values in documentation. Patch by Robert O’Shea

gh-95451 : Update library documentation with availability information on WebAssembly platforms wasm32-emscripten and wasm32-wasi .

gh-95415 : Use consistent syntax for platform availability. The directive now supports a content body and emits a warning when it encounters an unknown platform.

gh-94321 : Document the PEP 246 style protocol type sqlite3.PrepareProtocol .

gh-86128 : Document a limitation in ThreadPoolExecutor where its exit handler is executed before any handlers in atexit.

gh-61162 : Clarify sqlite3 behavior when How to use the connection context manager .

gh-87260 : Align sqlite3 argument specs with the actual implementation.

gh-86986 : The minimum Sphinx version required to build the documentation is now 3.2.

gh-88831 : Augmented documentation of asyncio.create_task(). Clarified the need to keep strong references to tasks and added a code snippet detailing how to do this.

gh-86438 : Clarify that -W and PYTHONWARNINGS are matched literally and case-insensitively, rather than as regular expressions, in warnings .

gh-93031 : Update tutorial introduction output to use 3.10+ SyntaxError invalid range.

gh-92240 : Added release dates for “What’s New in Python 3.X” for 3.0, 3.1, 3.2, 3.8 and 3.10

bpo-47161 : Document that pathlib.PurePath does not collapse initial double slashes because they denote UNC paths.

bpo-40838 : Document that inspect.getdoc() , inspect.getmodule() , and inspect.getsourcefile() might return None .

bpo-43689 : The Differ documentation now also mentions other whitespace characters, which make it harder to understand the diff output.

bpo-38056 : Overhaul the Error Handlers documentation in codecs .

bpo-13553 : Document tkinter.Tk args.

gh-95027 : On Windows, when the Python test suite is run with the -jN option, the ANSI code page is now used as the encoding for the stdout temporary file, rather than using UTF-8 which can lead to decoding errors. Patch by Victor Stinner.

gh-96624 : Fixed the failure of repeated runs of test.test_unittest caused by side effects in test_dotted_but_module_not_loaded .

gh-95243 : Mitigate the inherent race condition from using find_unused_port() in testSockName() by trying to find an unused port a few times before failing. Patch by Ross Burton.

gh-95573 : Lib/test/test_asyncio/test_ssl.py exposed a bug in the macOS kernel where intense concurrent load on non-blocking sockets occasionally causes errno.ENOBUFS (“No buffer space available”) to be emitted. FB11063974 filed with Apple, in the mean time as a workaround buffer size used in tests on macOS is decreased to avoid intermittent failures. Patch by Fantix King.

gh-95280 : Fix problem with test_ssl test_get_ciphers on systems that require perfect forward secrecy (PFS) ciphers.

gh-95212 : Make multiprocessing test case test_shared_memory_recreate parallel-safe.

gh-95218 : Move tests for importlib.resources into test_importlib.resources.

gh-93963 : Updated tests to use preferred location for importlib.resources ABCs.

gh-94675 : Add a regression test for re exponentional slowdown when using rjsmin.

gh-91330 : Added more tests for dataclasses to cover behavior with data descriptor-based fields.

gh-94208 : test_ssl is now checking for supported TLS version and protocols in more tests.

gh-94315 : Tests now check for DAC override capability instead of relying on os.geteuid() .

gh-54781 : Rename test_tk to test_tkinter, and rename test_ttk_guionly to test_ttk. Patch by Victor Stinner.

gh-93839 : Move Lib/ctypes/test/ to Lib/test/test_ctypes/ . Patch by Victor Stinner.

gh-93951 : In test_bdb.StateTestCase.test_skip, avoid including auxiliary importers.

gh-93957 : Provide nicer error reporting from subprocesses in test_venv.EnsurePipTest.test_with_pip.

gh-93884 : Add test cases for PyNumber_ToBase() that take a large number or a non-int object as parameter.

gh-93852 : test_asyncio, test_logging, test_socket and test_socketserver now create AF_UNIX domains in the current directory to no longer fail with OSError("AF_UNIX path too long") if the temporary directory (the TMPDIR environment variable) is too long. Patch by Victor Stinner.

gh-93353 : regrtest now checks if a test leaks temporary files or directories if run with -jN option. Patch by Victor Stinner.

gh-84461 : run_tests.py now handles cross compiling env vars correctly and pass HOSTRUNNER to regression tests.

gh-93616 : test_modulefinder now creates a temporary directory in ModuleFinderTest.setUp() instead of module scope.

gh-93575 : Fix issue with test_unicode test_raiseMemError. The test case now use test.support.calcobjsize to calculate size of PyUnicode structs. sys.getsizeof() may return different size when string has UTF-8 memory.

gh-90473 : WASI does not have a chmod(2) syscall. os.chmod() is now a dummy function on WASI. Skip all tests that depend on working os.chmod() .

gh-90473 : Skip tests on WASI that require symlinks with absolute paths.

gh-57539 : Increase calendar test coverage for calendar.LocaleTextCalendar.formatweekday() .

gh-90473 : Skip symlink tests on WASI. wasmtime uses openat2(2) with RESOLVE_BENEATH flag, which prevents symlinks with absolute paths.

gh-89858 : Fix test_embed for out-of-tree builds. Patch by Kumar Aditya.

gh-92886 : Fixing tests that fail when running with optimizations ( -O ) in test_imaplib.py .

gh-92886 : Fixing tests that fail when running with optimizations ( -O ) in test_zipimport.py

gh-92886 : Fixing tests that fail when running with optimizations ( -O ) in test_py_compile.py

gh-92886 : Fixing tests that fail when running with optimizations ( -O ) in test_sys_settrace.py .

gh-92886 : Fixing tests that fail when running with optimizations ( -O ) in _test_multiprocessing.py

gh-92670 : Skip test_shutil.TestCopy.test_copyfile_nonexistent_dir test on AIX as the test uses a trailing slash to force the OS consider the path as a directory, but on AIX the trailing slash has no effect and is considered as a file.

gh-92514 : Remove unused test.support.BasicTestRunner . Patch by Jelle Zijlstra.

bpo-47016 : Create a GitHub Actions workflow for verifying bundled pip and setuptools. Patch by Illia Volochii and Adam Turner.

gh-96761 : Fix the build process of clang compiler for _bootstrap_python if LTO optimization is applied. Patch by Matthias Görgens and Donghee Na.

gh-96883 : wasm32-emscripten builds for browsers now include concurrent.futures for asyncio and unittest.mock .

gh-85936 : CPython now uses the ThinLTO option as the default policy if the Clang compiler accepts the flag. Patch by Donghee Na.

gh-96729 : Ensure that Windows releases built with Tools\msi\buildrelease.bat are upgradable to and from official Python releases.

gh-96269 : Shared module targets now depend on new MODULE_DEPS variable, which includes EXPORTSYMS . This fixes a build order issue on unsupported AIX platform.

gh-84461 : wasm32-emscripten platform no longer builds resource module, getresuid() , getresgid() , and their setters. The APIs are stubs and not functional.

gh-95973 : Add a new --with-dsymutil configure option to link debug information in macOS. Patch by Pablo Galindo.

gh-90536 : Use the BOLT post-link optimizer to improve performance, particularly on medium-to-large applications.

gh-93744 : Remove the configure --with-cxx-main build option: it didn’t work for many years. Remove the MAINCC variable from configure and Makefile . Patch by Victor Stinner.

gh-94801 : Fix a regression in configure script that caused some header checks to ignore custom CPPFLAGS . The regression was introduced in gh-94802 .

gh-95145 : wasm32-wasi builds no longer depend on WASIX’s pthread stubs. Python now has its own stubbed pthread API.

gh-95174 : Python now detects missing dup function in WASI and works around some missing errno , select , and socket constants.

gh-95174 : Python now skips missing socket functions and methods on WASI. WASI can only create sockets from existing fd / accept and has no netdb.

gh-95085 : Platforms wasm32-unknown-emscripten and wasm32-unknown-wasi have been promoted to PEP 11 tier 3 platform support.

gh-94847 : Fixed _decimal module build issue on GCC when compiling with LTO and pydebug. Debug builds no longer force inlining of functions.

gh-94841 : Fix the possible performance regression of PyObject_Free() compiled with MSVC version 1932.

gh-94801 : configure now uses custom flags like ZLIB_CFLAGS and ZLIB_LIBS when searching for headers and libraries.

gh-94773 : deepfreeze.py now supports code object with frozensets that contain incompatible, unsortable types.

gh-94682 : Build and test with OpenSSL 1.1.1q

gh-90005 : Dependencies of readline and curses module are now detected in configure script with pkg-config . Only ncurses / ncursesw are detected automatically. The old curses library is not configured automatically. Workaround for missing termcap or tinfo library has been removed.

gh-90005 : Fix building _ctypes extension without pkg-config .

gh-90005 : _dbm module dependencies are now detected by configure.

gh-94404 : makesetup now works around an issue with sed on macOS and uses correct CFLAGS for object files that end up in a shared extension. Module CFLAGS are used before PY_STDMODULE_CFLAGS to avoid clashes with system headers.

gh-93939 : C extension modules are now built by configure and make instead of distutils and setup.py .

gh-93939 : The 2to3 , idle , and pydoc scripts are now generated and installed by Makefile instead of setup.py .

gh-94280 : Updated pegen regeneration script on Windows to find and use Python 3.9 or higher. Prior to this, pegen regeneration already required 3.9 or higher, but the script may have used lower versions of Python.

gh-93584 : Address race condition in Makefile when installing a PGO build. All test and install targets now depend on all target.

gh-93491 : configure now detects and reports PEP 11 support tiers.

gh-69093 : Fix Modules/Setup.stdlib.in rule for _sqlite3 extension.

gh-93207 : va_start() with two parameters, like va_start(args, format), is now required to build Python. va_start() is no longer called with a single parameter. Patch by Kumar Aditya.

gh-93202 : Python now always use the %zu and %zd printf formats to format a size_t or Py_ssize_t number. Building Python 3.12 requires a C11 compiler, so these printf formats are now always supported. Patch by Victor Stinner.

gh-90473 : Disable pymalloc and increase stack size on wasm32-wasi .

bpo-34449 : Drop invalid compiler switch -fPIC for HP aCC on HP-UX. Patch by Michael Osipov.

gh-98360 : Fixes multiprocessing spawning child processes on Windows from a virtual environment to ensure that child processes that also use multiprocessing to spawn more children will recognize that they are in a virtual environment.

gh-98414 : Fix py.exe launcher handling of -V: <company> / option when default preferences have been set in environment variables or configuration files.

gh-97728 : Fix possible crashes caused by the use of uninitialized variables when pass invalid arguments in os.system() on Windows and in Windows-specific modules (like winreg ).

gh-90989 : Made Python Launcher for Windows install per-user by default (unless an all users install already exists), and clarify some text in the installer.

gh-97649 : The Tools directory is no longer installed on Windows

gh-96965 : Update libffi to 3.4.3

gh-96577 : Fixes a potential buffer overrun in msilib .

gh-96559 : Fixes the Windows launcher not using the compatible interpretation of default tags found in configuration files when no tag was passed to the command.

gh-94781 : Fix pcbuild.proj to clean previous instances of output files in Python\deepfreeze and Python\frozen_modules directories on Windows. Patch by Charlie Zhao.

gh-89545 : Updates platform code getting the Windows version to use native Windows Management Instrumentation (WMI) queries to determine OS version, type, and architecture.

gh-95733 : Make certain requirements of the Windows Store package optional to allow installing on earlier updates of Windows.

gh-95656 : Enable the enable_load_extension() sqlite3 API.

gh-95587 : Fixes some issues where the Windows installer would incorrectly detect certain features of an existing install when upgrading.

gh-94399 : Restores the behaviour of Python Launcher for Windows for /usr/bin/env shebang lines, which will now search PATH for an executable matching the given command. If none is found, the usual search process is used.

gh-95445 : Fixes the unsuccessful removal of the HTML document directory when uninstalling with Windows msi.

gh-95359 : Fix Python Launcher for Windows handling of py.ini commands (it was incorrectly expecting a py_ prefix on keys) and crashes when reading per-user configuration file.

gh-95285 : Fix Python Launcher for Windows handling of command lines where it is only passed a short executable name.

gh-90844 : Allow virtual environments to correctly launch when they have spaces in the path.

gh-94772 : Fix incorrect handling of shebang lines in py.exe launcher

gh-94018 : zipfile will now remove trailing spaces from path components when extracting files on Windows.

gh-93824 : Drag and drop of files onto Python files in Windows Explorer has been enabled for Windows ARM64.

gh-43414 : os.get_terminal_size() now attempts to read the size from any provided handle, rather than only supporting file descriptors 0, 1 and 2.

gh-92817 : Ensures that py.exe will prefer an active virtual environment over default tags specified with environment variables or through a py.ini file.

gh-92984 : Explicitly disable incremental linking for non-Debug builds

gh-92841 : asyncio no longer throws RuntimeError: Event loop is closed on interpreter exit after asynchronous socket activity. Patch by Oleg Iarygin.

bpo-46907 : Update Windows installer to use SQLite 3.38.4.

gh-91061 : Accept os.PathLike for the argument to winsound.PlaySound

bpo-42658 : Support native Windows case-insensitive path comparisons by using LCMapStringEx instead of str.lower() in ntpath.normcase() . Add LCMapStringEx to the _winapi module.

bpo-38704 : Prevent installation on unsupported Windows versions.

gh-97897 : The macOS 13 SDK includes support for the mkfifoat and mknodat system calls. Using the dir_fd option with either os.mkfifo() or os.mknod() could result in a segfault if cpython is built with the macOS 13 SDK but run on an earlier version of macOS. Prevent this by adding runtime support for detection of these system calls (“weaklinking”) as is done for other newer syscalls on macOS.

gh-97527 : Fix a bug in the previous bugfix that caused IDLE to not start when run with 3.10.8, 3.12.0a1, and at least Microsoft Python 3.10.2288.0 installed without the Lib/test package. 3.11.0 was never affected.

gh-65802 : Document handling of extensions in Save As dialogs.

gh-95191 : Include prompts when saving Shell (interactive input and output).

gh-95511 : Fix the Shell context menu copy-with-prompts bug of copying an extra line when one selects whole lines.

gh-95471 : In the Edit menu, move Select All and add a new separator.

gh-95411 : Enable using IDLE’s module browser with .pyw files.

gh-89610 : Add .pyi as a recognized extension for IDLE on macOS. This allows opening stub files by double clicking on them in the Finder.

gh-68686 : Remove ptags and eptags scripts.

gh-97681 : Remove the Tools/demo/ directory which contained old demo scripts. A copy can be found in the old-demos project . Patch by Victor Stinner.

gh-97669 : Remove outdated example scripts of the Tools/scripts/ directory. A copy can be found in the old-demos project . Patch by Victor Stinner.

gh-95853 : The wasm_build.py script now pre-builds Emscripten ports, checks for broken EMSDK versions, and warns about pkg-config env vars.

gh-95853 : The new tool Tools/wasm/wasm_builder.py automates configure, compile, and test steps for building CPython on WebAssembly platforms.

gh-95731 : Fix handling of module docstrings in Tools/i18n/pygettext.py .

gh-93939 : Add script Tools/scripts/check_modules.py to check and validate builtin and shared extension modules. The script also handles Modules/Setup and will eventually replace setup.py .

gh-94538 : Fix Argument Clinic output to custom file destinations. Patch by Erlend E. Aasland.

gh-94430 : Allow parameters named module and self with custom C names in Argument Clinic. Patch by Erlend E. Aasland

gh-86087 : The Tools/scripts/parseentities.py script used to parse HTML4 entities has been removed.

gh-98393 : The PyUnicode_FSDecoder() function no longer accepts bytes-like paths, like bytearray and memoryview types: only the exact bytes type is accepted for bytes strings. Patch by Victor Stinner.

gh-91051 : Add PyType_Watch() and related APIs to allow callbacks on PyType_Modified() .

gh-95756 : Lazily create and cache co_ attributes for better performance for code getters.

gh-96512 : Configuration for the integer string conversion length limitation now lives in the PyConfig C API struct.

gh-95589 : Extensions classes that set tp_dictoffset and tp_weaklistoffset lose the support for multiple inheritance, but are now safe. Extension classes should use Py_TPFLAGS_MANAGED_DICT and Py_TPFLAGS_MANAGED_WEAKREF instead.

gh-95781 : An unrecognized format character in PyUnicode_FromFormat() and PyUnicode_FromFormatV() now sets a SystemError . In previous versions it caused all the rest of the format string to be copied as-is to the result string, and any extra arguments discarded.

gh-92678 : Restore the 3.10 behavior for multiple inheritance of C extension classes that store their dictionary at the end of the struct.

gh-92678 : Support C extensions using managed dictionaries by setting the Py_TPFLAGS_MANAGED_DICT flag.

gh-93274 : API for implementing vectorcall ( Py_TPFLAGS_HAVE_VECTORCALL , PyVectorcall_NARGS() and PyVectorcall_Call() ) was added to the limited API and stable ABI.

gh-95504 : Fix sign placement when specifying width or precision in PyUnicode_FromFormat() and PyUnicode_FromFormatV() . Patch by Philip Georgi.

gh-93012 : The Py_TPFLAGS_HAVE_VECTORCALL flag is now removed from a class when the class’s __call__() method is reassigned. This makes vectorcall safe to use with mutable types (i.e. heap types without the immutable flag). Mutable types that do not override tp_call now inherit the Py_TPFLAGS_HAVE_VECTORCALL flag.

gh-95388 : Creating immutable types with mutable bases is deprecated and is planned to be disabled in Python 3.14.

gh-92678 : Adds unstable C-API functions _PyObject_VisitManagedDict and _PyObject_ClearManagedDict to allow C extensions to allow the VM to manage their object’s dictionaries.

gh-94936 : Added PyCode_GetVarnames() , PyCode_GetCellvars() and PyCode_GetFreevars() for accessing co_varnames , co_cellvars and co_freevars respectively via the C API.

gh-94930 : Fix SystemError raised when PyArg_ParseTupleAndKeywords() is used with # in (...) but without PY_SSIZE_T_CLEAN defined.

gh-94731 : Python again uses C-style casts for most casting operations when compiled with C++. This may trigger compiler warnings, if they are enabled with e.g. -Wold-style-cast or -Wzero-as-null-pointer-constant options for g++ .

gh-93937 : The following frame functions and type are now directly available with #include <Python.h> , it’s no longer needed to add #include <frameobject.h> :

PyFrame_Check()

PyFrame_GetBack()

PyFrame_GetBuiltins()

PyFrame_GetGenerator()

PyFrame_GetGlobals()

PyFrame_GetLasti()

PyFrame_GetLocals()

PyFrame_Type

gh-91321 : Fix the compatibility of the Python C API with C++ older than C++11. Patch by Victor Stinner.

gh-91731 : Avoid defining the static_assert when compiling with C++ 11, where this is a keyword and redefining it can lead to undefined behavior. Patch by Pablo Galindo

gh-89546 : PyType_FromMetaclass() (and other PyType_From* functions) now check that offsets and the base class’s tp_basicsize fit in the new class’s tp_basicsize .

gh-93503 : Add two new public functions to the public C-API, PyEval_SetProfileAllThreads() and PyEval_SetTraceAllThreads() , that allow to set tracing and profiling functions in all running threads in addition to the calling one. Also, two analogous functions to the threading module ( threading.setprofile_all_threads() and threading.settrace_all_threads() ) that allow to do the same from Python. Patch by Pablo Galindo

gh-93442 : Add C++ overloads for _Py_CAST_impl() to handle 0/NULL. This will allow C++ extensions that pass 0 or NULL to macros using _Py_CAST() to continue to compile.

gh-93466 : Slot IDs in PyType_Spec may not be repeated. The documentation was updated to mention this. For some cases of repeated slots, PyType_FromSpec and related functions will now raise an exception.

gh-92898 : Fix C++ compiler warnings when casting function arguments to PyObject* . Patch by Serge Guelton.

gh-93103 : Deprecate global configuration variables, like Py_IgnoreEnvironmentFlag , in the documentation: the Py_InitializeFromConfig() API should be instead. Patch by Victor Stinner.

gh-77782 : Deprecate global configuration variable like Py_IgnoreEnvironmentFlag : the Py_InitializeFromConfig() API should be instead. Patch by Victor Stinner.

gh-92913 : Ensures changes to PyConfig.module_search_paths are ignored unless PyConfig.module_search_paths_set is set

gh-92781 : Avoid mixing declarations and code in the C API to fix the compiler warning: “ISO C90 forbids mixed declarations and code” [-Werror=declaration-after-statement]. Patch by Victor Stinner.

gh-92651 : Remove the token.h header file. There was never any public tokenizer C API. The token.h header file was only designed to be used by Python internals. Patch by Victor Stinner.

gh-92536 : Remove legacy Unicode APIs based on Py_UNICODE* .

gh-85858 : Remove the PyUnicode_InternImmortal() function and the SSTATE_INTERNED_IMMORTAL macro. Patch by Victor Stinner.

gh-92193 : Add new function PyFunction_SetVectorcall() to the C API which sets the vectorcall field of a given PyFunctionObject .

Warning: extensions using this API must preserve the behavior of the unaltered function!

gh-59121 : Fixed an assert that prevented PyRun_InteractiveOne from providing tracebacks when parsing from the provided FD.

bpo-45383 : The PyType_FromSpec() API will now find and use a metaclass based on the provided bases. An error will be raised if there is a metaclass conflict.

Python 3.11.0 beta 1 ¶

Release date: 2022-05-06

gh-57684 : Add the -P command line option and the PYTHONSAFEPATH environment variable to not prepend a potentially unsafe path to sys.path . Patch by Victor Stinner.

gh-89519 : Chaining classmethod descriptors (introduced in bpo-19072 ) is deprecated. It can no longer be used to wrap other descriptors such as property(). The core design of this feature was flawed, and it caused a number of downstream problems.

gh-92345 : pymain_run_python() now imports readline and rlcompleter before sys.path is extended to include the current working directory of an interactive interpreter. Non-interactive interpreters are not affected.

bpo-43857 : Improve the AttributeError message when deleting a missing attribute. Patch by Géry Ogam.

gh-92245 : Make sure that PEP 523 is respected in all cases. In 3.11a7, specialization may have prevented Python-to-Python calls respecting PEP 523.

gh-92203 : Add a closure keyword-only parameter to exec() . It can only be specified when exec-ing a code object that uses free variables. When specified, it must be a tuple, with exactly the number of cell variables referenced by the code object. closure has a default value of None , and it must be None if the code object doesn’t refer to any free variables.

gh-91173 : Disable frozen modules in debug builds. Patch by Kumar Aditya.

gh-92114 : Improve error message when subscript a type with __class_getitem__ set to None .

gh-92112 : Fix crash triggered by an evil custom mro() on a metaclass.

gh-92063 : The PRECALL_METHOD_DESCRIPTOR_FAST_WITH_KEYWORDS instruction now ensures methods are called only on objects of the correct type.

gh-92031 : Deoptimize statically allocated code objects during Py_FINALIZE() so that future _PyCode_Quicken calls always start with unquickened code.

gh-92036 : Fix a crash in subinterpreters related to the garbage collector. When a subinterpreter is deleted, untrack all objects tracked by its GC. To prevent a crash in deallocator functions expecting objects to be tracked by the GC, leak a strong reference to these objects on purpose, so they are never deleted and their deallocator functions are not called. Patch by Victor Stinner.

gh-92032 : The interpreter can now autocomplete soft keywords, as of now match , case , and _ (wildcard pattern) from PEP 634 .

gh-87999 : The warning emitted by the Python parser for a numeric literal immediately followed by keyword has been changed from deprecation warning to syntax warning.

gh-91869 : Fix an issue where specialized opcodes with extended arguments could produce incorrect tracing output or lead to assertion failures.

gh-91603 : Speed up types.UnionType instantiation. Based on patch provided by Yurii Karabas.

gh-89373 : If Python is built in debug mode, Python now ensures that deallocator functions leave the current exception unchanged. Patch by Victor Stinner.

gh-91632 : Fix a minor memory leak at exit: release the memory of the generic_alias_iterator type. Patch by Donghee Na.

gh-81548 : Octal escapes with value larger than 0o377 now produce a DeprecationWarning . In a future Python version they will be a SyntaxWarning and eventually a SyntaxError .

bpo-43950 : Use a single compact table for line starts, ends and column offsets. Reduces memory consumption for location info by half

gh-91102 : Use Argument Clinic for EncodingMap . Patch by Oleg Iarygin.

gh-91636 : Fixed a crash in a garbage-collection edge-case, in which a PyFunction_Type.tp_clear function could leave a python function object in an inconsistent state.

gh-91603 : Speed up isinstance() and issubclass() checks for types.UnionType . Patch by Yurii Karabas.

gh-91625 : Fixed a bug in which adaptive opcodes ignored any preceding EXTENDED_ARG s on specialization failure.

gh-78607 : The LLTRACE special build now looks for the name __lltrace__ defined in module globals, rather than the name __ltrace__ , which had been introduced as a typo.

gh-91576 : Speed up iteration of ascii strings by 50%. Patch by Kumar Aditya.

gh-89279 : Improve interpreter performance on Windows by inlining a few specific macros.

gh-91502 : Add a new _PyFrame_IsEntryFrame() API function, to check if a PyFrameObject is an entry frame. Patch by Pablo Galindo.

gh-91266 : Refactor the bytearray strip methods strip , lstrip and rstrip to use a common implementation.

gh-91479 : Replaced the __note__ field of BaseException (added in an earlier version of 3.11) with the final design of PEP 678 . Namely, BaseException gets an add_note() method, and its __notes__ field is created when necessary.

gh-46055 : Speed up right shift of negative integers, by removing unnecessary creation of temporaries. Original patch by Xinhang Xu, reworked by Mark Dickinson.

gh-91462 : Make the interpreter’s low-level tracing (lltrace) feature output more readable by displaying opcode names (rather than just numbers), and by displaying stack contents before each opcode.

gh-89455 : Fixed an uninitialized bool value in the traceback printing code path that was introduced by the initial bpo-45292 exception groups work.

gh-91421 : Fix a potential integer overflow in _Py_DecodeUTF8Ex.

gh-91428 : Add static const char *const _PyOpcode_OpName[256] = {...}; to opcode.h for debug builds to assist in debugging the Python interpreter. It is now more convenient to make various forms of debugging output more human-readable by including opcode names rather than just the corresponding decimal digits.

bpo-47120 : Make POP_JUMP_IF_TRUE , POP_JUMP_IF_FALSE , POP_JUMP_IF_NONE and POP_JUMP_IF_NOT_NONE virtual, mapping to new relative jump opcodes.

bpo-45317 : Add internal documentation explaining design of new (for 3.11) frame stack.

bpo-47197 : ctypes used to mishandle void return types, so that for instance a function declared like ctypes.CFUNCTYPE(None, ctypes.c_int) would be called with signature int f(int) instead of void f(int) . Wasm targets require function pointers to be called with the correct signatures so this led to crashes. The problem is now fixed.

bpo-47120 : Make opcodes JUMP_IF_TRUE_OR_POP and JUMP_IF_FALSE_OR_POP relative rather than absolute.

bpo-47177 : Replace the f_lasti member of the internal _PyInterpreterFrame structure with a prev_instr pointer, which reduces overhead in the main interpreter loop. The  f_lasti attribute of Python-layer frame objects is preserved for backward-compatibility.

bpo-46961 : Integer mod/remainder operations, including the three-argument form of pow() , now consistently return ints from the global small integer cache when applicable.

bpo-46962 : Classes and functions that unconditionally declared their docstrings ignoring the --without-doc-strings compilation flag no longer do so.

The classes affected are ctypes.UnionType , pickle.PickleBuffer , testcapi.RecursingInfinitelyError , and types.GenericAlias .

The functions affected are 24 methods in ctypes .

Patch by Oleg Iarygin.

bpo-46942 : Use Argument Clinic for the types.MethodType constructor. Patch by Oleg Iarygin.

bpo-46764 : Fix wrapping bound methods with @classmethod

bpo-43464 : Optimize set.intersection() for non-set arguments.

bpo-46721 : Optimize set.issuperset() for non-set argument.

bpo-46509 : Add type-specialized versions of the Py_DECREF() , and use them for float , int , str , bool , and None to avoid pointer-chasing at runtime where types are known at C compile time.

bpo-46045 : Do not use POSIX semaphores on NetBSD

bpo-36819 : Fix crashes in built-in encoders with error handlers that return position less or equal than the starting position of non-encodable characters.

bpo-34093 : marshal.dumps() uses FLAG_REF for all interned strings. This makes output more deterministic and helps reproducible build.

bpo-26579 : Added object.__getstate__ which provides the default implementation of the __getstate__() method.

Copying and pickling instances of subclasses of builtin types bytearray, set, frozenset, collections.OrderedDict, collections.deque, weakref.WeakSet, and datetime.tzinfo now copies and pickles instance attributes implemented as slots.

gh-87901 : Add the encoding parameter to os.popen() .

gh-90997 : Fix an issue where dis utilities may interpret populated inline cache entries as valid instructions.

gh-92332 : Deprecate typing.Text (removal of the class is currently not planned). Patch by Alex Waygood.

Deprecate nested classes in enum definitions becoming members – in 3.13 they will be normal classes; add member and nonmember functions to allow control over results now.

gh-92356 : Fixed a performance regression in ctypes function calls.

gh-90997 : Show the actual named values stored in inline caches when show_caches=True is passed to dis utilities.

gh-92301 : Prefer close_range() to iterating over procfs for file descriptor closing in subprocess for better performance.

gh-67248 : Sort the miscellaneous topics in Cmd.do_help()

gh-92210 : Port socket.__init__ to Argument Clinic. Patch by Cinder.

gh-80010 : Add support for generalized ISO 8601 parsing to datetime.datetime.fromisoformat() , datetime.date.fromisoformat() and datetime.time.fromisoformat() . Patch by Paul Ganssle.

gh-92118 : Fix a 3.11 regression in contextmanager() , which caused it to propagate exceptions with incorrect tracebacks.

gh-90887 : Adding COPYFILE_STAT , COPYFILE_ACL and COPYFILE_XATTR constants for os.fcopyfile() available in macOs.

gh-91215 : For @dataclass , add weakref_slot . The new parameter defaults to False . If true, and if slots=True , add a slot named "__weakref__" , which will allow instances to be weakref’d. Contributed by Eric V. Smith

gh-85984 : New function os.login_tty() for Unix.

gh-92128 : Add __class_getitem__() to logging.LoggerAdapter and logging.StreamHandler , allowing them to be parameterized at runtime. Patch by Alex Waygood.

gh-92049 : Forbid pickling constants re._constants.SUCCESS etc. Previously, pickling did not fail, but the result could not be unpickled.

gh-92062 : inspect.Parameter now raises ValueError if name is a keyword, in addition to the existing check that it is an identifier.

gh-87390 : Add an __unpacked__ attribute to types.GenericAlias . Patch by Jelle Zijlstra.

gh-88089 : Add support for generic typing.NamedTuple .

gh-91996 : New http.HTTPMethod enum to represent all the available HTTP request methods in a convenient way

gh-91984 : Modified test strings in test_argparse.py to not contain trailing spaces before end of line.

gh-91952 : Add encoding="locale" support to TextIOWrapper.reconfigure() .

gh-91954 : Add encoding and errors arguments to subprocess.getoutput() and subprocess.getstatusoutput() .

bpo-47029 : Always close the read end of the pipe used by multiprocessing.Queue after the last write of buffered data to the write end of the pipe to avoid BrokenPipeError at garbage collection and at multiprocessing.Queue.close() calls. Patch by Géry Ogam.

gh-91928 : Add datetime.UTC alias for datetime.timezone.utc .

Patch by Kabir Kwatra.

gh-68966 : The mailcap module is now deprecated and will be removed in Python 3.13. See PEP 594 for the rationale and the mimetypes module for an alternative. Patch by Victor Stinner.

gh-91401 : Provide a way to disable subprocess use of vfork() just in case it is ever needed and document the existing mechanism for posix_spawn() .

gh-64783 : Fix signal.NSIG value on FreeBSD to accept signal numbers greater than 32, like signal.SIGRTMIN and signal.SIGRTMAX . Patch by Victor Stinner.

gh-91910 : Add missing f prefix to f-strings in error messages from the multiprocessing and asyncio modules.

gh-91860 : Add typing.dataclass_transform() , implementing PEP 681 . Patch by Jelle Zijlstra.

gh-91832 : Add required attribute to argparse.Action repr output.

gh-91827 : In the tkinter module add method info_patchlevel() which returns the exact version of the Tcl library as a named tuple similar to sys.version_info .

gh-84461 : Add --enable-wasm-pthreads to enable pthreads support for WASM builds. Emscripten/node no longer has threading enabled by default. Include additional file systems.

gh-91821 : Fix unstable test_from_tuple test in test_decimal.py .

gh-91217 : Deprecate the xdrlib module.

gh-91217 : Deprecate the uu module.

gh-91760 : More strict rules will be applied for numerical group references and group names in regular expressions. For now, a deprecation warning is emitted for group references and group names which will be errors in future Python versions.

gh-84461 : Add provisional sys._emscripten_info named tuple with build-time and run-time information about Emscripten platform.

gh-90623 : signal.raise_signal() and os.kill() now check immediately for pending signals. Patch by Victor Stinner.

gh-91734 : Fix OSS audio support on Solaris.

gh-90633 : Include the passed value in the exception thrown by typing.assert_never() . Patch by Jelle Zijlstra.

gh-91700 : Compilation of regular expression containing a conditional expression (?(group)...) now raises an appropriate re.error if the group number refers to not defined group. Previously an internal RuntimeError was raised.

gh-91231 : Add an optional keyword shutdown_timeout parameter to the multiprocessing.BaseManager constructor. Kill the process if terminate() takes longer than the timeout. Patch by Victor Stinner.

gh-91621 : Fix typing.get_type_hints() for collections.abc.Callable . Patch by Shantanu Jain.

gh-90568 : Parsing \N escapes of Unicode Named Character Sequences in a regular expression raises now re.error instead of TypeError .

gh-91670 : Remove deprecated SO config variable in sysconfig .

gh-91217 : Deprecate the telnetlib module.

gh-91217 : Deprecate the sunau module.

gh-91217 : Deprecate the spwd module.

gh-91217 : Deprecate the sndhdr module, as well as inline needed functionality for email.mime.MIMEAudio .

gh-91616 : re module, fix fullmatch() mismatch when using Atomic Grouping or Possessive Quantifiers.

gh-91217 : Deprecate the ‘pipes’ module.

gh-91217 : Deprecate the ossaudiodev module.

bpo-47256 : re module, limit the maximum capturing group to 1,073,741,823 in 64-bit build, this increases the depth of backtracking.

gh-91217 : Deprecate the nis module.

gh-91595 : Fix the comparison of character and integer inside Tools.gdb.libpython.write_repr() . Patch by Yu Liu.

gh-74166 : Add option to raise all errors from create_connection() in an ExceptionGroup when it fails to create a connection. The default remains to raise only the last error that had occurred when multiple addresses were tried.

gh-91487 : Optimize asyncio UDP speed, over 100 times faster when transferring a large file.

gh-91575 : Update case-insensitive matching in the re module to the latest Unicode version.

gh-90622 : In concurrent.futures.process.ProcessPoolExecutor disallow the “fork” multiprocessing start method when the new max_tasks_per_child feature is used as the mix of threads+fork can hang the child processes. Default to using the safe “spawn” start method in that circumstance if no mp_context was supplied.

gh-89022 : In sqlite3 , SQLITE_MISUSE result codes are now mapped to InterfaceError instead of ProgrammingError . Also, more accurate exceptions are raised when binding parameters fail. Patch by Erlend E. Aasland.

gh-91526 : Stop calling os.device_encoding(file.fileno()) in TextIOWrapper . It was complex, never documented, and didn’t work for most cases. (Patch by Inada Naoki.)

gh-88116 : Change the frame-related functions in the inspect module to return a regular object (that is backwards compatible with the old tuple-like interface) that include the extended PEP 657 position information (end line number, column and end column). The affected functions are: inspect.getframeinfo() , inspect.getouterframes() , inspect.getinnerframes() , inspect.stack() and inspect.trace() . Patch by Pablo Galindo.

gh-69093 : Add indexing and slicing support to sqlite3.Blob . Patch by Aviv Palivoda and Erlend E. Aasland.

gh-69093 : Add context manager support to sqlite3.Blob . Patch by Aviv Palivoda and Erlend E. Aasland.

gh-91217 : Deprecate nntplib.

gh-91217 : Deprecate msilib.

gh-91404 : Improve the performance of re matching by using computed gotos (or “threaded code”) on supported platforms and removing expensive pointer indirections.

gh-91217 : Deprecate the imghdr module.

gh-91217 : Deprecate the crypt module.

gh-91276 : Make space for longer opcodes in dis output.

bpo-47000 : Make TextIOWrapper uses locale encoding when encoding="locale" is specified even in UTF-8 mode.

gh-91230 : warnings.catch_warnings() now accepts arguments for warnings.simplefilter() , providing a more concise way to locally ignore warnings or convert them to errors.

gh-91217 : Deprecate the chunk module.

Add the TCP_CONNECTION_INFO option (available on macOS) to socket .

bpo-47260 : Fix os.closerange() potentially being a no-op in a Linux seccomp sandbox.

bpo-47087 : Implement typing.Required and typing.NotRequired ( PEP 655 ). Patch by David Foster and Jelle Zijlstra.

bpo-47061 : Deprecate cgi and cgitb.

bpo-47061 : Deprecate audioop.

bpo-47000 : Add locale.getencoding() to get the current locale encoding. It is similar to locale.getpreferredencoding(False) but ignores the Python UTF-8 Mode .

bpo-42012 : Add wsgiref.types , containing WSGI-specific types for static type checking.

bpo-47227 : Suppress expression chaining for more re parsing errors.

bpo-47211 : Remove undocumented and never working function re.template() and flag re.TEMPLATE . This was later reverted in 3.11.0b2 and deprecated instead.

bpo-47135 : decimal.localcontext() now accepts context attributes via keyword arguments

bpo-43323 : Fix errors in the email module if the charset itself contains undecodable/unencodable characters.

bpo-46841 : Disassembly of quickened code.

bpo-46681 : Forward gzip.compress() compresslevel to zlib.

bpo-45100 : Add typing.get_overloads() and typing.clear_overloads() . Patch by Jelle Zijlstra.

bpo-44807 : typing.Protocol no longer silently replaces __init__() methods defined on subclasses. Patch by Adrian Garcia Badaracco.

bpo-46787 : Fix concurrent.futures.ProcessPoolExecutor exception memory leak

bpo-46720 : Add support for path-like objects to multiprocessing.set_executable() for Windows to be on a par with Unix-like systems. Patch by Géry Ogam.

bpo-46696 : Add SO_INCOMING_CPU constant to socket .

bpo-46053 : Fix OSS audio support on NetBSD.

bpo-45639 : image/avif and image/webp were added to mimetypes .

bpo-46285 : Add command-line option -p / --protocol to module http.server which specifies the HTTP version to which the server is conformant (HTTP/1.1 conformant servers can now be run from the command-line interface of module http.server ). Patch by Géry Ogam.

bpo-44791 : Accept ellipsis as the last argument of typing.Concatenate .

bpo-46547 : Remove variables leaking into pydoc.Helper class namespace.

bpo-46415 : Fix ipaddress.ip_{address,interface,network} raising TypeError instead of ValueError if given invalid tuple as address parameter.

bpo-46075 : CookieJar with DefaultCookiePolicy now can process cookies from localhost with domain=localhost explicitly specified in Set-Cookie header.

bpo-45995 : Add a “z” option to the string formatting specification that coerces negative zero floating-point values to positive zero after rounding to the format precision. Contributed by John Belmonte.

bpo-26175 : Fully implement the io.BufferedIOBase or io.TextIOBase interface for tempfile.SpooledTemporaryFile objects. This lets them work correctly with higher-level layers (like compression modules). Patch by Carey Metcalfe.

bpo-45138 : Fix a regression in the sqlite3 trace callback where bound parameters were not expanded in the passed statement string. The regression was introduced in Python 3.10 by bpo-40318 . Patch by Erlend E. Aasland.

bpo-44863 : Allow TypedDict subclasses to also include Generic as a base class in class based syntax. Thereby allowing the user to define a generic TypedDict , just like a user-defined generic but with TypedDict semantics.

bpo-44587 : Fix BooleanOptionalAction to not automatically add a default string. If a default string is desired, use a formatter to add it.

bpo-43827 : All positional-or-keyword parameters to ABCMeta.__new__ are now positional-only to avoid conflicts with keyword arguments to be passed to __init_subclass__() .

bpo-43218 : Prevent creation of a venv whose path contains the PATH separator. This could affect the usage of the activate script. Patch by Dustin Rodrigues.

bpo-38435 : Add a process_group parameter to subprocess.Popen to help move more things off of the unsafe preexec_fn parameter.

bpo-42066 : Fix cookies getting sorted in CookieJar.__iter__() which is an extra behavior and not mentioned in RFC 2965 or Netscape cookie protocol. Now the cookies in CookieJar follows the order of the Set-Cookie header. Patch by Iman Kermani.

bpo-40617 : Add create_window_function() to sqlite3.Connection for creating aggregate window functions. Patch by Erlend E. Aasland.

bpo-40676 : Convert csv to use Argument Clinic for csv.field_size_limit() , csv.get_dialect() , csv.unregister_dialect() and csv.list_dialects() .

bpo-39716 : Raise an ArgumentError when the same subparser name is added twice to an argparse.ArgumentParser . This is consistent with the (default) behavior when the same option string is added twice to an ArgumentParser .

bpo-36073 : Raise ProgrammingError instead of segfaulting on recursive usage of cursors in sqlite3 converters. Patch by Sergey Fedoseev.

bpo-34975 : Adds a start_tls() method to StreamWriter , which upgrades the connection with TLS using the given SSLContext .

bpo-22276 : Path methods glob() and rglob() return only directories if pattern ends with a pathname components separator ( / or sep ). Patch by Eisuke Kawashima.

bpo-24905 : Add blobopen() to sqlite3.Connection . sqlite3.Blob allows incremental I/O operations on blobs. Patch by Aviv Palivoda and Erlend E. Aasland.

gh-91888 : Add a new gh role to the documentation to link to GitHub issues.

gh-91783 : Document security issues concerning the use of the function shutil.unpack_archive()

gh-91547 : Remove “Undocumented modules” page.

gh-91298 : In importlib.resources.abc , refined the documentation of the Traversable Protocol, applying changes from importlib_resources 5.7.1.

bpo-44347 : Clarify the meaning of dirs_exist_ok , a kwarg of shutil.copytree() .

bpo-36329 : Remove ‘make -C Doc serve’ in favour of ‘make -C Doc htmlview’

bpo-47189 : Add a What’s New in Python 3.11 entry for the Faster CPython project. Documentation by Ken Jin and Kumar Aditya.

bpo-38668 : Update the introduction to documentation for os.path to remove warnings that became irrelevant after the implementations of PEP 383 and PEP 529 .

bpo-47115 : The documentation now lists which members of C structs are part of the Limited API/Stable ABI .

bpo-46962 : All docstrings in code snippets are now wrapped into PyDoc_STR to follow the guideline of PEP 7’s Documentation Strings paragraph . Patch by Oleg Iarygin.

bpo-26792 : Improve the docstrings of runpy.run_module() and runpy.run_path() . Original patch by Andrew Brezovsky.

gh-92169 : Use warnings_helper.import_deprecated() to import deprecated modules uniformly in tests. Patch by Hugo van Kemenade.

gh-84461 : When multiprocessing is enabled, libregrtest can now use a Python executable other than sys.executable via the --python flag.

gh-91904 : Fix initialization of PYTHONREGRTEST_UNICODE_GUARD which prevented running regression tests on non-UTF-8 locale.

gh-91752 : Added @requires_zlib to test.test_tools.test_freeze.TestFreeze.

gh-91607 : Fix test_concurrent_futures to test the correct multiprocessing start method context in several cases where the test logic mixed this up.

bpo-40280 : Threading tests are now skipped on WASM targets without pthread support.

bpo-47109 : Test for ctypes.macholib.dyld , ctypes.macholib.dylib , and ctypes.macholib.framework are brought from manual pre- unittest times to ctypes.test location and structure. Patch by Oleg Iarygin.

bpo-29890 : Add tests for ipaddress.IPv4Interface and ipaddress.IPv6Interface construction with tuple arguments. Original patch and tests by louisom.

gh-89452 : gdbm-compat is now preferred over ndbm if both are available on the system. This allows avoiding the problematic ndbm.h on macOS.

gh-91731 : Python is now built with -std=c11 compiler option, rather than -std=c99 . Patch by Victor Stinner.

bpo-47152 : Add script and make target for generating sre_constants.h .

bpo-47103 : Windows PGInstrument builds now copy a required DLL into the output directory, making it easier to run the profile stage of a PGO build.

bpo-46907 : Update Windows installer to use SQLite 3.38.3.

bpo-47239 : Fixed –list and –list-paths output for Python Launcher for Windows when used in an active virtual environment.

bpo-46907 : Update Windows installer to use SQLite 3.38.2.

bpo-46785 : Fix race condition between os.stat() and unlinking a file on Windows, by using errors codes returned by FindFirstFileW() when appropriate in win32_xstat_impl .

bpo-40859 : Update Windows build to use xz-5.2.5

bpo-46907 : Update macOS installer to SQLite 3.38.4.

gh-91583 : Fix regression in the code generated by Argument Clinic for functions with the defining_class parameter.

gh-91575 : Add script Tools/scripts/generate_re_casefix.py and the make target regen-re for generating additional data for case-insensitive matching according to the current Unicode version.

gh-91551 : Remove the ancient Pynche color editor. It has moved to https://gitlab.com/warsaw/pynche

gh-88279 : Deprecate the C functions: PySys_SetArgv() , PySys_SetArgvEx() , PySys_SetPath() . Patch by Victor Stinner.

gh-92154 : Added the PyCode_GetCode() function. This function does the equivalent of the Python code getattr(code_object, 'co_code') .

gh-92173 : Fix the closure argument to PyEval_EvalCodeEx() .

gh-91320 : Fix C++ compiler warnings about “old-style cast” ( g++ -Wold-style-cast ) in the Python C API. Use C++ reinterpret_cast<> and static_cast<> casts when the Python C API is used in C++. Patch by Victor Stinner.

gh-80527 : Mark functions as deprecated by PEP 623 : PyUnicode_AS_DATA() , PyUnicode_AS_UNICODE() , PyUnicode_GET_DATA_SIZE() , PyUnicode_GET_SIZE() . Patch by Victor Stinner.

gh-91768 : Py_REFCNT() , Py_TYPE() , Py_SIZE() and Py_IS_TYPE() functions argument type is now PyObject* , rather than const PyObject* . Patch by Victor Stinner.

gh-91020 : Add PyBytes_Type.tp_alloc to initialize PyBytesObject.ob_shash for bytes subclasses.

bpo-40421 : Add PyFrame_GetLasti C-API function to access frame object’s f_lasti attribute safely from C code.

bpo-35134 : Remove the Include/code.h header file. C extensions should only include the main <Python.h> header file. Patch by Victor Stinner.

bpo-47169 : PyOS_CheckStack() is now exported in the Stable ABI on Windows.

bpo-47169 : PyThread_get_thread_native_id() is excluded from the stable ABI on platforms where it doesn’t exist (like Solaris).

bpo-46343 : Added PyErr_GetHandledException() and PyErr_SetHandledException() as simpler alternatives to PyErr_GetExcInfo() and PyErr_SetExcInfo() .

They are included in the stable ABI.

Python 3.11.0 alpha 7 ¶

Release date: 2022-04-05

bpo-47212 : Raise IndentationError instead of SyntaxError for a bare except with no following indent. Improve SyntaxError locations for an un-parenthesized generator used as arguments. Patch by Matthieu Dartiailh.

bpo-47186 : Replace JUMP_IF_NOT_EG_MATCH by CHECK_EG_MATCH + jump.

bpo-47176 : Emscripten builds cannot handle signals in the usual way due to platform limitations. Python can now handle signals. To use, set Module.Py_EmscriptenSignalBuffer to be a single byte SharedArrayBuffer and set Py_EMSCRIPTEN_SIGNAL_HANDLING to 1. Writing a number into the SharedArrayBuffer will cause the corresponding signal to be raised into the Python thread.

bpo-47186 : Replace JUMP_IF_NOT_EXC_MATCH by CHECK_EXC_MATCH + jump.

bpo-47120 : Replace the absolute jump opcode JUMP_NO_INTERRUPT by the relative JUMP_BACKWARD_NO_INTERRUPT .

bpo-46841 : Avoid unnecessary allocations when comparing code objects.

bpo-47182 : Fix a crash when using a named unicode character like "\N{digit nine}" after the main interpreter has been initialized a second time.

bpo-47162 : WebAssembly cannot deal with bad function pointer casts (different count or types of arguments). Python can now use call trampolines to mitigate the problem. Define PY_CALL_TRAMPOLINE to enable call trampolines.

bpo-46775 : Some Windows system error codes(>= 10000) are now mapped into the correct errno and may now raise a subclass of OSError . Patch by Donghee Na.

bpo-47129 : Improve error messages in f-string syntax errors concerning empty expressions.

bpo-47117 : Fix a crash if we fail to decode characters in interactive mode if the tokenizer buffers are uninitialized. Patch by Pablo Galindo.

bpo-47127 : Speed up calls to c functions with keyword arguments by 25% with specialization. Patch by Kumar Aditya.

bpo-47120 : Replaced JUMP_ABSOLUTE by the relative jump JUMP_BACKWARD .

bpo-42197 : PyFrame_FastToLocalsWithError() and PyFrame_LocalsToFast() are no longer called during profiling nor tracing. C code can access the f_locals attribute of PyFrameObject by calling PyFrame_GetLocals() .

bpo-47070 : Improve performance of array_inplace_repeat by reducing the number of invocations of memcpy . Refactor the repeat and inplace repeat methods of array , bytes , bytearray and unicodeobject to use the common _PyBytes_Repeat .

bpo-47053 : Reduce de-optimization in the specialized BINARY_OP_INPLACE_ADD_UNICODE opcode.

bpo-47045 : Remove the f_state field from the _PyInterpreterFrame struct. Add the owner field to the _PyInterpreterFrame struct to make ownership explicit to simplify clearing and deallocing frames and generators.

bpo-46968 : Check for the existence of the “sys/auxv.h” header in faulthandler to avoid compilation problems in systems where this header doesn’t exist. Patch by Pablo Galindo

bpo-46329 : Use low bit of LOAD_GLOBAL to indicate whether to push a NULL before the global. Helps streamline the call sequence a bit.

bpo-46841 : Quicken bytecode in-place by storing it as part of the corresponding PyCodeObject .

bpo-47012 : Speed up iteration of bytes and bytearray by 30%. Patch by Kumar Aditya.

bpo-47009 : Improved the performance of list.append() and list comprehensions by optimizing for the common case, where no resize is needed. Patch by Dennis Sweeney.

bpo-47005 : Improve performance of bytearray_repeat and bytearray_irepeat by reducing the number of invocations of memcpy .

bpo-46829 : Deprecate passing a message into asyncio.Future.cancel() and asyncio.Task.cancel()

bpo-46993 : Speed up bytearray creation from list and tuple by 40%. Patch by Kumar Aditya.

bpo-39829 : Removed the __len__() call when initializing a list and moved initializing to list_extend . Patch by Jeremiah Pascual.

bpo-46944 : Speed up throwing exception in generator with METH_FASTCALL calling convention. Patch by Kumar Aditya.

bpo-46841 : Modify STORE_SUBSCR to use an inline cache entry (rather than its oparg) as an adaptive counter.

bpo-46841 : Use inline caching for PRECALL and CALL , and remove the internal machinery for managing the (now unused) non-inline caches.

bpo-46881 : Statically allocate and initialize the latin1 characters.

bpo-46838 : Improve syntax errors for incorrect function definitions. Patch by Pablo Galindo

bpo-43721 : Fix docstrings of getter , setter , and deleter to clarify that they create a new copy of the property.

bpo-43224 : Make grammar changes required for PEP 646.

bpo-47208 : Allow vendors to override CTYPES_MAX_ARGCOUNT .

bpo-23689 : re module: fix memory leak when a match is terminated by a signal or memory allocation failure. Patch by Ma Lin.

bpo-47167 : Allow overriding a future compliance check in asyncio.Task .

bpo-47151 : When subprocess tries to use vfork, it now falls back to fork if vfork returns an error. This allows use in situations where vfork isn’t allowed by the OS kernel.

bpo-47152 : Convert the re module into a package. Deprecate modules sre_compile , sre_constants and sre_parse .

bpo-4833 : Add ZipFile.mkdir()

bpo-27929 : Fix asyncio.loop.sock_connect() to only resolve names for socket.AF_INET or socket.AF_INET6 families. Resolution may not make sense for other families, like socket.AF_BLUETOOTH and socket.AF_UNIX .

bpo-14265 : Adds the fully qualified test name to unittest output

bpo-47061 : Deprecate the aifc module.

bpo-39622 : Handle Ctrl+C in asyncio programs to interrupt the main task.

bpo-47101 : hashlib.algorithms_available now lists only algorithms that are provided by activated crypto providers on OpenSSL 3.0. Legacy algorithms are not listed unless the legacy provider has been loaded into the default OSSL context.

bpo-47099 : All URLError exception messages raised in urllib.request.URLopener now contain a colon between ftp error and the rest of the message. Previously, open_ftp() missed the colon. Patch by Oleg Iarygin.

bpo-47099 : Exception chaining is changed from Exception.with_traceback() / sys.exc_info() to PEP 3134 . Patch by Oleg Iarygin.

bpo-47095 : hashlib ’s internal _blake2 module now prefers libb2 from https://www.blake2.net/ over Python’s vendored copy of blake2.

bpo-47098 : The Keccak Code Package for hashlib ’s internal _sha3 module has been replaced with tiny_sha3. The module is used as fallback when Python is built without OpenSSL.

bpo-47088 : Implement typing.LiteralString , part of PEP 675 . Patch by Jelle Zijlstra.

bpo-42885 : Optimize re.search() , re.split() , re.findall() , re.finditer() and re.sub() for regular expressions starting with \A or ^ .

bpo-23691 : Protect the re.finditer() iterator from re-entering.

bpo-47067 : Optimize calling GenericAlias objects by using PEP 590 vectorcall and by replacing PyObject_SetAttrString with PyObject_SetAttr .

bpo-28080 : Add the metadata_encoding parameter in the zipfile.ZipFile constructor and the --metadata-encoding option in the zipfile CLI to allow reading zipfiles using non-standard codecs to encode the filenames within the archive.

bpo-47000 : Make io.text_encoding() returns “utf-8” when UTF-8 mode is enabled.

bpo-42369 : Fix thread safety of zipfile._SharedFile.tell() to avoid a “zipfile.BadZipFile: Bad CRC-32 for file” exception when reading a ZipFile from multiple threads.

bpo-38256 : Fix binascii.crc32() when it is compiled to use zlib’c crc32 to work properly on inputs 4+GiB in length instead of returning the wrong result. The workaround prior to this was to always feed the function data in increments smaller than 4GiB or to just call the zlib module function.

We also have binascii.crc32() release the GIL when computing on larger inputs as zlib.crc32() and hashlib do.

This also boosts performance on Windows as it now uses the zlib crc32 implementation for binascii.crc32() for a 2-3x speedup.

That the stdlib has a crc32 API in two modules is a known historical oddity. This moves us closer to a single implementation behind them.

bpo-47066 : Global inline flags (e.g. (?i) ) can now only be used at the start of the regular expressions. Using them not at the start of expression was deprecated since Python 3.6.

bpo-39394 : A warning about inline flags not at the start of the regular expression now contains the position of the flag.

bpo-433030 : Add support of atomic grouping ( (?>...) ) and possessive quantifiers ( *+ , ++ , ?+ , {m,n}+ ) in regular expressions .

bpo-47062 : Implement asyncio.Runner context manager.

bpo-46382 : dataclass() slots=True now correctly omits slots already defined in base classes. Patch by Arie Bovenberg.

bpo-47057 : Use FASTCALL convention for FutureIter.throw()

bpo-47061 : Deprecate the various modules listed by PEP 594 :

aifc, asynchat, asyncore, audioop, cgi, cgitb, chunk, crypt, imghdr, msilib, nntplib, nis, ossaudiodev, pipes, smtpd, sndhdr, spwd, sunau, telnetlib, uu, xdrlib

bpo-34790 : Remove passing coroutine objects to asyncio.wait() .

bpo-47039 : Normalize repr() of asyncio future and task objects.

bpo-2604 : Fix bug where doctests using globals would fail when run multiple times.

bpo-45150 : Add hashlib.file_digest() helper for efficient hashing of file object.

bpo-34861 : Made cumtime the default sorting key for cProfile

bpo-45997 : Fix asyncio.Semaphore re-aquiring FIFO order.

bpo-47022 : The asynchat , asyncore and smtpd modules have been deprecated since at least Python 3.6. Their documentation and deprecation warnings and have now been updated to note they will removed in Python 3.12 ( PEP 594 ).

bpo-43253 : Fix a crash when closing transports where the underlying socket handle is already invalid on the Proactor event loop.

bpo-40280 : select.select() now passes NULL to select for each empty fdset.

bpo-47004 : Apply bugfixes from importlib_metadata 4.11.3, including bugfix for EntryPoint.extras, which was returning match objects and not the extras strings.

bpo-46998 : Allow subclassing of typing.Any . Patch by Shantanu Jain.

bpo-46995 : Deprecate missing asyncio.Task.set_name() for third-party task implementations, schedule making it mandatory in Python 3.13.

bpo-46994 : Accept explicit contextvars.Context in asyncio.create_task() and asyncio.loop.create_task() .

bpo-46981 : typing.get_args(typing.Tuple[()]) now returns () instead of ((),) .

bpo-46968 : Add os.sysconf_names['SC_MINSIGSTKSZ'] .

bpo-46985 : Upgrade pip wheel bundled with ensurepip (pip 22.0.4)

bpo-46968 : faulthandler : On Linux 5.14 and newer, dynamically determine size of signal handler stack size CPython allocates using getauxval(AT_MINSIGSTKSZ) . This changes allows for Python extension’s request to Linux kernel to use AMX_TILE instruction set on Sapphire Rapids Xeon processor to succeed, unblocking use of the ISA in frameworks.

bpo-46917 : The math.nan value is now always available. Patch by Victor Stinner.

bpo-46955 : Expose asyncio.base_events.Server as asyncio.Server . Patch by Stefan Zabka.

bpo-23325 : The signal module no longer assumes that SIG_IGN and SIG_DFL are small int singletons.

bpo-46932 : Update bundled libexpat to 2.4.7

bpo-46933 : The pwd module is now optional. os.path.expanduser() returns the path when the pwd module is not available.

bpo-40059 : PEP 680 , the tomllib module. Adds support for parsing TOML.

bpo-464471 : asyncio.timeout() and asyncio.timeout_at() context managers added. Patch by Tin Tvrtković and Andrew Svetlov.

bpo-46805 : Added raw datagram socket functions for asyncio: sock_sendto() , sock_recvfrom() and sock_recvfrom_into() .

bpo-46644 : No longer require valid typeforms to be callable. This allows typing.Annotated to wrap typing.ParamSpecArgs and dataclasses.InitVar . Patch by Gregory Beauregard.

bpo-46581 : Brings ParamSpec propagation for GenericAlias in line with Concatenate (and others).

bpo-45413 : Define posix_venv and nt_venv sysconfig installation schemes to be used for bootstrapping new virtual environments. Add venv sysconfig installation scheme to get the appropriate one of the above. The schemes are identical to the pre-existing posix_prefix and nt install schemes. The venv module now uses the venv scheme to create new virtual environments instead of hardcoding the paths depending only on the platform. Downstream Python distributors customizing the posix_prefix or nt install scheme in a way that is not compatible with the install scheme used in virtual environments are encouraged not to customize the venv schemes. When Python itself runs in a virtual environment, sysconfig.get_default_scheme() and sysconfig.get_preferred_scheme() with key="prefix" returns venv .

bpo-43224 : Implement support for PEP 646 in typing.py.

bpo-43224 : Allow unpacking types.GenericAlias objects, e.g. *tuple[int, str] .

bpo-46557 : Warnings captured by the logging module are now logged without a format string to prevent systems that group logs by the msg argument from grouping captured warnings together.

bpo-41370 : typing.get_type_hints() now supports evaluating strings as forward references in PEP 585 generic aliases .

bpo-46607 : Add DeprecationWarning to LegacyInterpolation , deprecated in the docstring since Python 3.2. Will be removed in Python 3.13. Use BasicInterpolation or ExtendedInterpolation instead.

bpo-26120 : pydoc now excludes __future__ imports from the module’s data items.

bpo-46480 : Add typing.assert_type() . Patch by Jelle Zijlstra.

bpo-46421 : Fix a unittest issue where if the command was invoked as python -m unittest and the filename(s) began with a dot (.), a ValueError is returned.

bpo-46245 : Add optional parameter dir_fd in shutil.rmtree() .

bpo-22859 : unittest.TestProgram.usageExit() is marked as deprecated, to be removed in Python 3.13.

bpo-46170 : Improve the error message when you try to subclass an instance of typing.NewType .

bpo-40296 : Fix supporting generic aliases in pydoc .

bpo-20392 : Fix inconsistency with uppercase file extensions in MimeTypes.guess_type() . Patch by Kumar Aditya.

bpo-46030 : Add LOCAL_CREDS , LOCAL_CREDS_PERSISTENT and SCM_CREDS2 FreeBSD constants to the socket module.

bpo-44439 : Fix .write() method of a member file in ZipFile , when the input data is an object that supports the buffer protocol, the file length may be wrong.

bpo-45171 : Fix handling of the stacklevel argument to logging functions in the logging module so that it is consistent across all logging functions and, as advertised, similar to the stacklevel argument used in warn() .

bpo-24959 : Fix bug where unittest sometimes drops frames from tracebacks of exceptions raised in tests.

bpo-44859 : Raise more accurate and PEP 249 compatible exceptions in sqlite3 .

Raise InterfaceError instead of ProgrammingError for SQLITE_MISUSE errors.

Don’t overwrite BufferError with ValueError when conversion to BLOB fails.

Raise ProgrammingError instead of Warning if user tries to execute() more than one SQL statement.

Raise ProgrammingError instead of ValueError if an SQL query contains null characters.

bpo-44493 : Add missing terminated NUL in sockaddr_un’s length

This was potentially observable when using non-abstract AF_UNIX datagram sockets to processes written in another programming language.

bpo-41930 : Add serialize() and deserialize() support to sqlite3 . Patch by Erlend E. Aasland.

bpo-33178 : Added ctypes.BigEndianUnion and ctypes.LittleEndianUnion classes, as originally documented in the library docs but not yet implemented.

bpo-43352 : Add an Barrier object in synchronization primitives of asyncio Lib in order to be consistent with Barrier from threading and multiprocessing libs*

bpo-35859 : re module, fix a few bugs about capturing group. In rare cases, capturing group gets an incorrect string. Patch by Ma Lin.

bpo-45099 : Document internal asyncio API.

bpo-47126 : Update PEP URLs to PEP 676 ’s new canonical form.

bpo-47040 : Clarified the old Python versions compatibility note of binascii.crc32() / zlib.adler32() / zlib.crc32() functions.

bpo-46033 : Clarify for statement execution in its doc.

bpo-45790 : Adjust inaccurate phrasing in Defining Extension Types: Tutorial about the ob_base field and the macros used to access its contents.

bpo-42340 : Document that in some circumstances KeyboardInterrupt may cause the code to enter an inconsistent state. Provided a sample workaround to avoid it if needed.

bpo-41233 : Link the errnos referenced in Doc/library/exceptions.rst to their respective section in Doc/library/errno.rst , and vice versa. Previously this was only done for EINTR and InterruptedError. Patch by Yan “yyyyyyyan” Orestes.

bpo-47205 : Skip test for sched_getaffinity() and sched_setaffinity() error case on FreeBSD.

bpo-46126 : Restore ‘descriptions’ when running tests internally.

bpo-47104 : Rewrite asyncio.to_thread() tests to use unittest.IsolatedAsyncioTestCase .

bpo-40280 : The test suite is now passing on the Emscripten platform. All fork, socket, and subprocess-based tests are skipped.

bpo-47037 : Skip strftime("%4Y") feature test on Windows. It can cause an assertion error in debug builds.

bpo-46587 : Skip tests if platform’s strftime does not support non-portable glibc extensions.

bpo-47015 : A test case for os.sendfile() is converted from deprecated asyncore (see PEP 594 ) to asyncio . Patch by Oleg Iarygin.

bpo-40280 : Add configure option --enable-wasm-dynamic-linking to enable dlopen and MAIN_MODULE / SIDE_MODULE on wasm32-emscripten .

bpo-46023 : makesetup now detects and skips all duplicated module definitions. The first entry wins.

bpo-40280 : Add SOABI wasm32-emscripten for Emscripten and wasm32-wasi for WASI on 32bit WASM as well as wasm64 counter parts.

bpo-47032 : Ensure Windows install builds fail correctly with a non-zero exit code when part of the build fails.

bpo-47024 : Update OpenSSL to 1.1.1n for macOS installers and all Windows builds.

bpo-46996 : The tkinter package now requires Tcl/Tk version 8.5.12 or newer.

bpo-46973 : Add regen-configure make target to regenerate configure script with Christian’s container image quay.io/tiran/cpython_autoconf:269 .

bpo-46917 : Building Python now requires support of IEEE 754 floating-point numbers. Patch by Victor Stinner.

bpo-45774 : configure now verifies that all SQLite C APIs needed for the sqlite3 extension module are found.

bpo-47194 : Update zlib to v1.2.12 to resolve CVE-2018-25032 .

bpo-47171 : Enables installing the py.exe launcher on Windows ARM64.

bpo-46566 : Upgraded Python Launcher for Windows to support a new -V:company/tag argument for full PEP 514 support and to detect ARM64 installs. The -64 suffix on arguments is deprecated, but still selects any non-32-bit install. Setting PYLAUNCHER_ALLOW_INSTALL and specifying a version that is not installed will attempt to install the requested version from the Microsoft Store.

bpo-47086 : The installer for Windows now includes documentation as loose HTML files rather than a single compiled .chm file.

bpo-46907 : Update Windows installer to use SQLite 3.38.1.

bpo-44549 : Update bzip2 to 1.0.8 in Windows builds to mitigate CVE-2016-3189 and CVE-2019-12900 .

bpo-46948 : Prevent CVE-2022-26488 by ensuring the Add to PATH option in the Windows installer uses the correct path when being repaired.

bpo-46890 : Fix a regression in the setting of sys._base_executable in framework builds, and thereby fix a regression in venv virtual environments with such builds.

bpo-46907 : Update macOS installer to SQLite 3.38.1.

bpo-40280 : Replace Emscripten’s limited shell with Katie Bell’s browser-ui REPL from python-wasm project.

bpo-40421 : Add PyFrame_GetBuiltins , PyFrame_GetGenerator and PyFrame_GetGlobals C-API functions to access frame object attributes safely from C code.

bpo-46850 : Move the private _PyFrameEvalFunction type, and private _PyInterpreterState_GetEvalFrameFunc() and _PyInterpreterState_SetEvalFrameFunc() functions to the internal C API. The _PyFrameEvalFunction callback function type now uses the _PyInterpreterFrame type which is part of the internal C API. Patch by Victor Stinner.

bpo-46850 : Move the private undocumented _PyEval_EvalFrameDefault() function to the internal C API. The function now uses the _PyInterpreterFrame type which is part of the internal C API. Patch by Victor Stinner.

bpo-46850 : Remove the private undocumented function _PyEval_CallTracing() from the C API. Call the public sys.call_tracing() function instead. Patch by Victor Stinner.

bpo-46850 : Remove the private undocumented function _PyEval_GetCoroutineOriginTrackingDepth() from the C API. Call the public sys.get_coroutine_origin_tracking_depth() function instead. Patch by Victor Stinner.

bpo-46850 : Remove the following private undocumented functions from the C API:

_PyEval_GetAsyncGenFirstiter()

_PyEval_GetAsyncGenFinalizer()

_PyEval_SetAsyncGenFirstiter()

_PyEval_SetAsyncGenFinalizer()

Call the public sys.get_asyncgen_hooks() and sys.set_asyncgen_hooks() functions instead. Patch by Victor Stinner.

bpo-46987 : Remove private functions _PySys_GetObjectId() and _PySys_SetObjectId() . Patch by Donghee Na.

bpo-46906 : Add new functions to pack and unpack C double (serialize and deserialize): PyFloat_Pack2() , PyFloat_Pack4() , PyFloat_Pack8() , PyFloat_Unpack2() , PyFloat_Unpack4() and PyFloat_Unpack8() . Patch by Victor Stinner.

Python 3.11.0 alpha 6 ¶

Release date: 2022-03-07

bpo-46940 : Avoid overriding AttributeError metadata information for nested attribute access calls. Patch by Pablo Galindo.

bpo-46927 : Include the type’s name in the error message for subscripting non-generic types.

bpo-46921 : Support vectorcall for super() . Patch by Ken Jin.

bpo-46841 : Fix incorrect handling of inline cache entries when specializing BINARY_OP .

bpo-46841 : Use an oparg to simplify the construction of helpful error messages in GET_AWAITABLE .

bpo-46903 : Make sure that str subclasses can be used as attribute names for instances with virtual dictionaries. Fixes regression in 3.11alpha

bpo-46841 : Add more detailed specialization failure stats for COMPARE_OP followed by EXTENDED_ARG .

bpo-46891 : Fix bug introduced during 3.11alpha where subclasses of types.ModuleType with __slots__ were not initialized correctly, resulting in an interpreter crash.

bpo-46841 : Use inline caching for LOAD_ATTR , LOAD_METHOD , and STORE_ATTR .

bpo-46841 : Use inline cache for BINARY_SUBSCR .

bpo-46841 : Use inline caching for COMPARE_OP .

bpo-46864 : Deprecate PyBytesObject.ob_shash . It will be removed in Python 3.13.

bpo-46841 : Use inline caching for UNPACK_SEQUENCE .

bpo-46845 : Reduces dict size by removing hash value from hash table when all inserted keys are Unicode. For example, sys.getsizeof(dict.fromkeys("abcdefg")) becomes 272 bytes from 352 bytes on 64bit platform.

bpo-46841 : Use inline cache for LOAD_GLOBAL .

bpo-46852 : Rename the private undocumented float.__set_format__() method to float.__setformat__() to fix a typo introduced in Python 3.7. The method is only used by test_float. Patch by Victor Stinner.

bpo-46852 : Remove the undocumented private float.__set_format__() method, previously known as float.__setformat__() in Python 3.7. Its docstring said: “You probably don’t want to use this function. It exists mainly to be used in Python’s test suite.” Patch by Victor Stinner.

bpo-40116 : Fix regression that dict.update(other) may don’t respect iterate order of other when other is key sharing dict.

bpo-46712 : Share global string identifiers in deep-frozen modules.

bpo-46430 : Fix memory leak in interned strings of deep-frozen modules.

bpo-46841 : Store BINARY_OP caches inline using a new CACHE instruction.

bpo-45107 : Specialize LOAD_METHOD for instances with a dict.

bpo-44337 : Reduce the memory usage of specialized LOAD_ATTR and STORE_ATTR instructions.

bpo-46729 : Add number of sub-exceptions to BaseException.__str__() .

bpo-45885 : Don’t un-adapt COMPARE_OP when collecting specialization stats.

bpo-46329 : Fix specialization stats gathering for PRECALL instructions.

bpo-46794 : Bump up the libexpat version into 2.4.6

bpo-46823 : Implement a specialized combined opcode LOAD_FAST__LOAD_ATTR_INSTANCE_VALUE . Patch by Dennis Sweeney.

bpo-46820 : Fix parsing a numeric literal immediately (without spaces) followed by “not in” keywords, like in 1not in x . Now the parser only emits a warning, not a syntax error.

bpo-46329 : Move KW_NAMES before PRECALL instruction in call sequence. Change operand of CALL to match PRECALL for easier specialization.

bpo-46808 : Remove the NEXT_BLOCK macro from compile.c, and make the compiler automatically generate implicit blocks when they are needed.

bpo-46329 : Add PUSH_NULL instruction. This is used as a prefix when evaluating a callable, so that the stack has the same shape for methods and other calls. PRECALL_FUNCTION and PRECALL_METHOD are merged into a single PRECALL instruction.

There is no change in semantics.

bpo-46762 : Fix an assert failure in debug builds when a ‘<’, ‘>’, or ‘=’ is the last character in an f-string that’s missing a closing right brace.

bpo-46730 : Message of AttributeError caused by getting, setting or deleting a property without the corresponding function now mentions that the attribute is in fact a property and also specifies type of the class that it belongs to.

bpo-46724 : Make sure that all backwards jumps use the JUMP_ABSOLUTE instruction, rather than JUMP_FORWARD with an argument of (2**32)+offset .

bpo-46732 : Correct the docstring for the __bool__() method. Patch by Jelle Zijlstra.

bpo-46072 : Add more detailed specialization failure statistics for BINARY_OP .

bpo-46707 : Avoid potential exponential backtracking when producing some syntax errors involving lots of brackets. Patch by Pablo Galindo.

bpo-46323 : ctypes now allocates memory on the stack instead of on the heap to pass arguments while calling a Python callback function. Patch by Donghee Na.

bpo-45923 : Add a quickened form of RESUME that skips quickening checks.

bpo-46702 : Specialize UNPACK_SEQUENCE for tuple and list unpackings.

bpo-46072 : Opcode pair stats are now gathered with --enable-pystats . Defining DYNAMIC_EXECUTION_PROFILE or DXPAIRS no longer has any effect.

bpo-46675 : Allow more than 16 items in a split dict before it is combined. The limit is now 254.

bpo-40479 : Add a missing call to va_end() in Modules/_hashopenssl.c .

bpo-46323 : Use PyObject_Vectorcall() while calling ctypes callback function. Patch by Donghee Na.

bpo-46615 : When iterating over sets internally in setobject.c , acquire strong references to the resulting items from the set. This prevents crashes in corner-cases of various set operations where the set gets mutated.

bpo-45828 : The bytecode compiler now attempts to apply runtime stack manipulations at compile-time (whenever it is feasible to do so).

bpo-30496 : Fixed a minor portability issue in the implementation of PyLong_FromLong() , and added a fast path for single-digit integers to PyLong_FromLongLong() .

bpo-25707 : Fixed a file leak in xml.etree.ElementTree.iterparse() when the iterator is not exhausted. Patch by Jacob Walls.

bpo-46877 : Export unittest.doModuleCleanups() in unittest . Patch by Kumar Aditya.

bpo-46848 : For performance, use the optimized string-searching implementations from find() and rfind() for find() and rfind() .

bpo-46736 : SimpleHTTPRequestHandler now uses HTML5 grammar. Patch by Donghee Na.

bpo-44886 : Inherit asyncio proactor datagram transport from asyncio.DatagramTransport .

bpo-46827 : Support UDP sockets in asyncio.loop.sock_connect() for selector-based event loops. Patch by Thomas Grainger.

bpo-46811 : Make test suite support Expat >=2.4.5

bpo-46252 : Raise TypeError if ssl.SSLSocket is passed to transport-based APIs.

bpo-46784 : Fix libexpat symbols collisions with user dynamically loaded or statically linked libexpat in embedded Python.

bpo-46786 : The HTML serialisation in xml.etree.ElementTree now writes embed , source , track and wbr as empty tags, as defined in HTML 5.

bpo-39327 : shutil.rmtree() can now work with VirtualBox shared folders when running from the guest operating-system.

bpo-45390 : Propagate asyncio.CancelledError message from inner task to outer awaiter.

bpo-46756 : Fix a bug in urllib.request.HTTPPasswordMgr.find_user_password() and urllib.request.HTTPPasswordMgrWithPriorAuth.is_authenticated() which allowed to bypass authorization. For example, access to URI example.org/foobar was allowed if the user was authorized for URI example.org/foo .

bpo-46737 : random.gauss() and random.normalvariate() now have default arguments.

bpo-46752 : Add task groups to asyncio (structured concurrency, inspired by Trio’s nurseries). This also introduces a change to task cancellation, where a cancelled task can’t be cancelled again until it calls .uncancel().

bpo-46724 : Fix dis behavior on negative jump offsets.

bpo-46333 : The __repr__() method of typing.ForwardRef now includes the module parameter of typing.ForwardRef when it is set.

bpo-46643 : In typing.get_type_hints() , support evaluating stringified ParamSpecArgs and ParamSpecKwargs annotations. Patch by Gregory Beauregard.

bpo-45863 : When the tarfile module creates a pax format archive, it will put an integer representation of timestamps in the ustar header (if possible) for the benefit of older unarchivers, in addition to the existing full-precision timestamps in the pax extended header.

bpo-46066 : Deprecate kwargs-based syntax for typing.TypedDict definitions. It had confusing semantics when specifying totality, and was largely unused. Patch by Jingchen Ye.

bpo-46676 : Make typing.ParamSpec args and kwargs equal to themselves. Patch by Gregory Beauregard.

bpo-46323 : ctypes.CFUNCTYPE() and ctypes.WINFUNCTYPE() now fail to create the type if its _argtypes_ member contains too many arguments. Previously, the error was only raised when calling a function. Patch by Victor Stinner.

bpo-46672 : Fix NameError in asyncio.gather() when initial type check fails.

bpo-46659 : The calendar.LocaleTextCalendar and calendar.LocaleHTMLCalendar classes now use locale.getlocale() , instead of using locale.getdefaultlocale() , if no locale is specified. Patch by Victor Stinner.

bpo-46659 : The locale.getdefaultlocale() function is deprecated and will be removed in Python 3.13. Use locale.setlocale() , locale.getpreferredencoding(False) and locale.getlocale() functions instead. Patch by Victor Stinner.

bpo-46655 : In typing.get_type_hints() , support evaluating bare stringified TypeAlias annotations. Patch by Gregory Beauregard.

bpo-45948 : Fixed a discrepancy in the C implementation of the xml.etree.ElementTree module. Now, instantiating an xml.etree.ElementTree.XMLParser with a target=None keyword provides a default xml.etree.ElementTree.TreeBuilder target as the Python implementation does.

bpo-46626 : Expose Linux’s IP_BIND_ADDRESS_NO_PORT option in socket .

bpo-46521 : Fix a bug in the codeop module that was incorrectly identifying invalid code involving string quotes as valid code.

bpo-46571 : Improve typing.no_type_check() .

Now it does not modify external classes and functions. We also now correctly mark classmethods as not to be type checked.

bpo-46400 : expat: Update libexpat from 2.4.1 to 2.4.4

bpo-46556 : Deprecate undocumented support for using a pathlib.Path object as a context manager.

bpo-46534 : Implement PEP 673 typing.Self . Patch by James Hilton-Balfe.

bpo-46522 : Make various module __getattr__ AttributeErrors more closely match a typical AttributeError

bpo-46475 : Add typing.Never and typing.assert_never() . Patch by Jelle Zijlstra.

bpo-46333 : The __eq__() and __hash__() methods of typing.ForwardRef now honor the module parameter of typing.ForwardRef . Forward references from different modules are now differentiated.

bpo-46246 : Add missing __slots__ to importlib.metadata.DeprecatedList . Patch by Arie Bovenberg.

bpo-46232 : The ssl module now handles certificates with bit strings in DN correctly.

bpo-46195 : typing.get_type_hints() no longer adds Optional to parameters with None as a default. This aligns to changes to PEP 484 in https://github.com/python/peps/pull/689

bpo-31369 : Add RegexFlag to re.__all__ and documented it. Add NOFLAG to indicate no flags being set.

bpo-45898 : ctypes no longer defines ffi_type_* symbols in cfield.c . The symbols have been provided by libffi for over a decade.

bpo-44953 : Calling operator.itemgetter objects and operator.attrgetter objects is now faster due to use of the vectorcall calling convention.

bpo-44289 : Fix an issue with is_tarfile() method when using fileobj argument: position in the fileobj was advanced forward which made it unreadable with tarfile.TarFile.open() .

bpo-44011 : Reimplement SSL/TLS support in asyncio, borrow the implementation from uvloop library.

bpo-41086 : Make the configparser.ConfigParser constructor raise TypeError if the interpolation parameter is not of type configparser.Interpolation

bpo-29418 : Implement inspect.ismethodwrapper() and fix inspect.isroutine() for cases where methodwrapper is given. Patch by Hakan Çelik.

bpo-14156 : argparse.FileType now supports an argument of ‘-’ in binary mode, returning the .buffer attribute of sys.stdin/sys.stdout as appropriate. Modes including ‘x’ and ‘a’ are treated equivalently to ‘w’ when argument is ‘-’. Patch contributed by Josh Rosenberg

bpo-42238 : Doc/tools/rstlint.py has moved to its own repository and is now packaged on PyPI as sphinx-lint .

bpo-46913 : Fix test_faulthandler.test_sigfpe() if Python is built with undefined behavior sanitizer (UBSAN): disable UBSAN on the faulthandler_sigfpe() function. Patch by Victor Stinner.

bpo-46760 : Remove bytecode offsets from expected values in test.test_dis module. Reduces the obstacles to modifying the VM or compiler.

bpo-46708 : Prevent default asyncio event loop policy modification warning after test_asyncio execution.

bpo-46678 : The function make_legacy_pyc in Lib/test/support/import_helper.py no longer fails when PYTHONPYCACHEPREFIX is set to a directory on a different device from where tempfiles are stored.

bpo-46623 : Skip test_pair() and test_speech128() of test_zlib on s390x since they fail if zlib uses the s390x hardware accelerator. Patch by Victor Stinner.

bpo-46860 : Respect --with-suffix when building on case-insensitive file systems.

bpo-46656 : Building Python now requires a C11 compiler. Optional C11 features are not required. Patch by Victor Stinner.

bpo-46656 : Building Python now requires support for floating-point Not-a-Number (NaN): remove the Py_NO_NAN macro. Patch by Victor Stinner.

bpo-46640 : Building Python now requires a C99 <math.h> header file providing a NAN constant, or the __builtin_nan() built-in function. Patch by Victor Stinner.

bpo-46608 : Exclude marshalled-frozen data if deep-freezing to save 300 KB disk space. This includes adding a new is_package field to _frozen . Patch by Kumar Aditya.

bpo-40280 : Fix wasm32-emscripten test failures and platform issues. - Disable syscalls that are not supported or don’t work, e.g. wait, getrusage, prlimit, mkfifo, mknod, setres[gu]id, setgroups. - Use fd_count to count open fds. - Add more checks for subprocess and fork. - Add workarounds for missing _multiprocessing and failing socket.accept(). - Enable bzip2. - Disable large file support. - Disable signal.alarm.

bpo-46430 : Intern strings in deep-frozen modules. Patch by Kumar Aditya.

bpo-46744 : The default all users install directory for ARM64 is now under the native Program Files folder, rather than Program Files (Arm) which is intended for ARM (32-bit) files.

bpo-46567 : Adds Tcl and Tk support for Windows ARM64. This also adds IDLE to the installation.

bpo-46638 : Ensures registry virtualization is consistently disabled. For 3.10 and earlier, it remains enabled (some registry writes are protected), while for 3.11 and later it is disabled (registry modifications affect all applications).

bpo-46630 : Make query dialogs on Windows start with a cursor in the entry box.

bpo-45447 : Apply IDLE syntax highlighting to .pyi files. Patch by Alex Waygood and Terry Jan Reedy.

bpo-46748 : Python’s public headers no longer import <stdbool.h> , leaving code that embeds/extends Python free to define bool , true and false .

bpo-46836 : Move the PyFrameObject type definition ( struct _frame ) to the internal C API pycore_frame.h header file. Patch by Victor Stinner.

bpo-45459 : Rename Include/buffer.h header file to Include/pybuffer.h to avoid conflicts with projects having an existing buffer.h header file. Patch by Victor Stinner.

bpo-45412 : Remove the HAVE_PY_SET_53BIT_PRECISION macro (moved to the internal C API). Patch by Victor Stinner.

bpo-46613 : Added function PyType_GetModuleByDef() , which allows access to module state when a method’s defining class is not available.

Python 3.11.0 alpha 5 ¶

Release date: 2022-02-03

bpo-45773 : Remove two invalid “peephole” optimizations from the bytecode compiler.

bpo-46564 : Do not create frame objects when creating super object. Patch by Kumar Aditya.

bpo-45885 : Added more fined-grained specialization failure stats regarding the COMPARE_OP bytecode.

bpo-44977 : The delegation of int() to __trunc__() is now deprecated. Calling int(a) when type(a) implements __trunc__() but not __int__() or __index__() now raises a DeprecationWarning .

bpo-46458 : Reorder code emitted by the compiler for a try - except block so that the else block’s code immediately follows the try body (without a jump). This is more optimal for the happy path.

bpo-46527 : Allow passing iterable as a keyword argument to enumerate() again. Patch by Jelle Zijlstra.

bpo-46528 : Replace several stack manipulation instructions ( DUP_TOP , DUP_TOP_TWO , ROT_TWO , ROT_THREE , ROT_FOUR , and ROT_N ) with new COPY and SWAP instructions.

bpo-46329 : Use two or three bytecodes to implement most calls.

Calls without named arguments are implemented as a sequence of two instructions: PRECALL; CALL . Calls with named arguments are implemented as a sequence of three instructions: PRECALL; KW_NAMES; CALL . There are two different PRECALL instructions: PRECALL_FUNTION and PRECALL_METHOD . The latter pairs with LOAD_METHOD .

This partition into pre-call and call allows better specialization, and thus better performance ultimately.

bpo-46503 : Fix an assert when parsing some invalid N escape sequences in f-strings.

bpo-46431 : Improve error message on invalid calls to BaseExceptionGroup.__new__() .

bpo-46476 : Fix memory leak in code objects generated by deepfreeze. Patch by Kumar Aditya.

bpo-46481 : Speed up calls to weakref.ref.__call__() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-46417 : Fix a race condition on setting a type __bases__ attribute: the internal function add_subclass() now gets the PyTypeObject.tp_subclasses member after calling PyWeakref_NewRef() which can trigger a garbage collection which can indirectly modify PyTypeObject.tp_subclasses . Patch by Victor Stinner.

bpo-46417 : python -X showrefcount now shows the total reference count after clearing and destroyed the main Python interpreter. Previously, it was shown before. Patch by Victor Stinner.

bpo-43683 : Add ASYNC_GEN_WRAP opcode to wrap the value to be yielded in async generators. Removes the need to special case async generators in the YIELD_VALUE instruction.

bpo-46407 : Optimize some modulo operations in Objects/longobject.c . Patch by Jeremiah Vivian.

bpo-46409 : Add new RETURN_GENERATOR bytecode to make generators. Simplifies calling Python functions in the VM, as they no longer any need to special case generator functions.

Also add JUMP_NO_INTERRUPT bytecode that acts like JUMP_ABSOLUTE , but does not check for interrupts.

bpo-46406 : The integer division // implementation has been optimized to better let the compiler understand its constraints. It can be 20% faster on the amd64 platform when dividing an int by a value smaller than 2**30 .

bpo-46383 : Fix invalid signature of _zoneinfo ’s module_free function to resolve a crash on wasm32-emscripten platform.

bpo-46361 : Ensure that “small” integers created by int.from_bytes() and decimal.Decimal are properly cached.

bpo-46161 : Fix the class building error when the arguments are constants and CALL_FUNCTION_EX is used.

bpo-46028 : Fixes calculation of sys._base_executable when inside a virtual environment that uses symlinks with different binary names than the base environment provides.

bpo-46091 : Correctly calculate indentation levels for lines with whitespace character that are ended by line continuation characters. Patch by Pablo Galindo

bpo-30512 : Add CAN Socket support for NetBSD.

bpo-44024 : Improve the TypeError message for non-string second arguments passed to the built-in functions getattr() and hasattr() . Patch by Géry Ogam.

bpo-46624 : Restore support for non-integer arguments of random.randrange() and random.randint() .

bpo-46591 : Make the IDLE doc URL on the About IDLE dialog clickable.

bpo-46565 : Remove loop variables that are leaking into modules’ namespaces.

bpo-46553 : In typing.get_type_hints() , support evaluating bare stringified ClassVar annotations. Patch by Gregory Beauregard.

bpo-46544 : Don’t leak x & uspace intermediate vars in textwrap.TextWrapper .

bpo-46487 : Add the get_write_buffer_limits method to asyncio.transports.WriteTransport and to the SSL transport.

bpo-45173 : Note the configparser deprecations will be removed in Python 3.12.

bpo-45162 : The deprecated unittest APIs removed in 3.11a1 have been temporarily restored to be removed in 3.12 while cleanups in external projects go in.

bpo-46539 : In typing.get_type_hints() , support evaluating stringified ClassVar and Final annotations inside Annotated . Patch by Gregory Beauregard.

bpo-46510 : Add missing test for types.TracebackType and types.FrameType . Calculate them directly from the caught exception without calling sys.exc_info() .

bpo-46491 : Allow typing.Annotated to wrap typing.Final and typing.ClassVar . Patch by Gregory Beauregard.

bpo-46483 : Remove __class_getitem__() from pathlib.PurePath as this class was not supposed to be generic.

bpo-46436 : Fix command-line option -d / --directory in module http.server which is ignored when combined with command-line option --cgi . Patch by Géry Ogam.

bpo-41403 : Make mock.patch() raise a TypeError with a relevant error message on invalid arg. Previously it allowed a cryptic AttributeError to escape.

bpo-46474 : In importlib.metadata.EntryPoint.pattern , avoid potential REDoS by limiting ambiguity in consecutive whitespace.

bpo-46474 : Removed private method from importlib.metadata.Path . Sync with importlib_metadata 4.10.0.

bpo-46470 : Remove unused branch from typing._remove_dups_flatten

bpo-46469 : asyncio generic classes now return types.GenericAlias in __class_getitem__ instead of the same class.

bpo-41906 : Support passing filter instances in the filters values of handlers and loggers in the dictionary passed to logging.config.dictConfig() .

bpo-46422 : Use dis.Positions in dis.Instruction instead of a regular tuple .

bpo-46434 : pdb now gracefully handles help when __doc__ is missing, for example when run with pregenerated optimized .pyc files.

bpo-43869 : Python uses the same time Epoch on all platforms. Add an explicit unit test to ensure that it’s the case. Patch by Victor Stinner.

bpo-46414 : Add typing.reveal_type() . Patch by Jelle Zijlstra.

bpo-40280 : subprocess now imports Windows-specific imports when msvcrt module is available, and POSIX-specific imports on all other platforms. This gives a clean exception when _posixsubprocess is not available (e.g. Emscripten browser target).

bpo-40066 : IntEnum , IntFlag , and StrEnum use the mixed-in type for their str() and format() output.

bpo-46316 : Optimize pathlib.Path.iterdir() by removing an unnecessary check for special entries.

bpo-29688 : Document pathlib.Path.absolute() (which has always existed).

bpo-43012 : The pathlib module’s obsolete and internal _Accessor class has been removed to prepare the terrain for upcoming enhancements to the module.

bpo-46258 : Speed up math.isqrt() for small positive integers by replacing two division steps with a lookup table.

bpo-46242 : Improve error message when creating a new enum.Enum type subclassing an existing Enum with _member_names_ using enum.Enum.__call__() .

bpo-43118 : Fix a bug in inspect.signature() that was causing it to fail on some subclasses of classes with a __text_signature__ referencing module globals. Patch by Weipeng Hong.

bpo-26552 : Fixed case where failing asyncio.ensure_future() did not close the coroutine. Patch by Kumar Aditya.

bpo-21987 : Fix an issue with tarfile.TarFile.getmember() getting a directory name with a trailing slash.

bpo-46124 : Update zoneinfo to rely on importlib.resources traversable API.

bpo-46103 : Now inspect.getmembers() only gets __bases__ attribute from class type. Patch by Weipeng Hong.

bpo-46080 : Fix exception in argparse help text generation if a argparse.BooleanOptionalAction argument’s default is argparse.SUPPRESS and it has help specified. Patch by Felix Fontein.

bpo-44791 : Fix substitution of ParamSpec in Concatenate with different parameter expressions. Substitution with a list of types returns now a tuple of types. Substitution with Concatenate returns now a Concatenate with concatenated lists of arguments.

bpo-46463 : Fixes escape4chm.py script used when building the CHM documentation file

bpo-43478 : Mocks can no longer be provided as the specs for other Mocks. As a result, an already-mocked object cannot be passed to mock.Mock() . This can uncover bugs in tests since these Mock-derived Mocks will always pass certain tests (e.g. isinstance) and builtin assert functions (e.g. assert_called_once_with) will unconditionally pass.

bpo-46616 : Ensures test_importlib.test_windows cleans up registry keys after completion.

bpo-44359 : test_ftplib now silently ignores socket errors to prevent logging unhandled threading exceptions. Patch by Victor Stinner.

bpo-46600 : Fix test_gdb.test_pycfunction() for Python built with clang -Og . Tolerate inlined functions in the gdb traceback. Patch by Victor Stinner.

bpo-46542 : Fix a Python crash in test_lib2to3 when using Python built in debug mode: limit the recursion limit. Patch by Victor Stinner.

bpo-46576 : test_peg_generator now disables compiler optimization when testing compilation of its own C extensions to significantly speed up the testing on non-debug builds of CPython.

bpo-46542 : Fix test_json tests checking for RecursionError : modify these tests to use support.infinite_recursion() . Patch by Victor Stinner.

bpo-13886 : Skip test_builtin PTY tests on non-ASCII characters if the readline module is loaded. The readline module changes input() behavior, but test_builtin is not intended to test the readline module. Patch by Victor Stinner.

bpo-40280 : Add test.support.requires_fork() decorators to mark tests that require a working os.fork() .

bpo-40280 : Add test.support.requires_subprocess() decorator to mark tests which require working subprocess module or os.spawn* . The wasm32-emscripten platform has no support for processes.

bpo-46126 : Disable ‘descriptions’ when running tests internally.

bpo-46602 : Tidied up configure.ac so that conftest.c is truncated rather than appended. This assists in the case where the ‘rm’ of conftest.c fails to happen between tests. Downstream issues such as a clobbered SOABI can result.

bpo-46600 : Fix the test checking if the C compiler supports -Og option in the ./configure script to also use -Og on clang which supports it. Patch by Victor Stinner.

bpo-38472 : Fix GCC detection in setup.py when cross-compiling. The C compiler is now run with LC_ALL=C. Previously, the detection failed with a German locale.

bpo-46513 : configure no longer uses AC_C_CHAR_UNSIGNED macro and pyconfig.h no longer defines reserved symbol __CHAR_UNSIGNED__ .

bpo-46471 : Use global singletons for single byte bytes objects in deepfreeze.

bpo-46443 : Deepfreeze now uses cached small integers as it saves some space for common small integers.

bpo-46429 : Merge all deep-frozen files into one for space savings. Patch by Kumar Aditya.

bpo-45569 : The build now defaults to using 30-bit digits for Python integers. Previously either 15-bit or 30-bit digits would be selected, depending on the platform. 15-bit digits may still be selected using the --enable-big-digits=15 option to the configure script, or by defining PYLONG_BITS_IN_DIGIT in pyconfig.h .

bpo-45925 : Update Windows installer to use SQLite 3.37.2.

bpo-43112 : Detect musl libc as a separate SOABI (tagged as linux-musl ).

bpo-33125 : The traditional EXE/MSI based installer for Windows is now available for ARM64

bpo-46362 : os.path.abspath(“C:CON”) is now fixed to return “\.CON”, not the same path. The regression was true of all legacy DOS devices such as COM1, LPT1, or NUL.

bpo-44934 : The installer now offers a command-line only option to add the installation directory to the end of PATH instead of at the start.

bpo-45925 : Update macOS installer to SQLite 3.37.2.

bpo-45296 : Clarify close, quit, and exit in IDLE. In the File menu, ‘Close’ and ‘Exit’ are now ‘Close Window’ (the current one) and ‘Exit’ is now ‘Exit IDLE’ (by closing all windows). In Shell, ‘quit()’ and ‘exit()’ mean ‘close Shell’. If there are no other windows, this also exits IDLE.

bpo-40170 : Remove the PyHeapType_GET_MEMBERS() macro. It was exposed in the public C API by mistake, it must only be used by Python internally. Use the PyTypeObject.tp_members member instead. Patch by Victor Stinner.

bpo-40170 : Move _Py_GetAllocatedBlocks() and _PyObject_DebugMallocStats() private functions to the internal C API. Patch by Victor Stinner.

bpo-46433 : The internal function _PyType_GetModuleByDef now correctly handles inheritance patterns involving static types.

bpo-45459 : Py_buffer and various Py_buffer related functions are now part of the limited API and stable ABI.

bpo-14916 : Fixed bug in the tokenizer that prevented PyRun_InteractiveOne from parsing from the provided FD.

Python 3.11.0 alpha 4 ¶

Release date: 2022-01-13

bpo-46070 : Py_EndInterpreter() now explicitly untracks all objects currently tracked by the GC. Previously, if an object was used later by another interpreter, calling PyObject_GC_UnTrack() on the object crashed if the previous or the next object of the PyGC_Head structure became a dangling pointer. Patch by Victor Stinner.

bpo-46347 : Fix memory leak in PyEval_EvalCodeEx.

bpo-46339 : Fix a crash in the parser when retrieving the error text for multi-line f-strings expressions that do not start in the first line of the string. Patch by Pablo Galindo

bpo-46331 : Do not set line number of instruction storing doc-string. Fixes regression introduced in 3.11 alpha.

bpo-46314 : Remove spurious “call” event when creating a lambda function that was accidentally introduced in 3.11a4.

bpo-46289 : ASDL declaration of FormattedValue has changed to reflect conversion field is not optional.

bpo-46297 : Fixed an interpreter crash on bootup with multiple PythonPaths set in the Windows registry. Patch by Derzsi Dániel.

bpo-46237 : Fix the line number of tokenizer errors inside f-strings. Patch by Pablo Galindo.

bpo-46263 : We always expect the “use_frozen_modules” config to be set, now that getpath.c was rewritten in pure Python and the logic improved.

bpo-46006 : Fix a regression when a type method like __init__() is modified in a subinterpreter. Fix a regression in _PyUnicode_EqualToASCIIId() and type update_slot() . Revert the change which made the Unicode dictionary of interned strings compatible with subinterpreters: the internal interned dictionary is shared again by all interpreters. Patch by Victor Stinner.

bpo-45923 : Add RESUME opcode. This is a logical no-op. It is emitted by the compiler anywhere a Python function can be entered. It is used by the interpreter to perform tracing and optimizer checks.

bpo-46208 : Fix the regression of os.path.normpath(“A/../../B”) not returning expected “../B” but “B”.

bpo-46240 : Correct the error message for unclosed parentheses when the tokenizer doesn’t reach the end of the source when the error is reported. Patch by Pablo Galindo

bpo-46009 : Remove the GEN_START opcode.

bpo-46235 : Certain sequence multiplication operations like [0] * 1_000 are now faster due to reference-counting optimizations. Patch by Dennis Sweeney.

bpo-46221 : PREP_RERAISE_STAR no longer pushes lasti to the stack.

bpo-46202 : Remove POP_EXCEPT_AND_RERAISE and replace it by an equivalent sequence of other opcodes.

bpo-46085 : Fix iterator cache mechanism of OrderedDict .

bpo-46055 : Speed up shifting operation involving integers less than PyLong_BASE . Patch by Xinhang Xu.

bpo-46110 : Add a maximum recursion check to the PEG parser to avoid stack overflow. Patch by Pablo Galindo

bpo-46107 : Fix bug where ExceptionGroup.split() and ExceptionGroup.subgroup() did not copy the exception group’s __note__ field to the parts.

bpo-45711 : The interpreter state’s representation of handled exceptions (a.k.a exc_info, or _PyErr_StackItem) now has only the exc_value field, exc_type and exc_traceback have been removed as their values can be derived from exc_value .

bpo-44525 : Replace the four call bytecode instructions which one pre-call instruction and two call instructions.

Removes CALL_FUNCTION , CALL_FUNCTION_KW , CALL_METHOD and CALL_METHOD_KW .

Adds CALL_NO_KW and CALL_KW call instructions, and PRECALL_METHOD prefix for pairing with LOAD_METHOD .

bpo-46039 : Remove the YIELD_FROM instruction and replace it with the SEND instruction which performs the same operation, but without the loop.

bpo-45635 : The code called from _PyErr_Display() was refactored to improve error handling. It now exits immediately upon an unrecoverable error.

bpo-46054 : Fix parser error when parsing non-utf8 characters in source files. Patch by Pablo Galindo.

bpo-46042 : Improve the location of the caret in SyntaxError exceptions emitted by the symbol table. Patch by Pablo Galindo.

bpo-46049 : Ensure ._pth files work as intended on platforms other than Windows.

bpo-46048 : Fixes parsing of ._pth files on startup so that single-character paths are correctly read.

bpo-37971 : Fix a bug where the line numbers given in a traceback when a decorator application raised an exception were wrong.

bpo-46031 : Add POP_JUMP_IF_NOT_NONE and POP_JUMP_IF_NONE opcodes to speed up conditional jumps.

bpo-45654 : Deepfreeze runpy , patch by Kumar Aditya.

bpo-46025 : Fix a crash in the atexit module involving functions that unregister themselves before raising exceptions. Patch by Pablo Galindo.

bpo-46000 : Improve compatibility of the curses module with NetBSD curses.

bpo-44525 : Specialize the CALL_FUNCTION instruction for calls to builtin types with a single argument. Speeds up range(x) , list(x) , and specifically type(obj) .

bpo-42918 : Fix bug where the built-in compile() function did not always raise a SyntaxError when passed multiple statements in ‘single’ mode. Patch by Weipeng Hong.

bpo-45953 : The main interpreter in _PyRuntimeState.interpreters is now statically allocated (as part of _PyRuntime). Likewise for the initial thread state of each interpreter. This means less allocation during runtime init, as well as better memory locality for these key state objects.

bpo-45292 : Complete the PEP 654 implementation: add except* .

bpo-43413 : Revert changes in set.__init__ . Subclass of set needs to define a __init__() method if it defines a __new__() method with additional keyword parameters.

bpo-43931 : Added the Py_Version constant which bears the same value as PY_VERSION_HEX . Patch by Gabriele N. Tornetta.

bpo-46342 : The @typing.final decorator now sets the __final__ attribute on the decorated object to allow runtime introspection. Patch by Jelle Zijlstra.

bpo-46328 : Added the sys.exception() method which returns the active exception instance.

bpo-46307 : Add string.Template.is_valid() and string.Template.get_identifiers() methods.

bpo-46306 : Assume that types.CodeType always has types.CodeType.co_firstlineno in doctest .

bpo-40479 : Fix hashlib usedforsecurity option to work correctly with OpenSSL 3.0.0 in FIPS mode.

bpo-46070 : Fix possible segfault when importing the asyncio module from different sub-interpreters in parallel. Patch by Erlend E. Aasland.

bpo-46244 : Removed __slots__ from typing.ParamSpec and typing.TypeVar . They served no purpose. Patch by Arie Bovenberg.

bpo-46278 : Reflect context argument in AbstractEventLoop.call_*() methods. Loop implementations already support it.

bpo-46269 : Remove special-casing of __new__ in enum.Enum.__dir__() .

bpo-46266 : Improve day constants in calendar .

Now all constants ( MONDAY … SUNDAY ) are documented, tested, and added to __all__ .

bpo-46257 : Optimized the mean, variance, and stdev functions in the statistics module. If the input is an iterator, it is consumed in a single pass rather than eating memory by conversion to a list. The single pass algorithm is about twice as fast as the previous two pass code.

bpo-41011 : Added two new variables to pyvenv.cfg which is generated by venv module: executable for the executable and command for the command line used to create the environment.

bpo-46239 : Improve error message when importing asyncio.windows_events on non-Windows.

bpo-46238 : Reuse _winapi constants in asyncio.windows_events .

bpo-46222 : Adding SF_NOCACHE sendfile constant for FreeBSD for the posixmodule.

bpo-37295 : Add fast path for 0 <= k <= n <= 67 for math.comb() .

bpo-46176 : Adding the MAP_STACK constant for the mmap module.

bpo-43424 : Deprecate webbrowser.MacOSXOSAScript._name and use name instead.

bpo-45321 : Added missing error codes to module xml.parsers.expat.errors .

bpo-46125 : Refactor tests to test traversable API directly. Includes changes from importlib 5.4.0.

bpo-46118 : Moved importlib.resources and its related functionality to a package.

bpo-37578 : Add include_hidden parameter to glob() and iglob() to match hidden files and directories when using special characters like * , ** , ? and [] .

bpo-20369 : concurrent.futures.wait() no longer blocks forever when given duplicate Futures. Patch by Kumar Aditya.

bpo-46105 : Honor spec when generating requirement specs with urls and extras (importlib_metadata 4.8.3).

bpo-44893 : EntryPoint objects are no longer tuples. Recommended means to access is by attribute (‘.name’, ‘.group’) or accessor (‘.load()’). Access by index is deprecated and will raise deprecation warning.

bpo-22815 : Print unexpected successes together with failures and errors in summary in unittest.TextTestResult .

bpo-22047 : Calling add_argument_group() on an argument group is deprecated. Calling add_argument_group() or add_mutually_exclusive_group() on a mutually exclusive group is deprecated.

These features were never supported and do not always work correctly. The functions exist on the API by accident through inheritance and will be removed in the future.

bpo-26952 : argparse raises ValueError with clear message when trying to render usage for an empty mutually exclusive group. Previously it raised a cryptic IndexError .

bpo-45615 : Functions in the traceback module raise TypeError rather than AttributeError when an exception argument is not of type BaseException .

bpo-16594 : Add allow allow_reuse_port flag in socketserver.

bpo-27718 : Fix help for the signal module. Some functions (e.g. signal() and getsignal() ) were omitted.

bpo-46032 : The registry() method of functools.singledispatch() functions checks now the first argument or the first parameter annotation and raises a TypeError if it is not supported. Previously unsupported “types” were ignored (e.g. typing.List[int] ) or caused an error at calling time (e.g. list[int] ).

bpo-46014 : Add ability to use typing.Union and types.UnionType as dispatch argument to functools.singledispatch . Patch provided by Yurii Karabas.

bpo-27062 : Add __all__ to inspect , patch by Kumar Aditya.

bpo-46018 : Ensure that math.expm1() does not raise on underflow.

bpo-46016 : Adding F_DUP2FD and F_DUP2FD_CLOEXEC constants from FreeBSD into the fcntl module.

bpo-45755 : typing generic aliases now reveal the class attributes of the original generic class when passed to dir() . This was the behavior up to Python 3.6, but was changed in 3.7-3.9.

bpo-45874 : The empty query string, consisting of no query arguments, is now handled correctly in urllib.parse.parse_qsl . This caused problems before when strict parsing was enabled.

bpo-44674 : Change how dataclasses disallows mutable default values. It used to use a list of known types (list, dict, set). Now it disallows unhashable objects to be defaults. It’s using unhashability as a proxy for mutability. Patch by Eric V. Smith, idea by Raymond Hettinger.

bpo-23882 : Remove namespace package (PEP 420) support from unittest discovery. It was introduced in Python 3.4 but has been broken since Python 3.7.

bpo-25066 : Added a __repr__() method to multiprocessing.Event objects, patch by Kumar Aditya.

bpo-45643 : Added signal.SIGSTKFLT on platforms where this signal is defined.

bpo-44092 : Fetch across rollback no longer raises InterfaceError . Instead we leave it to the SQLite library to handle these cases. Patch by Erlend E. Aasland.

bpo-42413 : Replace concurrent.futures.TimeoutError and asyncio.TimeoutError with builtin TimeoutError , keep these names as deprecated aliases.

bpo-46196 : Document method cmd.Cmd.columnize() .

bpo-46120 : State that | is preferred for readability over Union in the typing docs.

bpo-46109 : Extracted importlib.resources and importlib.resources.abc documentation into separate files.

bpo-19737 : Update the documentation for the globals() function.

bpo-46296 : Add a test case for enum with _use_args_ == True and _member_type_ == object .

bpo-46205 : Fix hang in runtest_mp due to race condition

bpo-46263 : Fix test_capi on FreeBSD 14-dev: instruct jemalloc to not fill freed memory with junk byte.

bpo-46262 : Cover ValueError path in tests for enum.Flag._missing_() .

bpo-46150 : Now fakename in test_pathlib.PosixPathTest.test_expanduser is checked to be non-existent.

bpo-46129 : Rewrite asyncio.locks tests with unittest.IsolatedAsyncioTestCase usage.

bpo-23819 : Fixed asyncio tests in python optimized mode. Patch by Kumar Aditya.

bpo-46114 : Fix test case for OpenSSL 3.0.1 version. OpenSSL 3.0 uses 0xMNN00PP0L .

bpo-44133 : When Python is configured with --without-static-libpython , the Python static library (libpython.a) is no longer built. Patch by Victor Stinner.

bpo-44133 : When Python is built without --enable-shared , the python program is now linked to object files, rather than being linked to the Python static library (libpython.a), to make sure that all symbols are exported. Previously, the linker omitted some symbols like the Py_FrozenMain() function. Patch by Victor Stinner.

bpo-40280 : The configure script has a new option --with-emscripten-target to select browser or node as Emscripten build target.

bpo-46315 : Added and fixed #ifdef HAVE_FEATURE checks for functionality that is not available on WASI platform.

bpo-45723 : Fixed a regression in configure check for select.epoll() .

bpo-46263 : configure no longer sets MULTIARCH on FreeBSD platforms.

bpo-46106 : Updated OpenSSL to 1.1.1m in Windows builds, macOS installer builds, and CI. Patch by Kumar Aditya.

bpo-46088 : Automatically detect or install bootstrap Python runtime when building from Visual Studio.

bpo-46072 : Add a –with-pystats configure option to turn on internal statistics gathering.

bpo-40280 : A new directory Tools/wasm contains WebAssembly-related helpers like config.site override for wasm32-emscripten, wasm assets generator to bundle the stdlib, and a README.

bpo-46023 : makesetup no longer builds extensions that have been marked as disabled . This allows users to disable modules in Modules/Setup.local .

bpo-45949 : Use pure Python freeze_module for all but importlib bootstrap files. --with-freeze-module configure option is no longer needed for cross builds.

bpo-46217 : Removed parameter that is unsupported on Windows 8.1 and early Windows 10 and may have caused build or runtime failures.

bpo-40477 : The Python Launcher app for macOS now properly launches scripts and, if necessary, the Terminal app when running on recent macOS releases.

bpo-46236 : Fix a bug in PyFunction_GetAnnotations() that caused it to return a tuple instead of a dict .

bpo-46140 : PyBuffer_GetPointer() , PyBuffer_FromContiguous() , PyBuffer_ToContiguous() and PyMemoryView_FromBuffer() now take buffer info by const Py_buffer * instead of Py_buffer * , as they do not need mutability. PyBuffer_FromContiguous() also now takes the source buffer as const void * , and similarly PyBuffer_GetPointer() takes the strides as const Py_ssize_t * .

bpo-45855 : Document that the no_block argument to PyCapsule_Import() is a no-op now.

bpo-45855 : Replaced deprecated usage of PyImport_ImportModuleNoBlock() with PyImport_ImportModule() in stdlib modules. Patch by Kumar Aditya.

bpo-46007 : The PyUnicode_CHECK_INTERNED() macro has been excluded from the limited C API. It was never usable there, because it used internal structures which are not available in the limited C API. Patch by Victor Stinner.

Python 3.11.0 alpha 3 ¶

Release date: 2021-12-08

bpo-46009 : Restore behavior from 3.9 and earlier when sending non-None to newly started generator. In 3.9 this did not affect the state of the generator. In 3.10.0 and 3.10.1 gen_func().send(0) is equivalent to gen_func().throw(TypeError(...) which exhausts the generator. In 3.10.2 onward, the behavior has been reverted to that of 3.9.

bpo-46004 : Fix the SyntaxError location for errors involving for loops with invalid targets. Patch by Pablo Galindo

bpo-45711 : _PyErr_ChainStackItem() no longer normalizes exc_info (including setting the traceback on the exception instance) because exc_info is always normalized.

bpo-45607 : The __note__ field was added to BaseException . It is None by default but can be set to a string which is added to the exception’s traceback.

bpo-45947 : Place pointers to dict and values immediately before GC header. This reduces number of dependent memory loads to access either dict or values from 3 to 1.

bpo-45915 : is_valid_fd now uses faster fcntl(fd, F_GETFD) on Linux, macOS, and Windows.

bpo-44530 : Reverts a change to the code.__new__ audit event from an earlier prerelease.

bpo-42268 : Fail the configure step if the selected compiler doesn’t support memory sanitizer. Patch by Pablo Galindo

bpo-45711 : The three values of exc_info are now always consistent with each other. In particular, the type and traceback fields are now derived from the exception instance. This impacts the return values of sys.exc_info() and PyErr_GetExcInfo() if the exception instance is modified while the exception is handled, as well as PyErr_SetExcInfo() , which now ignores the type and traceback arguments provided to it.

bpo-45727 : Refine the custom syntax error that suggests that a comma may be missing to trigger only when the expressions are detected between parentheses or brackets. Patch by Pablo Galindo

bpo-45885 : Specialized the COMPARE_OP opcode using the PEP 659 machinery.

bpo-45786 : Allocate space for the interpreter frame in the frame object, to avoid an additional allocation when the frame object outlives the frame activation.

bpo-45614 : Fix traceback display for exceptions with invalid module name.

bpo-45813 : Fix crash when calling coro.cr_frame.clear() after coroutine has been freed.

bpo-45811 : Improve the tokenizer errors when encountering invisible control characters in the parser. Patch by Pablo Galindo

bpo-45848 : Allow the parser to obtain error lines directly from encoded files. Patch by Pablo Galindo

bpo-45709 : Restore behavior from 3.10 when tracing an exception raised within a with statement.

bpo-44525 : Adds new COPY_FREE_VARS opcode, to make copying of free variables from function to frame explicit. Helps optimization of calls to Python function.

bpo-45829 : Specialize BINARY_SUBSCR for classes with a __getitem__ method implemented in Python

bpo-45826 : Fixed a crash when calling .with_traceback(None) on NameError . This occurs internally in unittest.TestCase.assertRaises() .

bpo-45822 : Fixed a bug in the parser that was causing it to not respect PEP 263 coding cookies when no flags are provided. Patch by Pablo Galindo

bpo-45820 : Fix a segfault when the parser fails without reading any input. Patch by Pablo Galindo

bpo-45636 : Simplify the implementation of BINARY_OP by indexing into an array of function pointers (rather than switching on the oparg).

bpo-42540 : Fix crash when os.fork() is called with an active non-default memory allocator.

bpo-45738 : Fix computation of error location for invalid continuation characters in the parser. Patch by Pablo Galindo.

bpo-45636 : Remove an existing “fast path” for old-style string formatting, since it no longer appears to have any measurable impact.

bpo-45753 : Make recursion checks a bit more efficient by tracking amount of calls left before overflow.

bpo-45773 : Fix a compiler hang when attempting to optimize certain jump patterns.

bpo-45764 : The parser now gives a better error message when leaving out the opening parenthesis ( after a def -statement:

bpo-45609 : Specialized the STORE_SUBSCR opcode using the PEP 659 machinery.

bpo-45636 : Replace all numeric BINARY_* and INPLACE_* instructions with a single BINARY_OP implementation.

bpo-45582 : Path calculation (known as getpath ) has been reimplemented as a frozen Python module. This should have no visible impact, but may affect calculation of all paths referenced in sys and sysconfig .

bpo-45450 : Improve the syntax error message for parenthesized arguments. Patch by Pablo Galindo.

bpo-27946 : Fix possible crash when getting an attribute of xml.etree.ElementTree.Element simultaneously with replacing the attrib dict.

bpo-45711 : Make asyncio normalize exceptions as soon as they are captured with PyErr_Fetch() , and before they are stored as an exc_info triplet. This brings asyncio in line with the rest of the codebase, where an exc_info triplet is always normalized.

bpo-23819 : Replaced asserts with exceptions in asyncio, patch by Kumar Aditya.

bpo-13236 : unittest.TextTestResult and unittest.TextTestRunner flush now the output stream more often.

bpo-45917 : Added math.exp2() :, which returns 2 raised to the power of x.

bpo-37658 : Fix issue when on certain conditions asyncio.wait_for() may allow a coroutine to complete successfully, but fail to return the result, potentially causing memory leaks or other issues.

bpo-45876 : Improve the accuracy of stdev() and pstdev() in the statistics module. When the inputs are floats or fractions, the output is a correctly rounded float

bpo-44649 : Handle dataclass(slots=True) with a field that has default a default value, but for which init=False.

bpo-45803 : Added missing kw_only parameter to dataclasses.make_dataclass().

bpo-45837 : The turtle.RawTurtle.settiltangle() is deprecated since Python 3.1, it now emits a deprecation warning and will be removed in Python 3.13.

Use turtle.RawTurtle.tiltangle() instead.

turtle.RawTurtle.tiltangle() was earlier incorrectly marked as deprecated, its docstring has been corrected.

bpo-45831 : faulthandler can now write ASCII-only strings (like filenames and function names) with a single write() syscall when dumping a traceback. It reduces the risk of getting an unreadable dump when two threads or two processes dump a traceback to the same file (like stderr) at the same time. Patch by Victor Stinner.

bpo-45828 : sqlite C callbacks now use unraisable exceptions if callback tracebacks are enabled. Patch by Erlend E. Aasland.

bpo-41735 : Fix thread lock in zlib.Decompress.flush() method before PyObject_GetBuffer .

bpo-45235 : Reverted an argparse bugfix that caused regression in the handling of default arguments for subparsers. This prevented leaf level arguments from taking precedence over root level arguments.

bpo-45754 : Fix a regression in Python 3.11a1 and 3.11a2 where sqlite3 incorrectly would use SQLITE_LIMIT_LENGTH when checking SQL statement lengths. Now, SQLITE_LIMIT_SQL_LENGTH is used. Patch by Erlend E. Aasland.

bpo-45766 : Added proportional option to statistics.linear_regression() .

bpo-45765 : In importlib.metadata, fix distribution discovery for an empty path.

bpo-45757 : Fix bug where dis produced an incorrect oparg when EXTENDED_ARG is followed by an opcode that does not use its argument.

bpo-45644 : In-place JSON file formatting using python3 -m json.tool infile infile now works correctly, previously it left the file empty. Patch by Chris Wesseling.

bpo-45703 : When a namespace package is imported before another module from the same namespace is created/installed in a different sys.path location while the program is running, calling the importlib.invalidate_caches() function will now also guarantee the new module is noticed.

bpo-45535 : Improve output of dir() with Enums.

bpo-45664 : Fix types.resolve_bases() and types.new_class() for types.GenericAlias instance as a base.

bpo-45663 : Fix dataclasses.is_dataclass() for dataclasses which are subclasses of types.GenericAlias .

bpo-45662 : Fix the repr of dataclasses.InitVar with a type alias to the built-in class, e.g. InitVar[list[int]] .

bpo-43137 : Launch GNOME web browsers via gio tool instead of obsolete gvfs-open

bpo-45429 : On Windows, time.sleep() now uses a waitable timer which supports high-resolution timers. Patch by Donghee Na and Eryk Sun.

bpo-37295 : Optimize math.comb() and math.perm() .

bpo-45514 : Deprecated legacy functions in importlib.resources .

bpo-45507 : Add tests for truncated/missing trailers in gzip.decompress implementation.

bpo-45359 : Implement PEP 585 for graphlib.TopologicalSorter .

bpo-44733 : Add max_tasks_per_child to concurrent.futures.ProcessPoolExecutor . This allows users to specify the maximum number of tasks a single process should execute before the process needs to be restarted.

bpo-28806 : Improve netrc library. netrc file no longer needs to contain all tokens. And if the login name is anonymous, security check is no longer need.

bpo-43498 : Avoid a possible “RuntimeError: dictionary changed size during iteration” when adjusting the process count of ProcessPoolExecutor .

bpo-42158 : Add MIME types for N-quads, N-triples, Notation3 and TriG to mimetypes .

bpo-30533 : Add inspect.getmembers_static() , it return all members without triggering dynamic lookup via the descriptor protocol. Patch by Weipeng Hong.

bpo-42238 : make -C Doc suspicious will be removed soon in favor of make -C Doc check , mark it as deprecated.

bpo-45840 : Improve cross-references in the documentation for the data model.

bpo-45640 : Properly marked-up grammar tokens in the documentation are now clickable and take you to the definition of a given piece of grammar. Patch by Arthur Milchior.

bpo-45788 : Link doc for sys.prefix to sysconfig doc on installation paths.

bpo-45772 : socket.socket documentation is corrected to a class from a function.

bpo-45392 : Update the docstring of the type built-in to remove a redundant line and to mention keyword arguments for the constructor.

bpo-45250 : Update the documentation to note that CPython does not consistently require iterators to define __iter__ .

bpo-25381 : In the extending chapter of the extending doc, update a paragraph about the global variables containing exception information.

bpo-43905 : Expanded astuple() and asdict() docs, warning about deepcopy being applied and providing a workaround.

bpo-45695 : Out-of-tree builds with a read-only source directory are now tested by CI.

bpo-19460 : Add new Test for Lib/email/mime/nonmultipart.py::MIMENonMultipart .

bpo-45835 : Fix race condition in test_queue tests with multiple “feeder” threads.

bpo-45783 : The test for the freeze tool now handles file moves and deletions.

bpo-45745 : Remove the --findleaks command line option of regrtest: use the --fail-env-changed option instead. Since Python 3.7, it was a deprecated alias to the --fail-env-changed option.

bpo-45701 : Add tests with tuple type with functools.lru_cache() to test_functools .

bpo-44035 : CI now verifies that autoconf files have been regenerated with a current and unpatched autoconf package.

bpo-45950 : The build system now uses a _bootstrap_python interpreter for freezing and deepfreezing again. To speed up build process the build tools _bootstrap_python and _freeze_module are no longer build with LTO.

bpo-45881 : The configure script now accepts --with-build-python and --with-freeze-module options to make cross compiling easier.

bpo-40280 : Emscripten platform now uses .wasm suffix by default.

bpo-40280 : Disable unusable core extension modules on WASM/Emscripten targets.

bpo-40280 : configure now checks for socket shutdown function. The check makes it possible to disable SYS_shutdown with ac_cv_func_shutdown=no in CONFIG_SITE.

bpo-40280 : configure now checks for functions fork1, getegid, geteuid, getgid, getppid, getuid, opendir, pipe, system, wait, ttyname .

bpo-33393 : Update config.guess to 2021-06-03 and config.sub to 2021-08-14. Makefile now has an update-config target to make updating more convenient.

bpo-45866 : make regen-all now produces the same output when run from a directory other than the source tree: when building Python out of the source tree. pegen now strips directory of the “generated by pygen from <FILENAME>” header Patch by Victor Stinner.

bpo-40280 : configure now accepts machine wasm32 or wasm64 and OS wasi or emscripten for cross building, e.g. wasm32-unknown-emscripten , wasm32-wasi , or wasm32-unknown-wasi .

bpo-41498 : Python now compiles on platforms without sigset_t . Several functions in signal are not available when sigset_t is missing.

Based on patch by Roman Yurchak for pyodide.

bpo-45881 : setup.py now uses CC from environment first to discover multiarch and cross compile paths.

bpo-45886 : The _freeze_module program path can now be overridden on the command line, e.g. make FREEZE_MODULE=../x86_64/Program/_freeze_module .

bpo-45873 : Get rid of the _bootstrap_python build step. The deepfreeze.py script is now run using $(PYTHON_FOR_REGEN) which can be Python 3.7 or newer (on Windows, 3.8 or newer).

bpo-45847 : Port builtin hashlib extensions to PY_STDLIB_MOD macro and addext() .

bpo-45723 : Add autoconf helpers for saving and restoring environment variables:

SAVE_ENV : Save $CFLAGS , $LDFLAGS , $LIBS , and $CPPFLAGS .

RESTORE_ENV : Restore $CFLAGS , $LDFLAGS , $LIBS , and $CPPFLAGS .

WITH_SAVE_ENV([SCRIPT]) : Run SCRIPT wrapped with SAVE_ENV and RESTORE_ENV .

bpo-45573 : Mandatory core modules, that are required to bootstrap Python, are now in Modules/Setup.bootstrap .

bpo-45573 : configure now creates Modules/Setup.stdlib with conditionally enabled/disabled extension module lines. The file is not used, yet.

bpo-45573 : configure now uses a unified format to set state, compiler flags, and linker flags in Makefile. The new macro PY_STDLIB_MOD sets three variables that are consumed by Modules/Setup and setup.py .

bpo-45816 : Python now supports building with Visual Studio 2022 (MSVC v143, VS Version 17.0). Patch by Jeremiah Vivian.

bpo-45800 : Settings for pyexpat C extension are now detected by configure . The bundled expat library is built in Makefile .

bpo-45798 : Settings for decimal internal C extension are now detected by configure . The bundled libmpdec library is built in Makefile .

bpo-45723 : configure has a new option --with-pkg-config to disable or require pkg-config.

bpo-45774 : The build dependencies for sqlite3 are now detected by configure and pkg-config . Patch by Erlend E. Aasland.

bpo-45763 : The build dependencies for zlib , bz2 , and lzma are now detected by configure .

bpo-45747 : gdbm and dbm build dependencies are now detected by configure .

bpo-45743 : On macOS, the build system no longer passes search_paths_first to the linker. The flag has been the default since Xcode 4 / macOS 10.6.

bpo-45723 : configure.ac is now compatible with autoconf 2.71. Deprecated checks STDC_HEADERS and AC_HEADER_TIME have been removed.

bpo-45723 : configure now prints a warning when pkg-config is missing.

bpo-45731 : configure --enable-loadable-sqlite-extensions is now handled by new PY_SQLITE_ENABLE_LOAD_EXTENSION macro instead of logic in setup.py.

bpo-45723 : configure.ac now uses custom helper macros and AC_CACHE_CHECK to simplify and speed up configure runs.

bpo-45696 : Skip the marshal step for frozen modules by generating C code that produces a set of ready-to-use code objects. This speeds up startup time by another 10% or more.

bpo-45561 : Run smelly.py tool from $(srcdir).

bpo-46105 : Fixed calculation of sys.path in a venv on Windows.

bpo-45901 : When installed through the Microsoft Store and set as the default app for *.py files, command line arguments will now be passed to Python when invoking a script without explicitly launching Python (that is, script.py args rather than python script.py args ).

bpo-45616 : Fix Python Launcher’s ability to distinguish between versions 3.1 and 3.10 when either one is explicitly requested. Previously, 3.1 would be used if 3.10 was requested but not installed, and 3.10 would be used if 3.1 was requested but 3.10 was installed.

bpo-45850 : Implement changes to build with deep-frozen modules on Windows. Note that we now require Python 3.10 as the “bootstrap” or “host” Python.

bpo-45732 : Updates bundled Tcl/Tk to 8.6.12.

bpo-45720 : Internal reference to shlwapi.dll was dropped to help improve startup time. This DLL will no longer be loaded at the start of every Python process.

bpo-45732 : Update python.org macOS installer to use Tcl/Tk 8.6.12.

bpo-39026 : Fix Python.h to build C extensions with Xcode: remove a relative include from Include/cpython/pystate.h .

Python 3.11.0 alpha 2 ¶

Release date: 2021-11-05

bpo-45716 : Improve the SyntaxError message when using True , None or False as keywords in a function call. Patch by Pablo Galindo.

bpo-45688 : sys.stdlib_module_names now contains the macOS-specific module _scproxy .

bpo-45379 : Clarify ImportError message when we try to explicitly import a frozen module but frozen modules are disabled.

bpo-44525 : Specialize simple calls to Python functions (no starargs, keyword dict, or closure)

bpo-45530 : Cases of sorting using tuples as keys may now be significantly faster in some cases. Patch by Tim Peters.

The order of the result may differ from earlier releases if the tuple elements don’t define a total ordering (see Value comparisons for information on total ordering). It’s generally true that the result of sorting simply isn’t well-defined in the absence of a total ordering on list elements.

bpo-45526 : In obmalloc, set ADDRESS_BITS to not ignore any bits (ignored 16 before). That is safer in the case that the kernel gives user-space virtual addresses that span a range greater than 48 bits.

bpo-30570 : Fixed a crash in issubclass() from infinite recursion when searching pathological __bases__ tuples.

bpo-45521 : Fix a bug in the obmalloc radix tree code. On 64-bit machines, the bug causes the tree to hold 46-bits of virtual addresses, rather than the intended 48-bits.

bpo-45494 : Fix parser crash when reporting errors involving invalid continuation characters. Patch by Pablo Galindo.

bpo-45445 : Python now fails to initialize if it finds an invalid -X option in the command line. Patch by Pablo Galindo.

bpo-45340 : Object attributes are held in an array instead of a dictionary. An object’s dictionary are created lazily, only when needed. Reduces the memory consumption of a typical Python object by about 30%. Patch by Mark Shannon.

bpo-45408 : Fix a crash in the parser when reporting tokenizer errors that occur at the same time unclosed parentheses are detected. Patch by Pablo Galindo.

bpo-29410 : Add SipHash13 for string hash algorithm and use it by default.

bpo-45385 : Fix reference leak from descr_check. Patch by Donghee Na.

bpo-45367 : Specialized the BINARY_MULTIPLY opcode to BINARY_MULTIPLY_INT and BINARY_MULTIPLY_FLOAT using the PEP 659 machinery.

bpo-21736 : Frozen stdlib modules now have __file__ to the .py file they would otherwise be loaded from, if possible. For packages, __path__ now has the correct entry instead of being an empty list, which allows unfrozen submodules to be imported. These are set only if the stdlib directory is known when the runtime is initialized. Note that the file at __file__ is not guaranteed to exist. None of this affects non-stdlib frozen modules nor, for now, frozen modules imported using PyImport_ImportFrozenModule() . Also, at the moment co_filename is not updated for the module.

bpo-45020 : For frozen stdlib modules, record the original module name as module.__spec__.loader_state.origname . If the value is different than module.__spec__.name then the module was defined as an alias in Tools/scripts/freeze_modules.py. If it is None then the module comes from a source file outside the stdlib.

bpo-45324 : In FrozenImporter.find_spec(), we now preserve the information needed in exec_module() to load the module. This change mostly impacts internal details, rather than changing the importer’s behavior.

bpo-45292 : Implement PEP 654 . Add ExceptionGroup and BaseExceptionGroup . Update traceback display code.

bpo-40116 : Change to the implementation of split dictionaries. Classes where the instances differ either in the exact set of attributes, or in the order in which those attributes are set, can still share keys. This should have no observable effect on users of Python or the C-API. Patch by Mark Shannon.

bpo-44050 : Extensions that indicate they use global state (by setting m_size to -1) can again be used in multiple interpreters. This reverts to behavior of Python 3.8.

bpo-44525 : Setup initial specialization infrastructure for the CALL_FUNCTION opcode. Implemented initial specializations for C function calls:

CALL_FUNCTION_BUILTIN_O for METH_O flag.

CALL_FUNCTION_BUILTIN_FAST for METH_FASTCALL flag without keywords.

CALL_FUNCTION_LEN for len(o) .

CALL_FUNCTION_ISINSTANCE for isinstance(o, t) .

bpo-44511 : Improve the generated bytecode for class and mapping patterns.

bpo-43706 : Speed up calls to enumerate() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-45679 : Fix caching of multi-value typing.Literal . Literal[True, 2] is no longer equal to Literal[1, 2] .

bpo-42064 : Convert sqlite3 to multi-phase initialisation (PEP 489). Patches by Erlend E. Aasland.

bpo-45438 : Fix typing.Signature string representation for generic builtin types.

bpo-45613 : sqlite3 now sets sqlite3.threadsafety based on the default threading mode the underlying SQLite library has been compiled with. Patch by Erlend E. Aasland.

bpo-45574 : Fix warning about print_escape being unused.

bpo-45581 : sqlite3.connect() now correctly raises MemoryError if the underlying SQLite API signals memory error. Patch by Erlend E. Aasland.

bpo-45557 : pprint.pprint() now handles underscore_numbers correctly. Previously it was always setting it to False .

bpo-44019 : Add operator.call() to operator.__all__ . Patch by Kreusada.

bpo-42174 : shutil.get_terminal_size() now falls back to sane values if the column or line count are 0.

bpo-35673 : Improve the introspectability of the __loader__ attribute for namespace packages. importlib.machinery.NamespaceLoader is now public, and implements the importlib.abc.InspectLoader interface. _NamespaceLoader is kept for backward compatibility.

bpo-45515 : Add references to zoneinfo in the datetime documentation, mostly replacing outdated references to dateutil.tz . Change by Paul Ganssle.

bpo-45475 : Reverted optimization of iterating gzip.GzipFile , bz2.BZ2File , and lzma.LZMAFile (see bpo-43787 ) because it caused regression when user iterate them without having reference of them. Patch by Inada Naoki.

bpo-45489 : Update ForwardRef to support | operator. Patch by Donghee Na.

bpo-42222 : Removed deprecated support for float arguments in randrange() .

bpo-45428 : Fix a regression in py_compile when reading filenames from standard input.

bpo-45467 : Fix incremental decoder and stream reader in the “raw-unicode-escape” codec. Previously they failed if the escape sequence was split.

bpo-45461 : Fix incremental decoder and stream reader in the “unicode-escape” codec. Previously they failed if the escape sequence was split.

bpo-45239 : Fixed email.utils.parsedate_tz() crashing with UnboundLocalError on certain invalid input instead of returning None . Patch by Ben Hoyt.

bpo-45417 : Fix quadratic behaviour in the enum module: Creation of enum classes with a lot of entries was quadratic.

bpo-45249 : Fix the behaviour of traceback.print_exc() when displaying the caret when the end_offset in the exception is set to 0. Patch by Pablo Galindo

bpo-45416 : Fix use of asyncio.Condition with explicit asyncio.Lock objects, which was a regression due to removal of explicit loop arguments. Patch by Joongi Kim.

bpo-20028 : Empty escapechar/quotechar is not allowed when initializing csv.Dialect . Patch by Vajrasky Kok and Donghee Na.

bpo-44904 : Fix bug in the doctest module that caused it to fail if a docstring included an example with a classmethod property . Patch by Alex Waygood.

bpo-45406 : Make inspect.getmodule() catch FileNotFoundError raised by inspect.getabsfile() , and return None to indicate that the module could not be determined.

bpo-45411 : Add extensions for files containing subtitles - .srt & .vtt - to the mimetypes.py module.

bpo-10716 : Migrated pydoc to HTML5 (without changing the look of it). Side effect is to update xmlrpc’s ServerHTMLDoc which now uses the CSS too. cgitb now relies less on pydoc (as it can’t use the CSS file).

bpo-27580 : Add support of null characters in csv .

bpo-45262 : Prevent use-after-free in asyncio. Make sure the cached running loop holder gets cleared on dealloc to prevent use-after-free in get_running_loop

bpo-45386 : Make xmlrpc.client more robust to C runtimes where the underlying C strftime function results in a ValueError when testing for year formatting options.

bpo-20028 : Improve error message of csv.Dialect when initializing. Patch by Vajrasky Kok and Donghee Na.

bpo-45343 : Update bundled pip to 21.2.4 and setuptools to 58.1.0

bpo-45328 : Fixed http.client.HTTPConnection to work properly in OSs that don’t support the TCP_NODELAY socket option.

bpo-45243 : Add setlimit() and getlimit() to sqlite3.Connection for setting and getting SQLite limits by connection basis. Patch by Erlend E. Aasland.

bpo-45320 : Removed from the inspect module:

use inspect.signature() or inspect.getfullargspec() instead.

the formatargspec function, deprecated since Python 3.5; use the inspect.signature() function and Signature object directly.

the undocumented Signature.from_builtin and Signature.from_function functions, deprecated since Python 3.5; use the Signature.from_callable() method instead.

bpo-45192 : Fix the tempfile._infer_return_type function so that the dir argument of the tempfile functions accepts an object implementing the os.PathLike protocol.

Patch by Kyungmin Lee.

bpo-45160 : When tracing a tkinter variable used by a ttk OptionMenu, callbacks are no longer made twice.

bpo-25625 : Added non parallel-safe chdir() context manager to change the current working directory and then restore it on exit. Simple wrapper around chdir() .

bpo-24139 : Add support for SQLite extended result codes in sqlite3.Error . Patch by Erlend E. Aasland.

bpo-24444 : Fixed an error raised in argparse help display when help for an option is set to 1+ blank spaces or when choices arg is an empty container.

bpo-44547 : Implement Fraction.__int__ , so that a fractions.Fraction instance f passes an isinstance(f, typing.SupportsInt) check.

bpo-40321 : Adds support for HTTP 308 redirects to urllib . See RFC 7538 for details. Patch by Jochem Schulenklopper.

bpo-41374 : Ensure that socket.TCP_* constants are exposed on Cygwin 3.1.6 and greater.

bpo-35970 : Add help flag to the base64 module’s command line interface. Patch contributed by Robert Kuska.

bpo-45726 : Improve documentation for functools.singledispatch() and functools.singledispatchmethod .

bpo-45680 : Amend the docs on GenericAlias objects to clarify that non-container classes can also implement __class_getitem__ . Patch contributed by Alex Waygood.

bpo-45618 : Update Sphinx version used to build the documentation to 4.2.0. Patch by Maciej Olko.

bpo-45655 : Add a new “relevant PEPs” section to the top of the documentation for the typing module. Patch by Alex Waygood.

bpo-45604 : Add level argument to multiprocessing.log_to_stderr function docs.

bpo-45516 : Add protocol description to the importlib.abc.TraversableResources documentation.

bpo-45464 : Mention in the documentation of Built-in Exceptions that inheriting from multiple exception types in a single subclass is not recommended due to possible memory layout incompatibility.

bpo-45449 : Add note about PEP 585 in collections.abc .

bpo-45516 : Add protocol description to the importlib.abc.Traversable documentation.

bpo-20692 : Add Programming FAQ entry explaining that int literal attribute access requires either a space after or parentheses around the literal.

bpo-45678 : Add tests for scenarios in which functools.singledispatchmethod is stacked on top of a method that has already been wrapped by two other decorators. Patch by Alex Waygood.

bpo-45578 : Add tests for dis.distb()

bpo-45678 : Add tests to ensure that functools.singledispatchmethod correctly wraps the attributes of the target function.

bpo-45668 : PGO tests now pass when Python is built without test extension modules.

bpo-45577 : Add subtests for all pickle protocols in test_zoneinfo .

bpo-45566 : Fix test_frozen_pickle in test_dataclasses to check all pickle versions.

bpo-43592 : test.libregrtest now raises the soft resource limit for the maximum number of file descriptors when the default is too low for our test suite as was often the case on macOS.

bpo-39679 : Add more test cases for @functools.singledispatchmethod when combined with @classmethod or @staticmethod .

bpo-45410 : When libregrtest spawns a worker process, stderr is now written into stdout to keep messages order. Use a single pipe for stdout and stderr, rather than two pipes. Previously, messages were out of order which made analysis of buildbot logs harder Patch by Victor Stinner.

bpo-45402 : Fix test_tools.test_sundry() when Python is built out of tree: fix how the freeze_modules.py tool locates the _freeze_module program. Patch by Victor Stinner.

bpo-45403 : Fix test_sys.test_stdlib_dir() when Python is built outside the source tree: compare normalized paths. Patch by Victor Stinner.

bpo-45400 : Fix test_name_error_suggestions_do_not_trigger_for_too_many_locals() of test_exceptions if a directory name contains “a1” (like “Python-3.11.0a1”): use a stricter regular expression. Patch by Victor Stinner.

bpo-10572 : Rename sqlite3 tests from test_sqlite to test_sqlite3 , and relocate them to Lib/test/test_sqlite3 . Patch by Erlend E. Aasland.

bpo-43158 : setup.py now uses values from configure script to build the _uuid extension module. Configure now detects util-linux’s libuuid , too.

bpo-45666 : Fix warning of swprintf and %s usage in _testembed.c

bpo-45548 : Modules/Setup and Modules/makesetup have been improved. The Setup file now contains working rules for all extensions. Outdated comments have been removed. Rules defined by makesetup track dependencies correctly.

bpo-45548 : The math and cmath implementation now require a C99 compatible libm and no longer ship with workarounds for missing acosh, asinh, atanh, expm1, and log1p functions.

bpo-45595 : setup.py and makesetup now track build dependencies on all Python header files and module specific header files.

bpo-45571 : Modules/Setup now use PY_CFLAGS_NODIST instead of PY_CFLAGS to compile shared modules.

bpo-45570 : pyexpat and _elementtree no longer define obsolete macros HAVE_EXPAT_CONFIG_H and USE_PYEXPAT_CAPI . XML_POOR_ENTROPY is now defined in expat_config.h .

bpo-43974 : setup.py no longer defines Py_BUILD_CORE_MODULE . Instead every module, that uses the internal API, defines the macro.

bpo-45548 : Fill in missing entries in Modules/Setup.

bpo-45532 : Update sys.version to use main as fallback information. Patch by Jeong YunWon.

bpo-45536 : The configure script now checks whether OpenSSL headers and libraries provide required APIs. Most common APIs are verified. The check detects outdated or missing OpenSSL. Failures do not stop configure.

bpo-45221 : Fixed regression in handling of LDFLAGS and CPPFLAGS options where argparse.parse_known_args() could interpret an option as one of the built-in command line argument, for example -h for help.

bpo-45440 : Building Python now requires a C99 <math.h> header file providing the following functions: copysign() , hypot() , isfinite() , isinf() , isnan() , round() . Patch by Victor Stinner.

bpo-45405 : Prevent internal configure error when running configure with recent versions of non-Apple clang. Patch by David Bohman.

bpo-45433 : Avoid linking libpython with libcrypt.

bpo-43652 : Update Tcl/Tk to 8.6.11, actually this time. The previous update incorrectly included 8.6.10.

bpo-45337 : venv now warns when the created environment may need to be accessed at a different path, due to redirections, links or junctions. It also now correctly installs or upgrades components when the alternate path is required.

bpo-43851 : Build SQLite SQLITE_OMIT_AUTOINIT on Windows. Patch by Erlend E. Aasland.

bpo-44828 : Avoid tkinter file dialog failure on macOS 12 Monterey when using the Tk 8.6.11 provided by python.org macOS installers. Patch by Marc Culler of the Tk project.

bpo-45495 : Add context keywords ‘case’ and ‘match’ to completions list.

bpo-29103 : PyType_FromSpec* now copies the class name from the spec to a buffer owned by the class, so the original can be safely deallocated. Patch by Petr Viktorin.

bpo-45522 : The internal freelists for frame, float, list, dict, async generators, and context objects can now be disabled.

bpo-35134 : Exclude PyWeakref_GET_OBJECT() from the limited C API. It never worked since the PyWeakReference structure is opaque in the limited C API.

bpo-35081 : Move the interpreteridobject.h header file from Include/ to Include/internal/ . It only provides private functions. Patch by Victor Stinner.

bpo-35134 : The non-limited API files cellobject.h , classobject.h , context.h , funcobject.h , genobject.h and longintrepr.h have been moved to the Include/cpython directory. Moreover, the eval.h header file was removed. These files must not be included directly, as they are already included in Python.h : Include Files . If they have been included directly, consider including Python.h instead. Patch by Victor Stinner.

bpo-45474 : The following items are no longer available when Py_LIMITED_API is defined:

PyMarshal_WriteLongToFile()

PyMarshal_WriteObjectToFile()

PyMarshal_ReadObjectFromString()

PyMarshal_WriteObjectToString()

the Py_MARSHAL_VERSION macro

These are not part of the limited API .

bpo-45434 : Remove the pystrhex.h header file. It only contains private functions. C extensions should only include the main <Python.h> header file. Patch by Victor Stinner.

bpo-45440 : Remove the Py_FORCE_DOUBLE() macro. It was used by the Py_IS_INFINITY() macro. Patch by Victor Stinner.

bpo-45434 : <Python.h> no longer includes the header files <stdlib.h> , <stdio.h> , <errno.h> and <string.h> when the Py_LIMITED_API macro is set to 0x030b0000 (Python 3.11) or higher. C extensions should explicitly include the header files after #include <Python.h> . Patch by Victor Stinner.

bpo-41123 : Remove Py_UNICODE_COPY() and Py_UNICODE_FILL() macros, deprecated since Python 3.3. Use PyUnicode_CopyCharacters() or memcpy() ( wchar_t* string), and PyUnicode_Fill() functions instead. Patch by Victor Stinner.

bpo-45412 : Remove the following math macros using the errno variable:

Py_ADJUST_ERANGE1()

Py_ADJUST_ERANGE2()

Py_OVERFLOWED()

Py_SET_ERANGE_IF_OVERFLOW()

Py_SET_ERRNO_ON_MATH_ERROR()

bpo-45395 : Custom frozen modules (the array set to PyImport_FrozenModules ) are now treated as additions, rather than replacing all the default frozen modules. Frozen stdlib modules can still be disabled by setting the “code” field of the custom array entry to NULL.

bpo-43760 : Add new PyThreadState_EnterTracing() , and PyThreadState_LeaveTracing() functions to the limited C API to suspend and resume tracing and profiling. Patch by Victor Stinner.

bpo-44220 : PyStructSequence_UnnamedField is added to the Stable ABI.

Python 3.11.0 alpha 1 ¶

Release date: 2021-10-05

bpo-42278 : Replaced usage of tempfile.mktemp() with TemporaryDirectory to avoid a potential race condition.

bpo-44600 : Fix incorrect line numbers while tracing some failed patterns in match statements. Patch by Charles Burkland.

bpo-41180 : Add auditing events to the marshal module, and stop raising code.__init__ events for every unmarshalled code object. Directly instantiated code objects will continue to raise an event, and audit event handlers should inspect or collect the raw marshal data. This reduces a significant performance overhead when loading from .pyc files.

bpo-44394 : Update the vendored copy of libexpat to 2.4.1 (from 2.2.8) to get the fix for the CVE-2013-0340 “Billion Laughs” vulnerability. This copy is most used on Windows and macOS.

bpo-43124 : Made the internal putcmd function in smtplib sanitize input for presence of \r and \n characters to avoid (unlikely) command injection.

bpo-44022 : http.client now avoids infinitely reading potential HTTP headers after a 100 Continue status response from the server.

bpo-43760 : The number of hardware branches per instruction dispatch is reduced from two to one by adding a special instruction for tracing. Patch by Mark Shannon.

bpo-45061 : Add a deallocator to the bool type to detect refcount bugs in C extensions which call Py_DECREF(Py_True) or Py_DECREF(Py_False) by mistake. Detect also refcount bugs when the empty tuple singleton or the Unicode empty string singleton is destroyed by mistake. Patch by Victor Stinner.

bpo-24076 : sum() was further optimised for summing up single digit integers.

bpo-45190 : Update Unicode databases to Unicode 14.0.0.

bpo-45167 : Fix deepcopying of types.GenericAlias objects.

bpo-45155 : int.to_bytes() and int.from_bytes() now take a default value of "big" for the byteorder argument. int.to_bytes() also takes a default value of 1 for the length argument.

bpo-44219 : Release the GIL while performing isatty system calls on arbitrary file descriptors. In particular, this affects os.isatty() , os.device_encoding() and io.TextIOWrapper . By extension, io.open() in text mode is also affected. This change solves a deadlock in os.isatty() . Patch by Vincent Michel in bpo-44219 .

bpo-44959 : Added fallback to extension modules with ‘.sl’ suffix on HP-UX

bpo-45121 : Fix issue where Protocol.__init__ raises RecursionError when it’s called directly or via super() . Patch provided by Yurii Karabas.

bpo-44348 : The deallocator function of the BaseException type now uses the trashcan mechanism to prevent stack overflow. For example, when a RecursionError instance is raised, it can be linked to another RecursionError through the __context__ attribute or the __traceback__ attribute, and then a chain of exceptions is created. When the chain is destroyed, nested deallocator function calls can crash with a stack overflow if the chain is too long compared to the available stack memory. Patch by Victor Stinner.

bpo-45123 : Fix PyAiter_Check to only check for the __anext__ presence (not for __aiter__). Rename PyAiter_Check to PyAIter_Check, PyObject_GetAiter -> PyObject_GetAIter.

bpo-1514420 : Interpreter no longer attempts to open files with names in angle brackets (like “<string>” or “<stdin>”) when formatting an exception.

bpo-41031 : Match C and Python code formatting of unprintable exceptions and exceptions in the __main__ module.

bpo-37330 : open() , io.open() , codecs.open() and fileinput.FileInput no longer accept 'U' (“universal newline”) in the file mode. This flag was deprecated since Python 3.3. Patch by Victor Stinner.

bpo-45083 : When the interpreter renders an exception, its name now has a complete qualname. Previously only the class name was concatenated to the module name, which sometimes resulted in an incorrect full name being displayed.

(This issue impacted only the C code exception rendering, the traceback module was using qualname already).

bpo-34561 : List sorting now uses the merge-ordering strategy from Munro and Wild’s powersort() . Unlike the former strategy, this is provably near-optimal in the entropy of the distribution of run lengths. Most uses of list.sort() probably won’t see a significant time difference, but may see significant improvements in cases where the former strategy was exceptionally poor. However, as these are all fast linear-time approximations to a problem that’s inherently at best quadratic-time to solve truly optimally, it’s also possible to contrive cases where the former strategy did better.

bpo-45056 : Compiler now removes trailing unused constants from co_consts.

bpo-45020 : Add a new command line option, “-X frozen_modules=[on|off]” to opt out of (or into) using optional frozen modules. This defaults to “on” (or “off” if it’s running out of the source tree).

bpo-45012 : In posix , release GIL during stat() , lstat() , and fstatat() syscalls made by os.DirEntry.stat() . Patch by Stanisław Skonieczny.

bpo-45018 : Fixed pickling of range iterators that iterated for over 2**32 times.

bpo-45000 : A SyntaxError is now raised when trying to delete __debug__ . Patch by Donghee Na.

bpo-44963 : Implement send() and throw() methods for anext_awaitable objects. Patch by Pablo Galindo.

bpo-44962 : Fix a race in WeakKeyDictionary, WeakValueDictionary and WeakSet when two threads attempt to commit the last pending removal. This fixes asyncio.create_task and fixes a data loss in asyncio.run where shutdown_asyncgens is not run

bpo-24234 : Implement the __bytes__() special method on the bytes type, so a bytes object b passes an isinstance(b, typing.SupportsBytes) check.

bpo-24234 : Implement the __complex__() special method on the complex type, so a complex number z passes an isinstance(z, typing.SupportsComplex) check.

bpo-44954 : Fixed a corner case bug where the result of float.fromhex('0x.8p-1074') was rounded the wrong way.

bpo-44947 : Refine the syntax error for trailing commas in import statements. Patch by Pablo Galindo.

bpo-44945 : Specialize the BINARY_ADD instruction using the PEP 659 machinery. Adds five new instructions:

BINARY_ADD_ADAPTIVE

BINARY_ADD_FLOAT

BINARY_ADD_INT

BINARY_ADD_UNICODE

BINARY_ADD_UNICODE_INPLACE_FAST

bpo-44929 : Fix some edge cases of enum.Flag string representation in the REPL. Patch by Pablo Galindo.

bpo-44914 : Class version tags are no longer recycled.

This means that a version tag serves as a unique identifier for the state of a class. We rely on this for effective specialization of the LOAD_ATTR and other instructions.

bpo-44698 : Restore behaviour of complex exponentiation with integer-valued exponent of type float or complex .

bpo-44895 : A debug variable PYTHONDUMPREFSFILE is added for creating a dump file which is generated by --with-trace-refs . Patch by Donghee Na.

bpo-44900 : Add five superinstructions for PEP 659 quickening:

LOAD_FAST LOAD_FAST

STORE_FAST LOAD_FAST

LOAD_FAST LOAD_CONST

LOAD_CONST LOAD_FAST

STORE_FAST STORE_FAST

bpo-44889 : Initial implementation of adaptive specialization of LOAD_METHOD . The following specialized forms were added:

LOAD_METHOD_CACHED

LOAD_METHOD_MODULE

LOAD_METHOD_CLASS

bpo-44890 : Specialization stats are always collected in debug builds.

bpo-44885 : Correct the ast locations of f-strings with format specs and repeated expressions. Patch by Pablo Galindo

bpo-44878 : Remove the loop from the bytecode interpreter. All instructions end with a DISPATCH macro, so the loop is now redundant.

bpo-44878 : Remove switch statement for interpreter loop when using computed gotos. This makes sure that we only have one dispatch table in the interpreter.

bpo-44874 : Deprecate the old trashcan macros ( Py_TRASHCAN_SAFE_BEGIN / Py_TRASHCAN_SAFE_END ). They should be replaced by the new macros Py_TRASHCAN_BEGIN and Py_TRASHCAN_END .

bpo-44872 : Use new trashcan macros (Py_TRASHCAN_BEGIN/END) in frameobject.c instead of the old ones (Py_TRASHCAN_SAFE_BEGIN/END).

bpo-33930 : Fix segmentation fault with deep recursion when cleaning method objects. Patch by Augusto Goulart and Pablo Galindo.

bpo-25782 : Fix bug where PyErr_SetObject hangs when the current exception has a cycle in its context chain.

bpo-44856 : Fix reference leaks in the error paths of update_bases() and __build_class__ . Patch by Pablo Galindo.

bpo-44826 : Initial implementation of adaptive specialization of STORE_ATTR

Three specialized forms of STORE_ATTR are added:

STORE_ATTR_SLOT

STORE_ATTR_SPLIT_KEYS

STORE_ATTR_WITH_HINT

bpo-44838 : Fixed a bug that was causing the parser to raise an incorrect custom SyntaxError for invalid ‘if’ expressions. Patch by Pablo Galindo.

bpo-44821 : Create instance dictionaries (__dict__) eagerly, to improve regularity of object layout and assist specialization.

bpo-44792 : Improve syntax errors for if expressions. Patch by Miguel Brito

bpo-34013 : Generalize the invalid legacy statement custom error message (like the one generated when “print” is called without parentheses) to include more generic expressions. Patch by Pablo Galindo

bpo-44732 : Rename types.Union to types.UnionType .

bpo-44725 : Expose specialization stats in python via _opcode.get_specialization_stats() .

bpo-44717 : Improve AttributeError on circular imports of submodules.

bpo-44698 : Fix undefined behaviour in complex object exponentiation.

bpo-44653 : Support typing types in parameter substitution in the union type.

bpo-44676 : Add ability to serialise types.Union objects. Patch provided by Yurii Karabas.

bpo-44633 : Parameter substitution of the union type with wrong types now raises TypeError instead of returning NotImplemented .

bpo-44661 : Update property_descr_set to use vectorcall if possible. Patch by Donghee Na.

bpo-44662 : Add __module__ to types.Union . This also fixes types.Union issues with typing.Annotated . Patch provided by Yurii Karabas.

bpo-44655 : Include the name of the type in unset __slots__ attribute errors. Patch by Pablo Galindo

bpo-44655 : Don’t include a missing attribute with the same name as the failing one when offering suggestions for missing attributes. Patch by Pablo Galindo

bpo-44646 : Fix the hash of the union type: it no longer depends on the order of arguments.

bpo-44636 : Collapse union of equal types. E.g. the result of int | int is now int . Fix comparison of the union type with non-hashable objects. E.g. int | str == {} no longer raises a TypeError.

bpo-44611 : On Windows, os.urandom() : uses BCryptGenRandom API instead of CryptGenRandom API which is deprecated from Microsoft Windows API. Patch by Donghee Na.

bpo-44635 : Convert None to type(None) in the union type constructor.

bpo-26280 : Implement adaptive specialization for BINARY_SUBSCR

Three specialized forms of BINARY_SUBSCR are added:

BINARY_SUBSCR_LIST_INT

BINARY_SUBSCR_TUPLE_INT

BINARY_SUBSCR_DICT

bpo-44589 : Mapping patterns in match statements with two or more equal literal keys will now raise a SyntaxError at compile-time.

bpo-44606 : Fix __instancecheck__ and __subclasscheck__ for the union type.

bpo-42073 : The @classmethod decorator can now wrap other classmethod-like descriptors.

bpo-41972 : Tuned the string-searching algorithm of fastsearch.h to have a shorter inner loop for most cases.

bpo-44590 : All necessary data for executing a Python function (local variables, stack, etc) is now kept in a per-thread stack. Frame objects are lazily allocated on demand. This increases performance by about 7% on the standard benchmark suite. Introspection and debugging are unaffected as frame objects are always available when needed. Patch by Mark Shannon.

bpo-44584 : The threading debug ( PYTHONTHREADDEBUG environment variable) is deprecated in Python 3.10 and will be removed in Python 3.12. This feature requires a debug build of Python. Patch by Victor Stinner.

bpo-43895 : An obsolete internal cache of shared object file handles added in 1995 that attempted, but did not guarantee, that a .so would not be dlopen’ed twice to work around flaws in mid-1990s posix-ish operating systems has been removed from dynload_shlib.c.

bpo-44490 : typing now searches for type parameters in types.Union objects. get_type_hints will also properly resolve annotations with nested types.Union objects. Patch provided by Yurii Karabas.

bpo-43950 : Code objects can now provide the column information for instructions when available. This is levaraged during traceback printing to show the expressions responsible for errors.

Contributed by Pablo Galindo, Batuhan Taskaya and Ammar Askar as part of PEP 657 .

bpo-44562 : Remove uses of PyObject_GC_Del() in error path when initializing types.GenericAlias .

bpo-41486 : Fix a memory consumption and copying performance regression in earlier 3.10 beta releases if someone used an output buffer larger than 4GiB with zlib.decompress on input data that expands that large.

bpo-43908 : Heap types with the Py_TPFLAGS_IMMUTABLETYPE flag can now inherit the PEP 590 vectorcall protocol. Previously, this was only possible for static types . Patch by Erlend E. Aasland.

bpo-44553 : Implement GC methods for types.Union to break reference cycles and prevent memory leaks.

bpo-44490 : Add __parameters__ attribute and __getitem__ operator to types.Union . Patch provided by Yurii Karabas.

bpo-44523 : Remove the pass-through for hash() of weakref.proxy objects to prevent unintended consequences when the original referred object dies while the proxy is part of a hashable object. Patch by Pablo Galindo.

bpo-44483 : Fix a crash in types.Union objects when creating a union of an object with bad __module__ field.

bpo-44486 : Modules will always have a dictionary, even when created by types.ModuleType.__new__()

bpo-44472 : Fix ltrace functionality when exceptions are raised. Patch by Pablo Galindo

bpo-12022 : A TypeError is now raised instead of an AttributeError in with and async with statements for objects which do not support the context manager or asynchronous context manager protocols correspondingly.

bpo-44297 : Make sure that the line number is set when entering a comprehension scope. This ensures that backtraces including generator expressions show the correct line number.

bpo-44456 : Improve the syntax error when mixing positional and keyword patterns. Patch by Pablo Galindo.

bpo-44409 : Fix error location information for tokenizer errors raised on initialization of the tokenizer. Patch by Pablo Galindo.

bpo-44396 : Fix a possible crash in the tokenizer when raising syntax errors for unclosed strings. Patch by Pablo Galindo.

bpo-44376 : Exact integer exponentiation (like i**2 or pow(i, 2) ) with a small exponent is much faster, due to reducing overhead in such cases.

bpo-44313 : Directly imported objects and modules (through import and from import statements) don’t generate LOAD_METHOD / CALL_METHOD for directly accessed objects on their namespace. They now use the regular LOAD_ATTR / CALL_FUNCTION .

bpo-44338 : Implement adaptive specialization for LOAD_GLOBAL

Two specialized forms of LOAD_GLOBAL are added:

LOAD_GLOBAL_MODULE

LOAD_GLOBAL_BUILTIN

bpo-44368 : Improve syntax errors for invalid “as” targets. Patch by Pablo Galindo

bpo-44349 : Fix an edge case when displaying text from files with encoding in syntax errors. Patch by Pablo Galindo.

bpo-44337 : Initial implementation of adaptive specialization of LOAD_ATTR

Four specialized forms of LOAD_ATTR are added:

LOAD_ATTR_SLOT

LOAD_ATTR_SPLIT_KEYS

LOAD_ATTR_WITH_HINT

LOAD_ATTR_MODULE

bpo-44335 : Fix a regression when identifying incorrect characters in syntax errors. Patch by Pablo Galindo

bpo-43693 : Computation of the offsets of cell variables is done in the compiler instead of at runtime. This reduces the overhead of handling cell and free variables, especially in the case where a variable is both an argument and cell variable.

bpo-44317 : Improve tokenizer error with improved locations. Patch by Pablo Galindo.

bpo-44304 : Fix a crash in the sqlite3 module that happened when the garbage collector clears sqlite.Statement objects. Patch by Pablo Galindo

bpo-44305 : Improve error message for try blocks without except or finally blocks. Patch by Pablo Galindo.

bpo-43413 : Constructors of subclasses of some builtin classes (e.g. tuple , list , frozenset ) no longer accept arbitrary keyword arguments. [reverted in 3.11a4] Subclass of set can now define a __new__() method with additional keyword parameters without overriding also __init__() .

bpo-43667 : Improve Unicode support in non-UTF locales on Oracle Solaris. This issue does not affect other Solaris systems.

bpo-43693 : A new opcode MAKE_CELL has been added that effectively moves some of the work done on function entry into the compiler and into the eval loop. In addition to creating the required cell objects, the new opcode converts relevant arguments (and other locals) to cell variables on function entry.

bpo-44232 : Fix a regression in type() when a metaclass raises an exception. The C function type_new() must properly report the exception when a metaclass constructor raises an exception and the winner class is not the metaclass. Patch by Victor Stinner.

bpo-44201 : Avoid side effects of checking for specialized syntax errors in the REPL that was causing it to ask for extra tokens after a syntax error had been detected. Patch by Pablo Galindo

bpo-43693 : PyCodeObject gained co_fastlocalnames and co_fastlocalkinds as the authoritative source of fast locals info. Marshaled code objects have changed accordingly.

bpo-44184 : Fix a crash at Python exit when a deallocator function removes the last strong reference to a heap type. Patch by Victor Stinner.

bpo-44187 : Implement quickening in the interpreter. This offers no advantages as yet, but is an enabler of future optimizations. See PEP 659 for full explanation.

bpo-44180 : The parser doesn’t report generic syntax errors that happen in a position further away that the one it reached in the first pass. Patch by Pablo Galindo

bpo-44168 : Fix error message in the parser involving keyword arguments with invalid expressions. Patch by Pablo Galindo

bpo-44156 : String caches in compile.c are now subinterpreter compatible.

bpo-44143 : Fixed a crash in the parser that manifest when raising tokenizer errors when an existing exception was present. Patch by Pablo Galindo.

bpo-44032 : Move ‘fast’ locals and other variables from the frame object to a per-thread datastack.

bpo-44114 : Fix incorrect dictkeys_reversed and dictitems_reversed function signatures in C code, which broke webassembly builds.

bpo-44110 : Improve str.__getitem__() error message

bpo-26110 : Add CALL_METHOD_KW opcode to speed up method calls with keyword arguments. Idea originated from PyPy. A side effect is executing CALL_METHOD is now branchless in the evaluation loop.

bpo-28307 : Compiler now optimizes simple C-style formatting with literal format containing only format codes %s, %r and %a by converting them to f-string expressions.

bpo-43149 : Correct the syntax error message regarding multiple exception types to not refer to “exception groups”. Patch by Pablo Galindo

bpo-43822 : The parser will prioritize tokenizer errors over custom syntax errors when raising exceptions. Patch by Pablo Galindo.

bpo-40222 : “Zero cost” exception handling.

Uses a lookup table to determine how to handle exceptions.

Removes SETUP_FINALLY and POP_TOP block instructions, eliminating the runtime overhead of try statements.

Reduces the size of the frame object by about 60%.

Patch by Mark Shannon

bpo-43918 : Document the signature and default argument in the docstring of the new anext builtin.

bpo-43833 : Emit a deprecation warning if the numeric literal is immediately followed by one of keywords: and, else, for, if, in, is, or. Raise a syntax error with more informative message if it is immediately followed by other keyword or identifier.

bpo-43879 : Add native_thread_id to PyThreadState. Patch by Gabriele N. Tornetta.

bpo-43693 : Compute cell offsets relative to locals in compiler. Allows the interpreter to treats locals and cells a single array, which is slightly more efficient. Also make the LOAD_CLOSURE opcode an alias for LOAD_FAST. Preserving LOAD_CLOSURE helps keep bytecode a bit more readable.

bpo-17792 : More accurate error messages for access of unbound locals or free vars.

bpo-28146 : Fix a confusing error message in str.format() .

bpo-11105 : When compiling ast.AST objects with recursive references through compile() , the interpreter doesn’t crash anymore instead it raises a RecursionError .

bpo-39091 : Fix crash when using passing a non-exception to a generator’s throw() method. Patch by Noah Oxer

bpo-33346 : Asynchronous comprehensions are now allowed inside comprehensions in asynchronous functions. Outer comprehensions implicitly become asynchronous.

bpo-45371 : Fix clang rpath issue in distutils . The UnixCCompiler now uses correct clang option to add a runtime library directory (rpath) to a shared library.

bpo-45329 : Fix freed memory access in pyexpat.xmlparser when building it with an installed expat library <= 2.2.0.

bpo-41710 : On Unix, if the sem_clockwait() function is available in the C library (glibc 2.30 and newer), the threading.Lock.acquire() method now uses the monotonic clock ( time.CLOCK_MONOTONIC ) for the timeout, rather than using the system clock ( time.CLOCK_REALTIME ), to not be affected by system clock changes. Patch by Victor Stinner.

bpo-1596321 : Fix the threading._shutdown() function when the threading module was imported first from a thread different than the main thread: no longer log an error at Python exit.

bpo-45274 : Fix a race condition in the Thread.join() method of the threading module. If the function is interrupted by a signal and the signal handler raises an exception, make sure that the thread remains in a consistent state to prevent a deadlock. Patch by Victor Stinner.

bpo-21302 : In Unix operating systems, time.sleep() now uses the nanosleep() function, if clock_nanosleep() is not available but nanosleep() is available. nanosleep() allows to sleep with nanosecond precision.

bpo-21302 : On Windows, time.sleep() now uses a waitable timer which has a resolution of 100 nanoseconds (10 -7 seconds). Previously, it had a resolution of 1 millisecond (10 -3 seconds). Patch by Benjamin Szőke and Victor Stinner.

bpo-45238 : Fix unittest.IsolatedAsyncioTestCase.debug() : it runs now asynchronous methods and callbacks.

bpo-36674 : unittest.TestCase.debug() raises now a unittest.SkipTest if the class or the test method are decorated with the skipping decorator.

bpo-45235 : Fix an issue where argparse would not preserve values in a provided namespace when using a subparser with defaults.

bpo-45183 : Have zipimport.zipimporter.find_spec() not raise an exception when the underlying zip file has been deleted and the internal cache has been reset via invalidate_cache().

bpo-45234 : Fixed a regression in copyfile() , copy() , copy2() raising FileNotFoundError when source is a directory, which should raise IsADirectoryError

bpo-45228 : Fix stack buffer overflow in parsing J1939 network address.

bpo-45225 : use map function instead of genexpr in capwords.

bpo-42135 : Fix typo: importlib.find_loader is really slated for removal in Python 3.12 not 3.10, like the others in PR 25169.

bpo-20524 : Improves error messages on .format() operation for str , float , int , and complex . New format now shows the problematic pattern and the object type.

bpo-45168 : Change dis.dis() output to omit op arg values that cannot be resolved due to co_consts , co_names etc not being provided. Previously the oparg itself was repeated in the value field, which is not useful and can be confusing.

bpo-21302 : In Unix operating systems, time.sleep() now uses the clock_nanosleep() function, if available, which allows to sleep for an interval specified with nanosecond precision.

bpo-45173 : Remove from the configparser module: the SafeConfigParser class, the filename property of the ParsingError class, the readfp() method of the ConfigParser class, deprecated since Python 3.2.

bpo-44987 : Pure ASCII strings are now normalized in constant time by unicodedata.normalize() . Patch by Donghee Na.

bpo-35474 : Calling mimetypes.guess_all_extensions() with strict=False no longer affects the result of the following call with strict=True . Also, mutating the returned list no longer affects the global state.

bpo-45166 : typing.get_type_hints() now works with Final wrapped in ForwardRef .

bpo-45162 : Remove many old deprecated unittest features:

“ fail* ” and “ assert* ” aliases of TestCase methods.

Broken from start TestCase method assertDictContainsSubset() .

Ignored <unittest.TestLoader.loadTestsFromModule> TestLoader.loadTestsFromModule() parameter use_load_tests .

Old alias _TextTestResult of TextTestResult .

bpo-38371 : Remove the deprecated split() method of _tkinter.TkappType . Patch by Erlend E. Aasland.

bpo-20499 : Improve the speed and accuracy of statistics.pvariance().

bpo-45132 : Remove __getitem__() methods of xml.dom.pulldom.DOMEventStream , wsgiref.util.FileWrapper and fileinput.FileInput , deprecated since Python 3.9.

bpo-45129 : Due to significant security concerns, the reuse_address parameter of asyncio.loop.create_datagram_endpoint() , disabled in Python 3.9, is now entirely removed. This is because of the behavior of the socket option SO_REUSEADDR in UDP.

bpo-45124 : The bdist_msi command, deprecated in Python 3.9, is now removed.

Use bdist_wheel (wheel packages) instead.

bpo-30856 : unittest.TestResult methods addFailure() , addError() , addSkip() and addSubTest() are now called immediately after raising an exception in test or finishing a subtest. Previously they were called only after finishing the test clean up.

bpo-45034 : Changes how error is formatted for struct.pack with 'H' and 'h' modes and too large / small numbers. Now it shows the actual numeric limits, while previously it was showing arithmetic expressions.

bpo-25894 : unittest now always reports skipped and failed subtests separately: separate characters in default mode and separate lines in verbose mode. Also the test description is now output for errors in test method, class and module cleanups.

bpo-45081 : Fix issue when dataclasses that inherit from typing.Protocol subclasses have wrong __init__ . Patch provided by Yurii Karabas.

bpo-45085 : The binhex module, deprecated in Python 3.9, is now removed. The following binascii functions, deprecated in Python 3.9, are now also removed:

a2b_hqx() , b2a_hqx() ;

rlecode_hqx() , rledecode_hqx() .

The binascii.crc_hqx() function remains available.

bpo-40360 : The lib2to3 package is now deprecated and may not be able to parse Python 3.10 or newer. See the PEP 617 (New PEG parser for CPython). Patch by Victor Stinner.

bpo-45075 : Rename traceback.StackSummary.format_frame() to traceback.StackSummary.format_frame_summary() . This method was added for 3.11 so it was not released yet.

Updated code and docs to better distinguish frame and FrameSummary.

bpo-31299 : Add option to completely drop frames from a traceback by returning None from a format_frame() override.

bpo-41620 : run() now always return a TestResult instance. Previously it returned None if the test class or method was decorated with a skipping decorator.

bpo-45021 : Fix a potential deadlock at shutdown of forked children when using concurrent.futures module

bpo-43913 : Fix bugs in cleaning up classes and modules in unittest :

Functions registered with addModuleCleanup() were not called unless the user defines tearDownModule() in their test module.

Functions registered with addClassCleanup() were not called if tearDownClass is set to None .

Buffering in TestResult did not work with functions registered with addClassCleanup() and addModuleCleanup() .

Errors in functions registered with addClassCleanup() and addModuleCleanup() were not handled correctly in buffered and debug modes.

Errors in setUpModule() and functions registered with addModuleCleanup() were reported in wrong order.

And several lesser bugs.

bpo-45030 : Fix integer overflow in pickling and copying the range iterator.

bpo-45001 : Made email date parsing more robust against malformed input, namely a whitespace-only Date: header. Patch by Wouter Bolsterlee.

bpo-45010 : Remove support of special method __div__ in unittest.mock . It is not used in Python 3.

bpo-39218 : Improve accuracy of variance calculations by using x*x instead of x**2 .

bpo-43613 : Improve the speed of gzip.compress() and gzip.decompress() by compressing and decompressing at once in memory instead of in a streamed fashion.

bpo-37596 : Ensure that set and frozenset objects are always marshalled reproducibly.

bpo-44019 : A new function operator.call has been added, such that operator.call(obj, *args, **kwargs) == obj(*args, **kwargs) .

bpo-42255 : webbrowser.MacOSX is deprecated and will be removed in Python 3.13. It is untested and undocumented and also not used by webbrowser itself. Patch by Donghee Na.

bpo-44955 : Method stopTestRun() is now always called in pair with method startTestRun() for TestResult objects implicitly created in run() . Previously it was not called for test methods and classes decorated with a skipping decorator.

bpo-39039 : tarfile.open raises ReadError when a zlib error occurs during file extraction.

bpo-44935 : subprocess on Solaris now also uses os.posix_spawn() for better performance.

bpo-44911 : IsolatedAsyncioTestCase will no longer throw an exception while cancelling leaked tasks. Patch by Bar Harel.

bpo-41322 : Added DeprecationWarning for tests and async tests that return a value!=None (as this may indicate an improperly written test, for example a test written as a generator function).

bpo-44524 : Make exception message more useful when subclass from typing special form alias. Patch provided by Yurii Karabas.

bpo-38956 : argparse.BooleanOptionalAction ’s default value is no longer printed twice when used with argparse.ArgumentDefaultsHelpFormatter .

bpo-44860 : Fix the posix_user scheme in sysconfig to not depend on sys.platlibdir .

bpo-44859 : Improve error handling in sqlite3 and raise more accurate exceptions.

MemoryError is now raised instead of sqlite3.Warning when memory is not enough for encoding a statement to UTF-8 in Connection.__call__() and Cursor.execute() .

UnicodEncodeError is now raised instead of sqlite3.Warning when the statement contains surrogate characters in Connection.__call__() and Cursor.execute() .

TypeError is now raised instead of ValueError for non-string script argument in Cursor.executescript() .

ValueError is now raised for script containing the null character instead of truncating it in Cursor.executescript() .

Correctly handle exceptions raised when getting boolean value of the result of the progress handler.

Add many tests covering different corner cases.

bpo-44581 : Upgrade bundled pip to 21.2.3 and setuptools to 57.4.0

bpo-44849 : Fix the os.set_inheritable() function on FreeBSD 14 for file descriptor opened with the O_PATH flag: ignore the EBADF error on ioctl() , fallback on the fcntl() implementation. Patch by Victor Stinner.

bpo-44605 : The @functools.total_ordering() decorator now works with metaclasses.

bpo-44524 : Fixed an issue wherein the __name__ and __qualname__ attributes of subscribed specialforms could be None .

bpo-44839 : MemoryError raised in user-defined functions will now produce a MemoryError in sqlite3 . OverflowError will now be converted to DataError . Previously OperationalError was produced in these cases.

bpo-44822 : sqlite3 user-defined functions and aggregators returning strings with embedded NUL characters are no longer truncated. Patch by Erlend E. Aasland.

bpo-44801 : Ensure that the ParamSpec variable in Callable can only be substituted with a parameters expression (a list of types, an ellipsis, ParamSpec or Concatenate).

bpo-44806 : Non-protocol subclasses of typing.Protocol ignore now the __init__ method inherited from protocol base classes.

bpo-27275 : collections.OrderedDict.popitem() and collections.OrderedDict.pop() no longer call __getitem__ and __delitem__ methods of the OrderedDict subclasses.

bpo-44793 : Fix checking the number of arguments when subscribe a generic type with ParamSpec parameter.

bpo-44784 : In importlib.metadata tests, override warnings behavior under expected DeprecationWarnings (importlib_metadata 4.6.3).

bpo-44667 : The tokenize.tokenize() doesn’t incorrectly generate a NEWLINE token if the source doesn’t end with a new line character but the last line is a comment, as the function is already generating a NL token. Patch by Pablo Galindo

bpo-44771 : Added importlib.simple module implementing adapters from a low-level resources reader interface to a TraversableResources interface. Legacy API ( path , contents , …) is now supported entirely by the .files() API with a compatibility shim supplied for resource loaders without that functionality. Feature parity with importlib_resources 5.2.

bpo-44752 : rcompleter does not call getattr() on property objects to avoid the side-effect of evaluating the corresponding method.

bpo-44747 : Refactor usage of sys._getframe in typing module. Patch provided by Yurii Karabas.

bpo-42378 : Fixes the issue with log file being overwritten when logging.FileHandler is used in atexit with filemode set to 'w' . Note this will cause the message in atexit not being logged if the log stream is already closed due to shutdown of logging.

bpo-44720 : weakref.proxy objects referencing non-iterators now raise TypeError rather than dereferencing the null tp_iternext slot and crashing.

bpo-44704 : The implementation of collections.abc.Set._hash() now matches that of frozenset.__hash__() .

bpo-44666 : Fixed issue in compileall.compile_file() when sys.stdout is redirected. Patch by Stefan Hölzl.

bpo-44688 : sqlite3.Connection.create_collation() now accepts non-ASCII collation names. Patch by Erlend E. Aasland.

bpo-44690 : Adopt binacii.a2b_base64 ’s strict mode in base64.b64decode .

bpo-42854 : Fixed a bug in the _ssl module that was throwing OverflowError when using _ssl._SSLSocket.write() and _ssl._SSLSocket.read() for a big value of the len parameter. Patch by Pablo Galindo.

bpo-44686 : Replace unittest.mock._importer with pkgutil.resolve_name .

bpo-44353 : Make NewType.__call__ faster by implementing it in C. Patch provided by Yurii Karabas.

bpo-44682 : Change the pdb commands directive to disallow setting commands for an invalid breakpoint and to display an appropriate error.

bpo-44353 : Refactor typing.NewType from function into callable class. Patch provided by Yurii Karabas.

bpo-44678 : Added a separate error message for discontinuous padding in binascii.a2b_base64 strict mode.

bpo-44524 : Add missing __name__ and __qualname__ attributes to typing module classes. Patch provided by Yurii Karabas.

bpo-40897 : Give priority to using the current class constructor in inspect.signature() . Patch by Weipeng Hong.

bpo-44638 : Add a reference to the zipp project and hint as to how to use it.

bpo-44648 : Fixed wrong error being thrown by inspect.getsource() when examining a class in the interactive session. Instead of TypeError , it should be OSError with appropriate error message.

bpo-44608 : Fix memory leak in _tkinter._flatten() if it is called with a sequence or set, but not list or tuple.

bpo-44594 : Fix an edge case of ExitStack and AsyncExitStack exception chaining. They will now match with block behavior when __context__ is explicitly set to None when the exception is in flight.

bpo-42799 : In fnmatch , the cache size for compiled regex patterns ( functools.lru_cache() ) was bumped up from 256 to 32768, affecting functions: fnmatch.fnmatch() , fnmatch.fnmatchcase() , fnmatch.filter() .

bpo-41928 : Update shutil.copyfile() to raise FileNotFoundError instead of confusing IsADirectoryError when a path ending with a os.path.sep does not exist; shutil.copy() and shutil.copy2() are also affected.

bpo-44569 : Added the StackSummary.format_frame() function in traceback . This allows users to customize the way individual lines are formatted in tracebacks without re-implementing logic to handle recursive tracebacks.

bpo-44566 : handle StopIteration subclass raised from @contextlib.contextmanager generator

bpo-44558 : Make the implementation consistency of indexOf() between C and Python versions. Patch by Donghee Na.

bpo-41249 : Fixes TypedDict to work with typing.get_type_hints() and postponed evaluation of annotations across modules.

bpo-44554 : Refactor argument processing in pdb.main() to simplify detection of errors in input loading and clarify behavior around module or script invocation.

bpo-34798 : Break up paragraph about pprint.PrettyPrinter construction parameters to make it easier to read.

bpo-44539 : Added support for recognizing JPEG files without JFIF or Exif markers.

bpo-44461 : Fix bug with pdb ’s handling of import error due to a package which does not have a __main__ module

bpo-43625 : Fix a bug in the detection of CSV file headers by csv.Sniffer.has_header() and improve documentation of same.

bpo-44516 : Update vendored pip to 21.1.3

bpo-42892 : Fixed an exception thrown while parsing a malformed multipart email by email.message.EmailMessage .

bpo-44468 : typing.get_type_hints() now finds annotations in classes and base classes with unexpected __module__ . Previously, it skipped those MRO elements.

bpo-44491 : Allow clearing the sqlite3 authorizer callback by passing None to set_authorizer() . Patch by Erlend E. Aasland.

bpo-43977 : Set the proper Py_TPFLAGS_MAPPING and Py_TPFLAGS_SEQUENCE flags for subclasses created before a parent has been registered as a collections.abc.Mapping or collections.abc.Sequence .

bpo-44482 : Fix very unlikely resource leak in glob in alternate Python implementations.

bpo-44466 : The faulthandler module now detects if a fatal error occurs during a garbage collector collection. Patch by Victor Stinner.

bpo-44471 : A TypeError is now raised instead of an AttributeError in contextlib.ExitStack.enter_context() and contextlib.AsyncExitStack.enter_async_context() for objects which do not support the context manager or asynchronous context manager protocols correspondingly.

bpo-44404 : tkinter ’s after() method now supports callables without the __name__ attribute.

bpo-41546 : Make pprint (like the builtin print ) not attempt to write to stdout when it is None .

bpo-44458 : BUFFER_BLOCK_SIZE is now declared static, to avoid linking collisions when bz2, lmza or zlib are statically linked.

bpo-44464 : Remove exception for flake8 in deprecated importlib.metadata interfaces. Sync with importlib_metadata 4.6.

bpo-44446 : Take into account that lineno might be None in traceback.FrameSummary .

bpo-44439 : Fix in bz2.BZ2File.write() / lzma.LZMAFile.write() methods, when the input data is an object that supports the buffer protocol, the file length may be wrong.

bpo-44434 : _thread.start_new_thread() no longer calls PyThread_exit_thread() explicitly at the thread exit, the call was redundant. On Linux with the glibc, pthread_exit() aborts the whole process if dlopen() fails to open libgcc_s.so file (ex: EMFILE error). Patch by Victor Stinner.

bpo-42972 : The _thread.RLock type now fully implement the GC protocol: add a traverse function and the Py_TPFLAGS_HAVE_GC flag. Patch by Victor Stinner.

bpo-44422 : The threading.enumerate() function now uses a reentrant lock to prevent a hang on reentrant call. Patch by Victor Stinner.

bpo-38291 : Importing typing.io or typing.re now prints a DeprecationWarning .

bpo-37880 : argparse actions store_const and append_const each receive a default value of None when the const kwarg is not provided. Previously, this raised a TypeError .

bpo-44389 : Fix deprecation of ssl.OP_NO_TLSv1_3

bpo-27827 : pathlib.PureWindowsPath.is_reserved() now identifies a greater range of reserved filenames, including those with trailing spaces or colons.

bpo-44395 : Fix as_string() to pass unixfrom properly. Patch by Donghee Na.

bpo-34266 : Handle exceptions from parsing the arg of pdb ’s run/restart command.

bpo-44362 : Improve ssl module’s deprecation messages, error reporting, and documentation for deprecations.

bpo-44342 : [Enum] Change pickling from by-value to by-name.

bpo-44356 : [Enum] Allow multiple data-type mixins if they are all the same.

bpo-44351 : Restore back parse_makefile() in distutils.sysconfig because it behaves differently than the similar implementation in sysconfig .

bpo-35800 : smtpd.MailmanProxy is now removed as it is unusable without an external module, mailman . Patch by Donghee Na.

bpo-44357 : Added a function that returns cube root of the given number math.cbrt()

bpo-44339 : Change math.pow(±0.0, -math.inf) to return inf instead of raising ValueError . This brings the special-case handling of math.pow into compliance with the IEEE 754 standard.

bpo-44242 : Remove missing flag check from Enum creation and move into a verify decorator.

bpo-44246 : In importlib.metadata , restore compatibility in the result from Distribution.entry_points ( EntryPoints ) to honor expectations in older implementations and issuing deprecation warnings for these cases: A. EntryPoints objects are once again mutable, allowing for sort() and other list-based mutation operations. Avoid deprecation warnings by casting to a mutable sequence (e.g. list(dist.entry_points).sort() ). B. EntryPoints results once again allow for access by index. To avoid deprecation warnings, cast the result to a Sequence first (e.g. tuple(dist.entry_points)[0] ).

bpo-44246 : In importlib.metadata.entry_points, de-duplication of distributions no longer requires loading the full metadata for PathDistribution objects, improving entry point loading performance by ~10x.

bpo-43858 : Added a function that returns a copy of a dict of logging levels: logging.getLevelNamesMapping()

bpo-44260 : The random.Random constructor no longer reads system entropy without need.

bpo-44254 : On Mac, give turtledemo button text a color that works on both light or dark background. Programmers cannot control the latter.

bpo-44258 : Support PEP 515 for Fraction’s initialization from string.

bpo-44235 : Remove deprecated functions in the gettext . Patch by Donghee Na.

bpo-38693 : Prefer f-strings to .format in importlib.resources.

bpo-33693 : Importlib.metadata now prefers f-strings to .format.

bpo-44241 : Incorporate minor tweaks from importlib_metadata 4.1: SimplePath protocol, support for Metadata 2.2.

bpo-43216 : Remove the @asyncio.coroutine decorator enabling legacy generator-based coroutines to be compatible with async/await code; remove asyncio.coroutines.CoroWrapper used for wrapping legacy coroutine objects in the debug mode. The decorator has been deprecated since Python 3.8 and the removal was initially scheduled for Python 3.10. Patch by Illia Volochii.

bpo-44210 : Make importlib.metadata._meta.PackageMetadata public.

bpo-43643 : Declare readers.MultiplexedPath.name as a property per the spec.

bpo-27334 : The sqlite3 context manager now performs a rollback (thus releasing the database lock) if commit failed. Patch by Luca Citi and Erlend E. Aasland.

bpo-4928 : Documented existing behavior on POSIX: NamedTemporaryFiles are not deleted when creating process is killed with SIGKILL

bpo-44154 : Optimize fractions.Fraction pickling for large components.

bpo-33433 : For IPv4 mapped IPv6 addresses ( RFC 4291 Section 2.5.5.2), the ipaddress.IPv6Address.is_private check is deferred to the mapped IPv4 address. This solves a bug where public mapped IPv4 addresses were considered private by the IPv6 check.

bpo-44150 : Add optional weights argument to statistics.fmean().

bpo-44142 : ast.unparse() will now drop the redundant parentheses when tuples used as assignment targets (e.g in for loops).

bpo-44145 : hmac computations were not releasing the GIL while calling the OpenSSL HMAC_Update C API (a new feature in 3.9). This unintentionally prevented parallel computation as other hashlib algorithms support.

bpo-44095 : zipfile.Path now supports zipfile.Path.stem , zipfile.Path.suffixes , and zipfile.Path.suffix attributes.

bpo-44077 : It’s now possible to receive the type of service (ToS), a.k.a. differentiated services (DS), a.k.a. differentiated services code point (DSCP) and explicit congestion notification (ECN) IP header fields with socket.IP_RECVTOS .

bpo-37788 : Fix a reference leak when a Thread object is never joined.

bpo-38908 : Subclasses of typing.Protocol which only have data variables declared will now raise a TypeError when checked with isinstance unless they are decorated with runtime_checkable() . Previously, these checks passed silently. Patch provided by Yurii Karabas.

bpo-44098 : typing.ParamSpec will no longer be found in the __parameters__ of most typing generics except in valid use locations specified by PEP 612 . This prevents incorrect usage like typing.List[P][int] . This change means incorrect usage which may have passed silently in 3.10 beta 1 and earlier will now error.

bpo-44089 : Allow subclassing csv.Error in 3.10 (it was allowed in 3.9 and earlier but was disallowed in early versions of 3.10).

bpo-44081 : ast.unparse() now doesn’t use redundant spaces to separate lambda and the : if there are no parameters.

bpo-44061 : Fix regression in previous release when calling pkgutil.iter_modules() with a list of pathlib.Path objects

bpo-44059 : Register the SerenityOS Browser in the webbrowser module.

bpo-36515 : The hashlib module no longer does unaligned memory accesses when compiled for ARM platforms.

bpo-40465 : Remove random module features deprecated in Python 3.9.

bpo-44018 : random.seed() no longer mutates bytearray inputs.

bpo-38352 : Add IO , BinaryIO , TextIO , Match , and Pattern to typing.__all__ . Patch by Jelle Zijlstra.

bpo-44002 : urllib.parse now uses functool.lru_cache() for its internal URL splitting and quoting caches instead of rolling its own like its the ‘90s.

The undocumented internal urllib.parse Quoted class API is now deprecated, for removal in 3.14.

bpo-43972 : When http.server.SimpleHTTPRequestHandler sends a 301 (Moved Permanently) for a directory path not ending with / , add a Content-Length: 0 header. This improves the behavior for certain clients.

bpo-28528 : Fix a bug in pdb where checkline() raises AttributeError if it is called after reset() .

bpo-43853 : Improved string handling for sqlite3 user-defined functions and aggregates:

It is now possible to pass strings with embedded null characters to UDFs

Conversion failures now correctly raise MemoryError

bpo-43666 : AIX: Lib/_aix_support.get_platform() may fail in an AIX WPAR. The fileset bos.rte appears to have a builddate in both LPAR and WPAR so this fileset is queried rather than bos.mp64. To prevent a similar situation (no builddate in ODM) a value (9988) sufficient for completing a build is provided. Patch by M Felt.

bpo-43650 : Fix MemoryError in shutil.unpack_archive() which fails inside shutil._unpack_zipfile() on large files. Patch by Igor Bolshakov.

bpo-43612 : zlib.compress() now accepts a wbits parameter which allows users to compress data as a raw deflate block without zlib headers and trailers in one go. Previously this required instantiating a zlib.compressobj . It also provides a faster alternative to gzip.compress when wbits=31 is used.

bpo-43392 : importlib._bootstrap._find_and_load() now implements a two-step check to avoid locking when modules have been already imported and are ready. This improves performance of repeated calls to importlib.import_module() and importlib.__import__() .

bpo-43318 : Fix a bug where pdb does not always echo cleared breakpoints.

bpo-43234 : Prohibit passing non- concurrent.futures.ThreadPoolExecutor executors to loop.set_default_executor() following a deprecation in Python 3.8. Patch by Illia Volochii.

bpo-43232 : Prohibit previously deprecated potentially disruptive operations on asyncio.trsock.TransportSocket . Patch by Illia Volochii.

bpo-30077 : Added support for Apple’s aifc/sowt pseudo-compression

bpo-42971 : Add definition of errno.EQFULL for platforms that define this constant (such as macOS).

bpo-43086 : Added a new optional strict_mode parameter to binascii.a2b_base64 . When scrict_mode is set to True , the a2b_base64 function will accept only valid base64 content. More details about what “valid base64 content” is, can be found in the function’s documentation.

bpo-43024 : Improve the help signature of traceback.print_exception() , traceback.format_exception() and traceback.format_exception_only() .

bpo-33809 : Add the traceback.TracebackException.print() method which prints the formatted exception information.

bpo-42862 : sqlite3 now utilizes functools.lru_cache() to implement the connection statement cache. As a small optimisation, the default statement cache size has been increased from 100 to 128. Patch by Erlend E. Aasland.

bpo-41818 : Soumendra Ganguly: add termios.tcgetwinsize(), termios.tcsetwinsize().

bpo-40497 : subprocess.check_output() now raises ValueError when the invalid keyword argument check is passed by user code. Previously such use would fail later with a TypeError . Patch by Rémi Lapeyre.

bpo-37449 : ensurepip now uses importlib.resources.files() traversable APIs

bpo-40956 : Use Argument Clinic in sqlite3 . Patches by Erlend E. Aasland.

bpo-41730 : DeprecationWarning is now raised when importing tkinter.tix , which has been deprecated in documentation since Python 3.6.

bpo-20684 : Remove unused _signature_get_bound_param function from inspect - by Anthony Sottile.

bpo-41402 : Fix email.message.EmailMessage.set_content() when called with binary data and 7bit content transfer encoding.

bpo-32695 : The compresslevel and preset keyword arguments of tarfile.open() are now both documented and tested.

bpo-41137 : Use utf-8 encoding while reading .pdbrc files. Patch by Srinivas Reddy Thatiparthy

bpo-24391 : Improved reprs of threading synchronization objects: Semaphore , BoundedSemaphore , Event and Barrier .

bpo-5846 : Deprecated the following unittest functions, scheduled for removal in Python 3.13:

findTestCases()

makeSuite()

getTestCaseNames()

bpo-40563 : Support pathlike objects on dbm/shelve. Patch by Hakan Çelik and Henry-Joseph Audéoud.

bpo-34990 : Fixed a Y2k38 bug in the compileall module where it would fail to compile files with a modification time after the year 2038.

bpo-39549 : Whereas the code for reprlib.Repr had previously used a hardcoded string value of ‘…’, this PR updates it to use of a “fillvalue” attribute, whose value defaults to ‘…’ and can be reset in either individual reprlib.Repr instances or in subclasses thereof.

bpo-37022 : pdb now displays exceptions from repr() with its p and pp commands.

bpo-38840 : Fix test___all__ on platforms lacking a shared memory implementation.

bpo-39359 : Add one missing check that the password is a bytes object for an encrypted zipfile.

bpo-38741 : configparser : using ‘]’ inside a section header will no longer cut the section name short at the ‘]’

bpo-38415 : Added missing behavior to contextlib.asynccontextmanager() to match contextlib.contextmanager() so decorated functions can themselves be decorators.

bpo-30256 : Pass multiprocessing BaseProxy argument manager_owned through AutoProxy.

bpo-27513 : email.utils.getaddresses() now accepts email.header.Header objects along with string values. Patch by Zackery Spytz.

bpo-16379 : Add SQLite error code and name to sqlite3 exceptions. Patch by Aviv Palivoda, Daniel Shahaf, and Erlend E. Aasland.

bpo-26228 : pty.spawn no longer hangs on FreeBSD, macOS, and Solaris.

bpo-33349 : lib2to3 now recognizes async generators everywhere.

bpo-29298 : Fix TypeError when required subparsers without dest do not receive arguments. Patch by Anthony Sottile.

bpo-45216 : Remove extra documentation listing methods in difflib . It was rendering twice in pydoc and was outdated in some places.

bpo-45024 : collections.abc documentation has been expanded to explicitly cover how instance and subclass checks work, with additional doctest examples and an exhaustive list of ABCs which test membership purely by presence of the right special method s. Patch by Raymond Hettinger.

bpo-44957 : Promote PEP 604 union syntax by using it where possible. Also, mention X | Y more prominently in section about Union and mention X | None at all in section about Optional .

bpo-16580 : Added code equivalents for the int.to_bytes() and int.from_bytes() methods, as well as tests ensuring that these code equivalents are valid.

bpo-44903 : Removed the othergui.rst file, any references to it, and the list of GUI frameworks in the FAQ. In their place I’ve added links to the Python Wiki page on GUI frameworks .

bpo-33479 : Tkinter documentation has been greatly expanded with new “Architecture” and “Threading model” sections.

bpo-36700 : base64 RFC references were updated to point to RFC 4648 ; a section was added to point users to the new “security considerations” section of the RFC.

bpo-44740 : Replaced occurrences of uppercase “Web” and “Internet” with lowercase versions per the 2016 revised Associated Press Style Book.

bpo-44693 : Update the definition of __future__ in the glossary by replacing the confusing word “pseudo-module” with a more accurate description.

bpo-35183 : Add typical examples to os.path.splitext docs

bpo-30511 : Clarify that shutil.make_archive() is not thread-safe due to reliance on changing the current working directory.

bpo-44561 : Update of three expired hyperlinks in Doc/distributing/index.rst: “Project structure”, “Building and packaging the project”, and “Uploading the project to the Python Packaging Index”.

bpo-44651 : Delete entry “coercion” in Doc/glossary.rst for its outdated definition.

bpo-42958 : Updated the docstring and docs of filecmp.cmp() to be more accurate and less confusing especially in respect to shallow arg.

bpo-44631 : Refactored the repr() code of the _Environ (os module).

bpo-44613 : importlib.metadata is no longer provisional.

bpo-44558 : Match the docstring and python implementation of countOf() to the behavior of its c implementation.

bpo-44544 : List all kwargs for textwrap.wrap() , textwrap.fill() , and textwrap.shorten() . Now, there are nav links to attributes of TextWrap , which makes navigation much easier while minimizing duplication in the documentation.

bpo-38062 : Clarify that atexit uses equality comparisons internally.

bpo-40620 : Convert examples in tutorial controlflow.rst section 4.3 to be interpreter-demo style.

bpo-43066 : Added a warning to zipfile docs: filename arg with a leading slash may cause archive to be un-openable on Windows systems.

bpo-39452 : Rewrote Doc/library/__main__.rst . Broadened scope of the document to explicitly discuss and differentiate between __main__.py in packages versus the __name__ == '__main__' expression (and the idioms that surround it).

bpo-13814 : In the Design FAQ, answer “Why don’t generators support the with statement?”

bpo-27752 : Documentation of csv.Dialect is more descriptive.

bpo-44453 : Fix documentation for the return type of sysconfig.get_path() .

bpo-44392 : Added a new section in the C API documentation for types used in type hinting. Documented Py_GenericAlias and Py_GenericAliasType .

bpo-38291 : Mark typing.io and typing.re as deprecated since Python 3.8 in the documentation. They were never properly supported by type checkers.

bpo-44322 : Document that SyntaxError args have a details tuple and that details are adjusted for errors in f-string field replacement expressions.

bpo-42392 : Document the deprecation and removal of the loop parameter for many functions and classes in asyncio .

bpo-44195 : Corrected references to TraversableResources in docs. There is no TraversableReader .

bpo-41963 : Document that ConfigParser strips off comments when reading configuration files.

bpo-44072 : Correct where in the numeric ABC hierarchy ** support is added, i.e., in numbers.Complex, not numbers.Integral.

bpo-43558 : Add the remark to dataclasses documentation that the __init__() of any base class has to be called in __post_init__() , along with a code example.

bpo-44025 : Clarify when ‘_’ in match statements is a keyword, and when not.

bpo-41706 : Fix docs about how methods like __add__ are invoked when evaluating operator expressions.

bpo-41621 : Document that collections.defaultdict parameter default_factory defaults to None and is positional-only.

bpo-41576 : document BaseException in favor of bare except

bpo-21760 : The description for __file__ fixed. Patch by Furkan Onder

bpo-39498 : Add a “Security Considerations” index which links to standard library modules that have explicitly documented security considerations.

bpo-33479 : Remove the unqualified claim that tkinter is threadsafe. It has not been true for several years and likely never was. An explanation of what is true may be added later, after more discussion, and possibly after patching _tkinter.c,

bpo-40173 : Fix test.support.import_helper.import_fresh_module() .

bpo-45280 : Add a test case for empty typing.NamedTuple .

bpo-45269 : Cover case when invalid markers type is supplied to c_make_encoder .

bpo-45128 : Fix test_multiprocessing_fork failure due to test_logging and sys.modules manipulation.

bpo-45209 : Fix UserWarning: resource_tracker warning in _test_multiprocessing._TestSharedMemory.test_shared_memory_cleaned_after_process_termination

bpo-45185 : Enables TestEnumerations test cases in test_ssl suite.

bpo-45195 : Fix test_readline.test_nonascii(): sometimes, the newline character is not written at the end, so don’t expect it in the output. Patch by Victor Stinner.

bpo-45156 : Fixes infinite loop on unittest.mock.seal() of mocks created by create_autospec() .

bpo-45125 : Improves pickling tests and docs of SharedMemory and SharableList objects.

bpo-44860 : Update test_sysconfig.test_user_similar() for the posix_user scheme: platlib doesn’t use sys.platlibdir . Patch by Victor Stinner.

bpo-45052 : WithProcessesTestSharedMemory.test_shared_memory_basics test was ignored, because self.assertEqual(sms.size, sms2.size) line was failing. It is now removed and test is unskipped.

The main motivation for this line to be removed from the test is that the size of SharedMemory is not ever guaranteed to be the same. It is decided by the platform.

bpo-44895 : libregrtest now clears the type cache later to reduce the risk of false alarm when checking for reference leaks. Previously, the type cache was cleared too early and libregrtest raised a false alarm about reference leaks under very specific conditions. Patch by Irit Katriel and Victor Stinner.

bpo-45042 : Fixes that test classes decorated with @hashlib_helper.requires_hashdigest were skipped all the time.

bpo-25130 : Add calls of gc.collect() in tests to support PyPy.

bpo-45011 : Made tests relying on the _asyncio C extension module optional to allow running on alternative Python implementations. Patch by Serhiy Storchaka.

bpo-44949 : Fix auto history tests of test_readline: sometimes, the newline character is not written at the end, so don’t expect it in the output.

bpo-44891 : Tests were added to clarify id() is preserved when obj * 1 is used on str and bytes objects. Patch by Nikita Sobolev.

bpo-44852 : Add ability to wholesale silence DeprecationWarnings while running the regression test suite.

bpo-40928 : Notify users running test_decimal regression tests on macOS of potential harmless “malloc can’t allocate region” messages spewed by test_decimal.

bpo-44734 : Fixed floating-point precision issue in turtle tests.

bpo-44708 : Regression tests, when run with -w, are now re-running only the affected test methods instead of re-running the entire test file.

bpo-42095 : Added interop tests for Apple plists: generate plist files with Python plistlib and parse with Apple plutil; and the other way round.

bpo-44647 : Added a permanent Unicode-valued environment variable to regression tests to ensure they handle this use case in the future. If your test environment breaks because of that, report a bug to us, and temporarily set PYTHONREGRTEST_UNICODE_GUARD=0 in your test environment.

bpo-44515 : Adjust recently added contextlib tests to avoid assuming the use of a refcounted GC

bpo-44287 : Fix asyncio test_popen() of test_windows_utils by using a longer timeout. Use military grade battle-tested test.support.SHORT_TIMEOUT timeout rather than a hardcoded timeout of 10 seconds: it’s 30 seconds by default, but it is made longer on slow buildbots. Patch by Victor Stinner.

bpo-44451 : Reset DeprecationWarning filters in test.test_importlib.test_metadata_api.APITests.test_entry_points_by_index to avoid StopIteration error if DeprecationWarnings are ignored.

bpo-44363 : Account for address sanitizer in test_capi. test_capi now passes when run GCC address sanitizer.

bpo-44364 : Add non integral tests for math.sqrt() function.

bpo-43921 : Fix test_ssl.test_wrong_cert_tls13(): use suppress_ragged_eofs=False , since read() can raise ssl.SSLEOFError on Windows. Patch by Victor Stinner.

bpo-43921 : Fix test_pha_required_nocert() of test_ssl: catch two more EOF cases (when the recv() method returns an empty string). Patch by Victor Stinner.

bpo-44131 : Add test_frozenmain to test_embed to test the Py_FrozenMain() C function. Patch by Victor Stinner.

bpo-31904 : Ignore error string case in test_file_not_exists().

bpo-42083 : Add test to check that PyStructSequence_NewType accepts a PyStructSequence_Desc with doc field set to NULL .

bpo-35753 : Fix crash in doctest when doctest parses modules that include unwrappable functions by skipping those functions.

bpo-30256 : Add test for nested queues when using multiprocessing shared objects AutoProxy[Queue] inside ListProxy and DictProxy

bpo-45220 : Avoid building with the Windows 11 SDK previews automatically. This may be overridden by setting the DefaultWindowsSDKVersion environment variable before building.

bpo-45020 : Freeze stdlib modules that are imported during startup. This provides significant performance improvements to startup. If necessary, use the previously added “-X frozen_modules=off” commandline option to force importing the source modules.

bpo-45188 : Windows builds now regenerate frozen modules as the first part of the build. Previously the regeneration was later in the build, which would require it to be restarted if any modules had changed.

bpo-45163 : Fixes Haiku platform build.

bpo-45067 : The ncurses function extended_color_content was introduced in 2017

( https://invisible-island.net/ncurses/NEWS.html#index-t20170401 ). The

ncurses-devel package in CentOS 7 had a older version ncurses resulted in compilation error. For compiling ncurses with extended color support, we verify the version of the ncurses library >= 20170401.

bpo-45019 : Generate lines in relevant files for frozen modules. Up until now each of the files had to be edited manually. This change makes it easier to add to and modify the frozen modules.

bpo-44340 : Add support for building with clang thin lto via –with-lto=thin/full. Patch by Donghee Na and Brett Holman.

bpo-44535 : Enable building using a Visual Studio 2022 install on Windows.

bpo-43298 : Improved error message when building without a Windows SDK installed.

bpo-44381 : The Windows build now accepts EnableControlFlowGuard set to guard to enable CFG.

bpo-41282 : Fix broken make install that caused standard library extension modules to be unnecessarily and incorrectly rebuilt during the install phase of cpython.

bpo-45375 : Fixes an assertion failure due to searching for the standard library in unnormalised paths.

bpo-45022 : Update Windows release to include libffi 3.4.2

bpo-45007 : Update to OpenSSL 1.1.1l in Windows build

bpo-44848 : Upgrade Windows installer to use SQLite 3.36.0.

bpo-44572 : Avoid consuming standard input in the platform module

bpo-44582 : Accelerate speed of mimetypes initialization using a native implementation of the registry scan.

bpo-41299 : Fix 16 milliseconds jitter when using timeouts in threading , such as with threading.Lock.acquire() or threading.Condition.wait() .

bpo-42686 : Build sqlite3 with math functions enabled. Patch by Erlend E. Aasland.

bpo-40263 : This is a follow-on bug from https://bugs.python.org/issue26903 . Once that is applied we run into an off-by-one assertion problem. The assert was not correct.

bpo-45007 : Update macOS installer builds to use OpenSSL 1.1.1l.

bpo-34602 : When building CPython on macOS with ./configure --with-undefined-behavior-sanitizer --with-pydebug , the stack size is now quadrupled to allow for the entire test suite to pass.

bpo-44848 : Update macOS installer to use SQLite 3.36.0.

bpo-44689 : ctypes.util.find_library() now works correctly on macOS 11 Big Sur even if Python is built on an older version of macOS. Previously, when built on older macOS systems, find_library was not able to find macOS system libraries when running on Big Sur due to changes in how system libraries are stored.

bpo-41972 : The framework build’s user header path in sysconfig is changed to add a ‘pythonX.Y’ component to match distutils’s behavior.

bpo-43109 : Allow –with-lto configure option to work with Apple-supplied Xcode or Command Line Tools.

bpo-34932 : Add socket.TCP_KEEPALIVE support for macOS. Patch by Shane Harvey.

bpo-45296 : On Windows, change exit/quit message to suggest Ctrl-D, which works, instead of <Ctrl-Z Return>, which does not work in IDLE.

bpo-45193 : Make completion boxes appear on Ubuntu again.

bpo-40128 : Mostly fix completions on macOS when not using tcl/tk 8.6.11 (as with 3.9). The added update_idletask call should be harmless and possibly helpful otherwise.

bpo-33962 : Move the indent space setting from the Font tab to the new Windows tab. Patch by Mark Roseman and Terry Jan Reedy.

bpo-40468 : Split the settings dialog General tab into Windows and Shell/ED tabs. Move help sources, which extend the Help menu, to the Extensions tab. Make space for new options and shorten the dialog. The latter makes the dialog better fit small screens.

bpo-41611 : Avoid uncaught exceptions in AutoCompleteWindow.winconfig_event() .

bpo-41611 : Fix IDLE sometimes freezing upon tab-completion on macOS.

bpo-44010 : Highlight the new match statement’s soft keywords : match , case , and _ . However, this highlighting is not perfect and will be incorrect in some rare cases, including some _ -s in case patterns.

bpo-44026 : Include interpreter’s typo fix suggestions in message line for NameErrors and AttributeErrors. Patch by E. Paine.

bpo-44786 : Fix a warning in regular expression in the c-analyzer script.

bpo-44967 : pydoc now returns a non-zero status code when a module cannot be found.

bpo-44978 : Allow the Argument Clinic tool to handle __complex__ special methods.

bpo-43425 : Removed the ‘test2to3’ demo project that demonstrated using lib2to3 to support Python 2.x and Python 3.x from a single source in a distutils package. Patch by Donghee Na

bpo-44074 : Make patchcheck automatically detect the correct base branch name (previously it was hardcoded to ‘master’)

bpo-20291 : Added support for variadic positional parameters in Argument Clinic.

bpo-41710 : The PyThread_acquire_lock_timed() function now clamps the timeout if it is too large, rather than aborting the process. Patch by Victor Stinner.

bpo-44687 : BufferedReader.peek() no longer raises ValueError when the entire file has already been buffered.

bpo-45116 : Add the Py_ALWAYS_INLINE macro to ask the compiler to always inline a static inline function. The compiler can ignore it and decides to not inline the function. Patch by Victor Stinner.

bpo-45094 : Add the Py_NO_INLINE macro to disable inlining on a function. Patch by Victor Stinner.

bpo-45061 : Add a deallocator to the bool type to detect refcount bugs in C extensions which call Py_DECREF(Py_True); or Py_DECREF(Py_False); by mistake. Patch by Victor Stinner.

bpo-42035 : Add a new PyType_GetQualName() function to get type’s qualified name.

bpo-41103 : Reverts removal of the old buffer protocol because they are part of stable ABI.

bpo-44751 : Remove crypt.h include from the public Python.h header.

bpo-42747 : The Py_TPFLAGS_HAVE_VERSION_TAG type flag now does nothing. The Py_TPFLAGS_HAVE_AM_SEND flag (which was added in 3.10) is removed. Both were unnecessary because it is not possible to have type objects with the relevant fields missing.

bpo-44530 : Added the co_qualname to the PyCodeObject structure to propagate the qualified name from the compiler to code objects.

Patch by Gabriele N. Tornetta

bpo-44441 : Py_RunMain() now resets PyImport_Inittab to its initial value at exit. It must be possible to call PyImport_AppendInittab() or PyImport_ExtendInittab() at each Python initialization. Patch by Victor Stinner.

bpo-39947 : Remove 4 private trashcan C API functions which were only kept for the backward compatibility of the stable ABI with Python 3.8 and older, since the trashcan API was not usable with the limited C API on Python 3.8 and older. The trashcan API was excluded from the limited C API in Python 3.9.

Removed functions:

_PyTrash_deposit_object()

_PyTrash_destroy_chain()

_PyTrash_thread_deposit_object()

_PyTrash_thread_destroy_chain()

The trashcan C API was never usable with the limited C API, since old trashcan macros accessed directly PyThreadState members like _tstate->trash_delete_nesting , whereas the PyThreadState structure is opaque in the limited C API.

Exclude also the PyTrash_UNWIND_LEVEL constant from the C API.

bpo-40939 : Removed documentation for the removed PyParser_* C API.

bpo-43795 : The list in Contents of Limited API now shows the public name PyFrameObject rather than _frame . The non-existing entry _node no longer appears in the list.

bpo-44378 : Py_IS_TYPE() no longer uses Py_TYPE() to avoid a compiler warning: no longer cast const PyObject* to PyObject* . Patch by Victor Stinner.

bpo-39573 : Convert the Py_TYPE() and Py_SIZE() macros to static inline functions. The Py_SET_TYPE() and Py_SET_SIZE() functions must now be used to set an object type and size. Patch by Victor Stinner.

bpo-44263 : The PyType_Ready() function now raises an error if a type is defined with the Py_TPFLAGS_HAVE_GC flag set but has no traverse function ( PyTypeObject.tp_traverse ). Patch by Victor Stinner.

bpo-43795 : The undocumented function Py_FrozenMain() is removed from the Limited API.

bpo-44113 : Deprecate the following functions to configure the Python initialization:

Use the new PyConfig API of the Python Initialization Configuration instead ( PEP 587 ).

bpo-44094 : Remove PyErr_SetFromErrnoWithUnicodeFilename() , PyErr_SetFromWindowsErrWithUnicodeFilename() , and PyErr_SetExcFromWindowsErrWithUnicodeFilename() . They are not documented and have been deprecated since Python 3.3.

bpo-43795 : PyCodec_Unregister() is now properly exported as a function in the Windows Stable ABI DLL.

bpo-44029 : Remove deprecated Py_UNICODE APIs: PyUnicode_Encode , PyUnicode_EncodeUTF7 , PyUnicode_EncodeUTF8 , PyUnicode_EncodeUTF16 , PyUnicode_EncodeUTF32 , PyUnicode_EncodeLatin1 , PyUnicode_EncodeMBCS , PyUnicode_EncodeDecimal , PyUnicode_EncodeRawUnicodeEscape , PyUnicode_EncodeCharmap , PyUnicode_EncodeUnicodeEscape , PyUnicode_TransformDecimalToASCII , PyUnicode_TranslateCharmap , PyUnicodeEncodeError_Create , PyUnicodeTranslateError_Create . See PEP 393 and PEP 624 for reference.

bpo-42035 : Add a new PyType_GetName() function to get type’s short name.

Python 3.10.0 beta 1 ¶

Release date: 2021-05-03

bpo-43434 : Creating sqlite3.Connection objects now also produces sqlite3.connect and sqlite3.connect/handle auditing events . Previously these events were only produced by sqlite3.connect() calls. Patch by Erlend E. Aasland.

bpo-43998 : The ssl module sets more secure cipher suites defaults. Ciphers without forward secrecy and with SHA-1 MAC are disabled by default. Security level 2 prohibits weak RSA, DH, and ECC keys with less than 112 bits of security. SSLContext defaults to minimum protocol version TLS 1.2. Settings are based on Hynek Schlawack’s research.

bpo-43882 : The presence of newline or tab characters in parts of a URL could allow some forms of attacks.

Following the controlling specification for URLs defined by WHATWG urllib.parse() now removes ASCII newlines and tabs from URLs, preventing such attacks.

bpo-43472 : Ensures interpreter-level audit hooks receive the cpython.PyInterpreterState_New event when called through the _xxsubinterpreters module.

bpo-43362 : Fix invalid free in _sha3 module. The issue was introduced in 3.10.0a1. Python 3.9 and earlier are not affected.

bpo-43762 : Add audit events for sqlite3.connect/handle() , sqlite3.Connection.enable_load_extension() , and sqlite3.Connection.load_extension() . Patch by Erlend E. Aasland.

bpo-43756 : Add new audit event glob.glob/2 to incorporate the new root_dir and dir_fd arguments added to glob.glob() and glob.iglob() .

bpo-36384 : ipaddress module no longer accepts any leading zeros in IPv4 address strings. Leading zeros are ambiguous and interpreted as octal notation by some libraries. For example the legacy function socket.inet_aton() treats leading zeros as octal notation. glibc implementation of modern inet_pton() does not accept any leading zeros. For a while the ipaddress module used to accept ambiguous leading zeros.

bpo-43075 : Fix Regular Expression Denial of Service (ReDoS) vulnerability in urllib.request.AbstractBasicAuthHandler . The ReDoS-vulnerable regex has quadratic worst-case complexity and it allows cause a denial of service when identifying crafted invalid RFCs. This ReDoS issue is on the client side and needs remote attackers to control the HTTP server.

bpo-42800 : Audit hooks are now fired for frame.f_code, traceback.tb_frame, and generator code/frame attribute access.

bpo-37363 : Add audit events to the http.client module.

bpo-43977 : Prevent classes being both a sequence and a mapping when pattern matching.

bpo-43977 : Use tp_flags on the class object to determine if the subject is a sequence or mapping when pattern matching. Avoids the need to import collections.abc when pattern matching.

bpo-43892 : Restore proper validation of complex literal value patterns when parsing match blocks.

bpo-43933 : Set frame.f_lineno to the line number of the ‘with’ kweyword when executing the call to __exit__ .

bpo-43933 : If the current position in a frame has no line number then set the f_lineno attribute to None, instead of -1, to conform to PEP 626. This should not normally be possible, but might occur in some unusual circumstances.

bpo-43963 : Importing the _signal module in a subinterpreter has no longer side effects.

bpo-42739 : The internal representation of line number tables is changed to not use sentinels, and an explicit length parameter is added to the out of process API function PyLineTable_InitAddressRange . This makes the handling of line number tables more robust in some circumstances.

bpo-43908 : Make re types immutable. Patch by Erlend E. Aasland.

bpo-43908 : Make the array.array type immutable. Patch by Erlend E. Aasland.

bpo-43901 : Change class and module objects to lazy-create empty annotations dicts on demand. The annotations dicts are stored in the object’s __dict__ for backwards compatibility.

bpo-43892 : Match patterns now use new dedicated AST nodes ( MatchValue , MatchSingleton , MatchSequence , MatchStar , MatchMapping , MatchClass ) rather than reusing expression AST nodes. MatchAs and MatchOr are now defined as pattern nodes rather than as expression nodes. Patch by Nick Coghlan.

bpo-42725 : Usage of await / yield / yield from and named expressions within an annotation is now forbidden when PEP 563 is activated.

bpo-43754 : When performing structural pattern matching ( PEP 634 ), captured names are now left unbound until the entire pattern has matched successfully.

bpo-42737 : Annotations for complex targets (everything beside simple names) no longer cause any runtime effects with from __future__ import annotations .

bpo-43914 : SyntaxError exceptions raised by the interpreter will highlight the full error range of the expression that constitutes the syntax error itself, instead of just where the problem is detected. Patch by Pablo Galindo.

bpo-38605 : Revert making from __future__ import annotations the default. This follows the Steering Council decision to postpone PEP 563 changes to at least Python 3.11. See the original email for more information regarding the decision: https://mail.python.org/archives/list/python-dev@python.org/thread/CLVXXPQ2T2LQ5MP2Y53VVQFCXYWQJHKZ/ . Patch by Pablo Galindo.

bpo-43475 : Hashes of NaN values now depend on object identity. Formerly, they always hashed to 0 even though NaN values are not equal to one another. Having the same hash for unequal values caused pile-ups in hash tables.

bpo-43859 : Improve the error message for IndentationError exceptions. Patch by Pablo Galindo

bpo-41323 : Constant tuple folding in bytecode optimizer now reuses tuple in constant table.

bpo-43846 : Data stack usage is much reduced for large literal and call expressions.

bpo-38530 : When printing NameError raised by the interpreter, PyErr_Display() will offer suggestions of similar variable names in the function that the exception was raised from. Patch by Pablo Galindo

bpo-43823 : Improve syntax errors for invalid dictionary literals. Patch by Pablo Galindo.

bpo-43822 : Improve syntax errors in the parser for missing commas between expressions. Patch by Pablo Galindo.

bpo-43798 : ast.alias nodes now include source location metadata attributes e.g. lineno, col_offset.

bpo-43797 : Improve SyntaxError error messages for invalid comparisons. Patch by Pablo Galindo.

bpo-43760 : Move the flag for checking whether tracing is enabled to the C stack, from the heap. Should speed up dispatch in the interpreter.

bpo-43682 : Static methods ( @staticmethod ) and class methods ( @classmethod ) now inherit the method attributes ( __module__ , __name__ , __qualname__ , __doc__ , __annotations__ ) and have a new __wrapped__ attribute. Patch by Victor Stinner.

bpo-43751 : Fixed a bug where anext(ait, default) would erroneously return None.

bpo-42128 : __match_args__ is no longer allowed to be a list.

bpo-43683 : Add GEN_START opcode. Marks start of generator, including async, or coroutine and handles sending values to a newly created generator or coroutine.

bpo-43105 : Importlib now resolves relative paths when creating module spec objects from file locations.

bpo-43682 : Static methods ( @staticmethod ) are now callable as regular functions. Patch by Victor Stinner.

bpo-42609 : Prevented crashes in the AST validator and optimizer when compiling some absurdly long expressions like "+0"*1000000 . RecursionError is now raised instead.

bpo-38530 : When printing AttributeError , PyErr_Display() will offer suggestions of similar attribute names in the object that the exception was raised from. Patch by Pablo Galindo

bpo-44015 : In @dataclass(), raise a TypeError if KW_ONLY is specified more than once.

bpo-25478 : Added a total() method to collections.Counter() to compute the sum of the counts.

bpo-43733 : Change netrc.netrc to use UTF-8 encoding before using locale encoding.

bpo-43979 : Removed an unnecessary list comprehension before looping from urllib.parse.parse_qsl() . Patch by Christoph Zwerschke and Donghee Na.

bpo-43993 : Update bundled pip to 21.1.1.

bpo-43957 : [Enum] Deprecate TypeError when non-member is used in a containment check; In 3.12 True or False will be returned instead, and containment will return True if the value is either a member of that enum or one of its members’ value.

bpo-42904 : For backwards compatibility with previous minor versions of Python, if typing.get_type_hints() receives no namespace dictionary arguments, typing.get_type_hints() will search through the global then local namespaces during evaluation of stringized type annotations (string forward references) inside a class.

bpo-43945 : [Enum] Deprecate non-standard mixin format() behavior: in 3.12 the enum member, not the member’s value, will be used for format() calls.

bpo-41139 : Deprecate undocumented cgi.log() API.

bpo-43937 : Fixed the turtle module working with non-default root window.

bpo-43930 : Update bundled pip to 21.1 and setuptools to 56.0.0

bpo-43907 : Fix a bug in the pure-Python pickle implementation when using protocol 5, where bytearray instances that occur several time in the pickled object graph would incorrectly unpickle into repeated copies of the bytearray object.

bpo-43926 : In importlib.metadata , provide a uniform interface to Description , allow for any field to be encoded with multiline values, remove continuation lines from multiline values, and add a .json property for easy access to the PEP 566 JSON-compatible form. Sync with importlib_metadata 4.0 .

bpo-43920 : OpenSSL 3.0.0: load_verify_locations() now returns a consistent error message when cadata contains no valid certificate.

bpo-43607 : urllib can now convert Windows paths with \\?\ prefixes into URL paths.

bpo-43817 : Add inspect.get_annotations() , which safely computes the annotations defined on an object. It works around the quirks of accessing the annotations from various types of objects, and makes very few assumptions about the object passed in. inspect.get_annotations() can also correctly un-stringize stringized annotations.

inspect.signature() , inspect.from_callable() , and inspect.from_function() now call inspect.get_annotations() to retrieve annotations. This means inspect.signature() and inspect.from_callable() can now un-stringize stringized annotations, too.

bpo-43284 : platform.win32_ver derives the windows version from sys.getwindowsversion().platform_version which in turn derives the version from kernel32.dll (which can be of a different version than Windows itself). Therefore change the platform.win32_ver to determine the version using the platform module’s _syscmd_ver private function to return an accurate version.

bpo-42854 : The ssl module now uses SSL_read_ex and SSL_write_ex internally. The functions support reading and writing of data larger than 2 GB. Writing zero-length data no longer fails with a protocol violation error.

bpo-42333 : Port _ssl extension module to multiphase initialization.

bpo-43880 : ssl now raises DeprecationWarning for OP_NO_SSL/TLS* options, old TLS versions, old protocols, and other features that have been deprecated since Python 3.6, 3.7, or OpenSSL 1.1.0.

bpo-41559 : PEP 612 is now implemented purely in Python; builtin types.GenericAlias objects no longer include typing.ParamSpec in __parameters__ (with the exception of collections.abc.Callable ‘s GenericAlias ). This means previously invalid uses of ParamSpec (such as list[P] ) which worked in earlier versions of Python 3.10 alpha, will now raise TypeError during substitution.

bpo-43867 : The multiprocessing Server class now explicitly catches SystemExit and closes the client connection in this case. It happens when the Server.serve_client() method reaches the end of file (EOF).

bpo-40443 : Remove unused imports: pyclbr no longer uses copy, and typing no longer uses ast. Patch by Victor Stinner.

bpo-43820 : Remove an unneeded copy of the namespace passed to dataclasses.make_dataclass().

bpo-43787 : Add __iter__() method to bz2.BZ2File , gzip.GzipFile , and lzma.LZMAFile . It makes iterating them about 2x faster. Patch by Inada Naoki.

bpo-43680 : Deprecate io.OpenWrapper and _pyio.OpenWrapper: use io.open and _pyio.open instead. Until Python 3.9, _pyio.open was not a static method and builtins.open was set to OpenWrapper to not become a bound method when set to a class variable. _io.open is a built-in function whereas _pyio.open is a Python function. In Python 3.10, _pyio.open() is now a static method, and builtins.open() is now io.open().

bpo-43680 : The Python _pyio.open() function becomes a static method to behave as io.open() built-in function: don’t become a bound method when stored as a class variable. It becomes possible since static methods are now callable in Python 3.10. Moreover, _pyio.OpenWrapper() becomes a simple alias to _pyio.open() . Patch by Victor Stinner.

bpo-41515 : Fix KeyError raised in typing.get_type_hints() due to synthetic modules that don’t appear in sys.modules .

bpo-43776 : When subprocess.Popen args are provided as a string or as pathlib.Path , the Popen instance repr now shows the right thing.

bpo-42248 : [Enum] ensure exceptions raised in _missing__ are released

bpo-43744 : fix issue with enum member name matching the start of a private variable name

bpo-43772 : Fixed the return value of TypeVar.__ror__ . Patch by Jelle Zijlstra.

bpo-43764 : Add match_args parameter to @dataclass decorator to allow suppression of __match_args__ generation.

bpo-43799 : OpenSSL 3.0.0: define OPENSSL_API_COMPAT 1.1.1 to suppress deprecation warnings. Python requires OpenSSL 1.1.1 APIs.

bpo-43478 : Mocks can no longer be used as the specs for other Mocks. As a result, an already-mocked object cannot have an attribute mocked using autospec=True or be the subject of a create_autospec(...) call. This can uncover bugs in tests since these Mock-derived Mocks will always pass certain tests (e.g. isinstance() ) and builtin assert functions (e.g. assert_called_once_with) will unconditionally pass.

bpo-43794 : Add ssl.OP_IGNORE_UNEXPECTED_EOF constants (OpenSSL 3.0.0)

bpo-43785 : Improve bz2.BZ2File performance by removing the RLock from BZ2File. This makes BZ2File thread unsafe in the face of multiple simultaneous readers or writers, just like its equivalent classes in gzip and lzma have always been. Patch by Inada Naoki.

bpo-43789 : OpenSSL 3.0.0: Don’t call the password callback function a second time when first call has signaled an error condition.

bpo-43788 : The header files for ssl error codes are now OpenSSL version-specific. Exceptions will now show correct reason and library codes. The make_ssl_data.py script has been rewritten to use OpenSSL’s text file with error codes.

bpo-43766 : Implement PEP 647 in the typing module by adding TypeGuard .

bpo-25264 : os.path.realpath() now accepts a strict keyword-only argument. When set to True , OSError is raised if a path doesn’t exist or a symlink loop is encountered.

bpo-43780 : In importlib.metadata , incorporate changes from importlib_metadata 3.10: Add mtime-based caching during distribution discovery. Flagged use of dict result from entry_points() as deprecated.

The P.args and P.kwargs attributes of typing.ParamSpec are now instances of the new classes typing.ParamSpecArgs and typing.ParamSpecKwargs , which enables a more useful repr() . Patch by Jelle Zijlstra.

bpo-43731 : Add an encoding parameter logging.fileConfig() .

bpo-43712 : Add encoding and errors parameters to fileinput.input() and fileinput.FileInput .

bpo-38659 : A simple_enum decorator is added to the enum module to convert a normal class into an Enum. test_simple_enum added to test simple enums against a corresponding normal Enum. Standard library modules updated to use simple_enum .

bpo-43764 : Fix an issue where __match_args__ generation could fail for some dataclasses .

bpo-43752 : Fix sqlite3 regression for zero-sized blobs with converters, where b"" was returned instead of None . The regression was introduced by PR 24723. Patch by Erlend E. Aasland.

bpo-43655 : tkinter dialog windows are now recognized as dialogs by window managers on macOS and X Window.

bpo-43723 : The following threading methods are now deprecated and should be replaced:

currentThread => threading.current_thread()

activeCount => threading.active_count()

Condition.notifyAll => threading.Condition.notify_all()

Event.isSet => threading.Event.is_set()

Thread.setName => threading.Thread.name

thread.getName => threading.Thread.name

Thread.isDaemon => threading.Thread.daemon

Thread.setDaemon => threading.Thread.daemon

Patch by Jelle Zijlstra.

bpo-2135 : Deprecate find_module() and find_loader() implementations in importlib and zipimport.

bpo-43534 : turtle.textinput() and turtle.numinput() create now a transient window working on behalf of the canvas window.

bpo-43532 : Add the ability to specify keyword-only fields to dataclasses. These fields will become keyword-only arguments to the generated __init__.

bpo-43522 : Fix problem with hostname_checks_common_name . OpenSSL does not copy hostflags from struct SSL_CTX to struct SSL .

bpo-8978 : Improve error message for tarfile.open() when lzma / bz2 are unavailable. Patch by Anthony Sottile.

bpo-42967 : Allow bytes separator argument in urllib.parse.parse_qs and urllib.parse.parse_qsl when parsing str query strings. Previously, this raised a TypeError .

bpo-43296 : Improve sqlite3 error handling: sqlite3_value_blob() errors that set SQLITE_NOMEM now raise MemoryError . Patch by Erlend E. Aasland.

bpo-43312 : New functions sysconfig.get_preferred_scheme() and sysconfig.get_default_scheme() are added to query a platform for its preferred “user”, “home”, and “prefix” (default) scheme names.

bpo-43265 : Improve sqlite3.Connection.backup() error handling. The error message for non-existent target database names is now unknown database <database name> instead of SQL logic error . Patch by Erlend E. Aasland.

bpo-41282 : Install schemes in distutils.command.install are now loaded from sysconfig .

bpo-41282 : distutils.sysconfig has been merged to sysconfig .

bpo-43176 : Fixed processing of a dataclass that inherits from a frozen dataclass with no fields. It is now correctly detected as an error.

bpo-43080 : pprint now has support for dataclasses.dataclass . Patch by Lewis Gaul.

bpo-39950 : Add pathlib.Path.hardlink_to() method that supersedes link_to() . The new method has the same argument order as symlink_to() .

bpo-42904 : typing.get_type_hints() now checks the local namespace of a class when evaluating PEP 563 annotations inside said class.

bpo-42269 : Add slots parameter to dataclasses.dataclass decorator to automatically generate __slots__ for class. Patch provided by Yurii Karabas.

bpo-39529 : Deprecated use of asyncio.get_event_loop() without running event loop. Emit deprecation warning for asyncio functions which implicitly create a Future or Task objects if there is no running event loop and no explicit loop argument is passed: ensure_future() , wrap_future() , gather() , shield() , as_completed() and constructors of Future , Task , StreamReader , StreamReaderProtocol .

bpo-18369 : Certificate and PrivateKey classes were added to the ssl module. Certificates and keys can now be loaded from memory buffer, too.

bpo-41486 : Use a new output buffer management code for bz2 / lzma / zlib modules, and add .readall() function to _compression.DecompressReader class. These bring some performance improvements. Patch by Ma Lin.

bpo-31870 : The ssl.get_server_certificate() function now has a timeout parameter.

bpo-41735 : Fix thread locks in zlib module may go wrong in rare case. Patch by Ma Lin.

bpo-36470 : Fix dataclasses with InitVar s and replace() . Patch by Claudiu Popa.

bpo-40849 : Expose X509_V_FLAG_PARTIAL_CHAIN ssl flag

bpo-35114 : ssl.RAND_status() now returns a boolean value (as documented) instead of 1 or 0 .

bpo-39906 : pathlib.Path.stat() and chmod() now accept a follow_symlinks keyword-only argument for consistency with corresponding functions in the os module.

bpo-39899 : os.path.expanduser() now refuses to guess Windows home directories if the basename of current user’s home directory does not match their username.

pathlib.Path.expanduser() and home() now consistently raise RuntimeError exception when a home directory cannot be resolved. Previously a KeyError exception could be raised on Windows when the "USERNAME" environment variable was unset.

bpo-36076 : Added SNI support to ssl.get_server_certificate() .

bpo-38490 : Covariance, Pearson’s correlation, and simple linear regression functionality was added to statistics module. Patch by Tymoteusz Wołodźko.

bpo-33731 : Provide a locale.localize() function, which converts a normalized number string into a locale format.

bpo-32745 : Fix a regression in the handling of ctypes’ ctypes.c_wchar_p type: embedded null characters would cause a ValueError to be raised. Patch by Zackery Spytz.

bpo-43987 : Add “Annotations Best Practices” document as a new HOWTO.

bpo-43977 : Document the new Py_TPFLAGS_MAPPING and Py_TPFLAGS_SEQUENCE type flags.

bpo-43959 : The documentation on the PyContextVar C-API was clarified.

bpo-43938 : Update dataclasses documentation to express that FrozenInstanceError is derived from AttributeError.

bpo-43778 : Fix the Sphinx glossary_search extension: create the _static/ sub-directory if it doesn’t exist.

bpo-43755 : Update documentation to reflect that unparenthesized lambda expressions can no longer be the expression part in an if clause in comprehensions and generator expressions since Python 3.9.

bpo-43739 : Fixing the example code in Doc/extending/extending.rst to declare and initialize the pmodule variable to be of the right type.

bpo-43961 : Fix test_logging.test_namer_rotator_inheritance() on Windows: use os.replace() rather than os.rename() . Patch by Victor Stinner.

bpo-43842 : Fix a race condition in the SMTP test of test_logging. Don’t close a file descriptor (socket) from a different thread while asyncore.loop() is polling the file descriptor. Patch by Victor Stinner.

bpo-43843 : test.libregrtest now marks a test as ENV_CHANGED (altered the execution environment) if a thread raises an exception but does not catch it. It sets a hook on threading.excepthook() . Use --fail-env-changed option to mark the test as failed. Patch by Victor Stinner.

bpo-43811 : Tests multiple OpenSSL versions on GitHub Actions. Use ccache to speed up testing.

bpo-43791 : OpenSSL 3.0.0: Disable testing of legacy protocols TLS 1.0 and 1.1. Tests are failing with TLSV1_ALERT_INTERNAL_ERROR.

bpo-43567 : Improved generated code refresh (AST/tokens/opcodes/keywords) on Windows.

bpo-43669 : Implement PEP 644 . Python now requires OpenSSL 1.1.1 or newer.

bpo-35306 : Adds additional arguments to os.startfile() function.

bpo-43538 : Avoid raising errors from pathlib.Path.exists() when passed an invalid filename.

bpo-38822 : Fixed os.stat() failing on inaccessible directories with a trailing slash, rather than falling back to the parent directory’s metadata. This implicitly affected os.path.exists() and os.path.isdir() .

bpo-26227 : Fixed decoding of host names in socket.gethostbyaddr() and socket.gethostbyname_ex() .

bpo-40432 : Updated pegen regeneration script on Windows to find and use Python 3.8 or higher. Prior to this, pegen regeneration already required 3.8 or higher, but the script may have used lower versions of Python.

bpo-43745 : Actually updates Windows release to OpenSSL 1.1.1k. Earlier releases were mislabelled and actually included 1.1.1i again.

bpo-43652 : Update Tcl and Tk to 8.6.11 in Windows installer.

bpo-43492 : Upgrade Windows installer to use SQLite 3.35.5.

bpo-30555 : Fix WindowsConsoleIO errors in the presence of fd redirection. Patch by Segev Finer.

bpo-42119 : Fix check for macOS SDK paths when building Python. Narrow search to match contents of SDKs, namely only files in /System/Library , /System/IOSSupport , and /usr other than /usr/local . Previously, anything under /System was assumed to be in an SDK which causes problems with the new file system layout in 10.15+ where user file systems may appear to be mounted under /System . Paths in /Library were also incorrectly treated as SDK locations.

bpo-43568 : Drop support for MACOSX_DEPLOYMENT_TARGET < 10.3

bpo-44009 : Provide “python3.x-intel64” executable to allow reliably forcing macOS universal2 framework builds to run under Rosetta 2 Intel-64 emulation on Apple Silicon Macs. This can be useful for testing or when universal2 wheels are not yet available.

bpo-43851 : Build SQLite with SQLITE_OMIT_AUTOINIT on macOS. Patch by Erlend E. Aasland.

bpo-43492 : Update macOS installer to use SQLite 3.35.4.

bpo-42235 : Mac/BuildScript/build-installer.py will now use “–enable-optimizations” and --with-lto when building on macOS 10.15 or later.

bpo-37903 : Add mouse actions to the shell sidebar. Left click and optional drag selects one or more lines, as with the editor line number sidebar. Right click after selecting raises a context menu with ‘copy with prompts’. This zips together prompts from the sidebar with lines from the selected text.

bpo-43981 : Fix reference leak in test_sidebar and test_squeezer. Patches by Terry Jan Reedy and Pablo Galindo

bpo-37892 : Indent IDLE Shell input with spaces instead of tabs

bpo-43655 : IDLE dialog windows are now recognized as dialogs by window managers on macOS and X Window.

bpo-37903 : IDLE’s shell now shows prompts in a separate side-bar.

bpo-43916 : Add a new Py_TPFLAGS_DISALLOW_INSTANTIATION type flag to disallow creating type instances. Patch by Victor Stinner.

bpo-43774 : Remove the now unused PYMALLOC_DEBUG macro. Debug hooks on memory allocators are now installed by default if Python is built in debug mode (if Py_DEBUG macro is defined). Moreover, they can now be used on Python build in release mode (ex: using PYTHONMALLOC=debug environment variable).

bpo-43962 : _PyInterpreterState_IDIncref() now calls _PyInterpreterState_IDInitref() and always increments id_refcount. Previously, calling _xxsubinterpreters.get_current() could create an id_refcount inconsistency when a _xxsubinterpreters.InterpreterID object was deallocated. Patch by Victor Stinner.

bpo-28254 : Add new C-API functions to control the state of the garbage collector: PyGC_Enable() , PyGC_Disable() , PyGC_IsEnabled() , corresponding to the functions in the gc module.

bpo-43908 : Introduce Py_TPFLAGS_IMMUTABLETYPE flag for immutable type objects, and modify PyType_Ready() to set it for static types. Patch by Erlend E. Aasland.

bpo-43795 : PyMem_Calloc() is now available in the limited C API ( Py_LIMITED_API ).

bpo-43868 : PyOS_ReadlineFunctionPointer() is no longer exported by limited C API headers and by python3.dll on Windows. Like any function that takes FILE* , it is not part of the stable ABI.

bpo-43795 : Stable ABI and limited API definitions are generated from a central manifest ( PEP 652 ).

bpo-43753 : Add the Py_Is(x, y) function to test if the x object is the y object, the same as x is y in Python. Add also the Py_IsNone() , Py_IsTrue() , Py_IsFalse() functions to test if an object is, respectively, the None singleton, the True singleton or the False singleton. Patch by Victor Stinner.

Python 3.10.0 alpha 7 ¶

Release date: 2021-04-05

bpo-42988 : CVE-2021-3426 : Remove the getfile feature of the pydoc module which could be abused to read arbitrary files on the disk (directory traversal vulnerability). Moreover, even source code of Python modules can contain sensitive data like passwords. Vulnerability reported by David Schwörer.

bpo-43285 : ftplib no longer trusts the IP address value returned from the server in response to the PASV command by default. This prevents a malicious FTP server from using the response to probe IPv4 address and port combinations on the client network.

Code that requires the former vulnerable behavior may set a trust_server_pasv_ipv4_address attribute on their ftplib.FTP instances to True to re-enable it.

bpo-43439 : Add audit hooks for gc.get_objects() , gc.get_referrers() and gc.get_referents() . Patch by Pablo Galindo.

bpo-27129 : Update CPython bytecode magic number.

bpo-43672 : Raise ImportWarning when calling find_loader().

bpo-43660 : Fix crash that happens when replacing sys.stderr with a callable that can remove the object while an exception is being printed. Patch by Pablo Galindo.

bpo-27129 : The bytecode interpreter uses instruction, rather byte, offsets internally. This reduces the number of EXTENDED_ARG instructions needed and streamlines instruction dispatch a bit.

bpo-40645 : Fix reference leak in the _hashopenssl extension. Patch by Pablo Galindo.

bpo-42134 : Calls to find_module() by the import system now raise ImportWarning.

bpo-41064 : Improve the syntax error for invalid usage of double starred elements (‘**’) in f-strings. Patch by Pablo Galindo.

bpo-43575 : Speed up calls to map() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-42137 : The import system now prefers using __spec__ for ModuleType.__repr__ over module_repr() .

bpo-43452 : Added micro-optimizations to _PyType_Lookup() to improve cache lookup performance in the common case of cache hits.

bpo-43555 : Report the column offset for SyntaxError for invalid line continuation characters. Patch by Pablo Galindo.

bpo-43517 : Fix misdetection of circular imports when using from pkg.mod import attr , which caused false positives in non-trivial multi-threaded code.

bpo-43497 : Emit SyntaxWarnings for assertions with tuple constants, this is a regression introduced in python3.7

bpo-39316 : Tracing now has correct line numbers for attribute accesses when the attribute is on a different line from the object. Improves debugging and profiling for multi-line method chains.

bpo-35883 : Python no longer fails at startup with a fatal error if a command line argument contains an invalid Unicode character. The Py_DecodeLocale() function now escapes byte sequences which would be decoded as Unicode characters outside the [U+0000; U+10ffff] range.

bpo-43410 : Fix a bug that was causing the parser to crash when emitting syntax errors when reading input from stdin. Patch by Pablo Galindo

bpo-43406 : Fix a possible race condition where PyErr_CheckSignals tries to execute a non-Python signal handler.

bpo-42128 : Add __match_args__ to struct sequence objects . Patch by Pablo Galindo.

bpo-43390 : CPython now sets the SA_ONSTACK flag in PyOS_setsig for the VM’s default signal handlers. This is friendlier to other in-process code that an extension module or embedding use could pull in (such as Golang’s cgo) where tiny thread stacks are the norm and sigaltstack() has been used to provide for signal handlers. This is a no-op change for the vast majority of processes that don’t use sigaltstack.

bpo-43287 : Speed up calls to filter() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-37448 : Add a radix tree based memory map to track in-use obmalloc arenas. Use to replace the old implementation of address_in_range(). The radix tree approach makes it easy to increase pool sizes beyond the OS page size. Boosting the pool and arena size allows obmalloc to handle a significantly higher percentage of requests from its ultra-fast paths.

It also has the advantage of eliminating the memory unsanitary behavior of the previous address_in_range(). The old address_in_range() was marked with the annotations _Py_NO_SANITIZE_ADDRESS, _Py_NO_SANITIZE_THREAD, and _Py_NO_SANITIZE_MEMORY. Those annotations are no longer needed.

To disable the radix tree map, set a preprocessor flag as follows: -DWITH_PYMALLOC_RADIX_TREE=0 .

Co-authored-by: Tim Peters < tim . peters @ gmail . com >

bpo-29988 : Only handle asynchronous exceptions and requests to drop the GIL when returning from a call or on the back edges of loops. Makes sure that __exit__() is always called in with statements, even for interrupts.

bpo-43720 : Document various stdlib deprecations in imp, pkgutil, and importlib.util for removal in Python 3.12.

bpo-43433 : xmlrpc.client.ServerProxy no longer ignores query and fragment in the URL of the server.

bpo-31956 : The index() method of array.array now has optional start and stop parameters.

bpo-40066 : Enum: adjust repr() to show only enum and member name (not value, nor angle brackets) and str() to show only member name. Update and improve documentation to match.

bpo-42136 : Deprecate all module_repr() methods found in importlib as their use is being phased out by Python 3.12.

bpo-35930 : Raising an exception raised in a “future” instance will create reference cycles.

bpo-41369 : Finish updating the vendored libmpdec to version 2.5.1. Patch by Stefan Krah.

bpo-43422 : Revert the _decimal C API which was added in bpo-41324 .

bpo-43577 : Fix deadlock when using ssl.SSLContext debug callback with ssl.SSLContext.sni_callback() .

bpo-43571 : It’s now possible to create MPTCP sockets with IPPROTO_MPTCP

bpo-43542 : image/heic and image/heif were added to mimetypes .

bpo-40645 : The hmac module now uses OpenSSL’s HMAC implementation when digestmod argument is a hash name or builtin hash function.

bpo-43510 : Implement PEP 597 : Add EncodingWarning warning, -X warn_default_encoding option, PYTHONWARNDEFAULTENCODING environment variable and encoding="locale" argument value.

bpo-43521 : ast.unparse can now render NaNs and empty sets.

bpo-42914 : pprint.pprint() gains a new boolean underscore_numbers optional argument to emit integers with thousands separated by an underscore character for improved readability (for example 1_000_000 instead of 1000000 ).

bpo-41361 : rotate() calls are now slightly faster due to faster argument parsing.

bpo-43423 : subprocess.communicate() no longer raises an IndexError when there is an empty stdout or stderr IO buffer during a timeout on Windows.

bpo-27820 : Fixed long-standing bug of smtplib.SMTP where doing AUTH LOGIN with initial_response_ok=False will fail.

The cause is that SMTP.auth_login _always_ returns a password if provided with a challenge string, thus non-compliant with the standard for AUTH LOGIN.

Also fixes bug with the test for smtpd.

bpo-43445 : Add frozen modules to sys.stdlib_module_names . For example, add "_frozen_importlib" and "_frozen_importlib_external" names.

bpo-43245 : Add keyword arguments support to ChainMap.new_child() .

bpo-29982 : Add optional parameter ignore_cleanup_errors to tempfile.TemporaryDirectory() and allow multiple cleanup() attempts. Contributed by C.A.M. Gerlach.

bpo-43428 : Include changes from importlib_metadata 3.7 :

Performance enhancements to distribution discovery.

entry_points only returns unique distributions.

Introduces new EntryPoints object for containing a set of entry points with convenience methods for selecting entry points by group or name. entry_points now returns this object if selection parameters are supplied but continues to return a dict object for compatibility. Users are encouraged to rely on the selection interface. The dict object result is likely to be deprecated in the future.

Added packages_distributions function to return a mapping of packages to the distributions that provide them.

bpo-43332 : Improves the networking efficiency of http.client when using a proxy via set_tunnel() . Fewer small send calls are made during connection setup.

bpo-43420 : Improve performance of fractions.Fraction arithmetics for large components. Contributed by Sergey B. Kirpichev.

bpo-43356 : Allow passing a signal number to _thread.interrupt_main() .

bpo-43399 : Fix ElementTree.extend not working on iterators when using the Python implementation

bpo-43369 : Improve sqlite3 error handling: If sqlite3_column_text() and sqlite3_column_blob() set SQLITE_NOMEM , MemoryError is now raised. Patch by Erlend E. Aasland.

bpo-43368 : Fix a regression introduced in PR 24562, where an empty bytestring was fetched as None instead of b'' in sqlite3 . Patch by Mariusz Felisiak.

bpo-41282 : Fixed stacklevel of DeprecationWarning emitted from import distutils .

bpo-42129 : importlib.resources now honors namespace packages, merging resources from each location in the namespace as introduced in importlib_resources 3.2 and including incidental changes through 5.0.3.

bpo-43295 : datetime.datetime.strptime() now raises ValueError instead of IndexError when matching 'z' with the %z format specifier.

bpo-43125 : Return empty string if base64mime.body_encode receive empty bytes

bpo-43084 : curses.window.enclose() returns now True or False (as was documented) instead of 1 or 0 .

bpo-42994 : Add MIME types for opus, AAC, 3gpp and 3gpp2

bpo-14678 : Add an invalidate_caches() method to the zipimport.zipimporter class to support importlib.invalidate_caches(). Patch by Desmond Cheong.

bpo-42782 : Fail fast in shutil.move() to avoid creating destination directories on failure.

bpo-40066 : Enum’s repr() and str() have changed: repr() is now EnumClass.MemberName and str() is MemberName . Additionally, stdlib Enum’s whose contents are available as module attributes, such as RegexFlag.IGNORECASE , have their repr() as module.name , e.g. re.IGNORECASE .

bpo-26053 : Fixed bug where the pdb interactive run command echoed the args from the shell command line, even if those have been overridden at the pdb prompt.

bpo-24160 : Fixed bug where breakpoints did not persist across multiple debugger sessions in pdb ’s interactive mode.

bpo-40701 : When the tempfile.tempdir global variable is set to a value of type bytes, it is now handled consistently. Previously exceptions could be raised from some tempfile APIs when the directory did not already exist in this situation. Also ensures that the tempfile.gettempdir() and tempfile.gettempdirb() functions always return str and bytes respectively.

bpo-39342 : Expose X509_V_FLAG_ALLOW_PROXY_CERTS as VERIFY_ALLOW_PROXY_CERTS to allow proxy certificate validation as explained in https://www.openssl.org/docs/man1.1.1/man7/proxy-certificates.html .

bpo-31861 : Add builtins.aiter and builtins.anext. Patch by Joshua Bronson (@jab), Daniel Pope (@lordmauve), and Justin Wang (@justin39).

bpo-43199 : Answer “Why is there no goto?” in the Design and History FAQ.

bpo-43407 : Clarified that a result from time.monotonic() , time.perf_counter() , time.process_time() , or time.thread_time() can be compared with the result from any following call to the same function - not just the next immediate call.

bpo-43354 : Fix type documentation for Fault.faultCode ; the type has to be int instead of str .

bpo-41933 : Clarified wording of s * n in the Common Sequence Operations

bpo-37945 : Fix test_getsetlocale_issue1813() of test_locale: skip the test if setlocale() fails. Patch by Victor Stinner.

bpo-41561 : Add workaround for Ubuntu’s custom OpenSSL security level policy.

bpo-43179 : Introduce and correctly use ALIGNOF_X in place of SIZEOF_X for alignment-related code in optimized string routines. Patch by Jessica Clarke.

bpo-43631 : Update macOS, Windows, and CI to OpenSSL 1.1.1k.

bpo-43617 : Improve configure.ac: Check for presence of autoconf-archive package and remove our copies of M4 macros.

bpo-43466 : The configure script now supports --with-openssl-rpath option.

bpo-43372 : Use _freeze_importlib to generate code for the __hello__ module. This approach ensures the code matches the interpreter version. Previously, PYTHON_FOR_REGEN was used to generate the code, which might be wrong. The marshal format for code objects has changed with bpo-42246 , commit 877df851. Update the code and the expected code sizes in ctypes test_frozentable.

bpo-43440 : Build sqlite3 with the R*Tree module enabled. Patch by Erlend E. Aasland.

bpo-42225 : Document that IDLE can fail on Unix either from misconfigured IP masquerade rules or failure displaying complex colored (non-ascii) characters.

bpo-43688 : The limited C API is now supported if Python is built in debug mode (if the Py_DEBUG macro is defined). In the limited C API, the Py_INCREF() and Py_DECREF() functions are now implemented as opaque function calls, rather than accessing directly the PyObject.ob_refcnt member, if Python is built in debug mode and the Py_LIMITED_API macro targets Python 3.10 or newer. It became possible to support the limited C API in debug mode because the PyObject structure is the same in release and debug mode since Python 3.8 (see bpo-36465 ).

The limited C API is still not supported in the --with-trace-refs special build ( Py_TRACE_REFS macro).

bpo-43244 : Remove the pyarena.h header file with functions:

PyArena_New()

PyArena_Free()

PyArena_Malloc()

PyArena_AddPyObject()

These functions were undocumented, excluded from the limited C API, and were only used internally by the compiler. Patch by Victor Stinner.

bpo-43244 : Remove the compiler and parser functions using struct _mod type, because the public AST C API was removed:

PyAST_Compile()

PyAST_CompileEx()

PyAST_CompileObject()

PyFuture_FromAST()

PyFuture_FromASTObject()

PyParser_ASTFromFile()

PyParser_ASTFromFileObject()

PyParser_ASTFromFilename()

PyParser_ASTFromString()

PyParser_ASTFromStringObject()

These functions were undocumented and excluded from the limited C API. Patch by Victor Stinner.

bpo-43244 : Remove ast.h , asdl.h , and Python-ast.h header files. These functions were undocumented and excluded from the limited C API. Most names defined by these header files were not prefixed by Py and so could create names conflicts. For example, Python-ast.h defined a Yield macro which was conflict with the Yield name used by the Windows <winbase.h> header. Use the Python ast module instead. Patch by Victor Stinner.

bpo-43541 : Fix a PyEval_EvalCodeEx() regression: fix reference counting on builtins. Patch by Victor Stinner.

bpo-43244 : Remove the symtable.h header file and the undocumented functions:

PyST_GetScope()

PySymtable_Build()

PySymtable_BuildObject()

PySymtable_Free()

Py_SymtableString()

Py_SymtableStringObject()

The Py_SymtableString() function was part the stable ABI by mistake but it could not be used, because the symtable.h header file was excluded from the limited C API.

The Python symtable module remains available and is unchanged.

bpo-43244 : Remove the PyAST_Validate() function. It is no longer possible to build a AST object ( mod_ty type) with the public C API. The function was already excluded from the limited C API ( PEP 384 ). Patch by Victor Stinner.

Python 3.10.0 alpha 6 ¶

Release date: 2021-03-01

bpo-42967 : Fix web cache poisoning vulnerability by defaulting the query args separator to & , and allowing the user to choose a custom separator.

bpo-43321 : Fix SystemError raised when PyArg_Parse*() is used with # but without PY_SSIZE_T_CLEAN defined.

bpo-36346 : PyArg_Parse*() functions now emits DeprecationWarning when u or Z format is used. See PEP 623 for detail.

bpo-43277 : Add a new PySet_CheckExact() function to the C-API to check if an object is an instance of set but not an instance of a subtype. Patch by Pablo Galindo.

bpo-42990 : The types.FunctionType constructor now inherits the current builtins if the globals dictionary has no "__builtins__" key, rather than using {"None": None} as builtins: same behavior as eval() and exec() functions. Defining a function with def function(...): ... in Python is not affected, globals cannot be overridden with this syntax: it also inherits the current builtins. Patch by Victor Stinner.

bpo-42990 : Functions have a new __builtins__ attribute which is used to look for builtin symbols when a function is executed, instead of looking into __globals__['__builtins__'] . Patch by Mark Shannon and Victor Stinner.

bpo-43149 : Improve the error message in the parser for exception groups without parentheses. Patch by Pablo Galindo.

bpo-43121 : Fixed an incorrect SyntaxError message for missing comma in literals. Patch by Pablo Galindo.

bpo-42819 : readline : Explicitly disable bracketed paste in the interactive interpreter, even if it’s set in the inputrc, is enabled by default (eg GNU Readline 8.1), or a user calls readline.read_init_file() . The Python REPL has not implemented bracketed paste support. Also, bracketed mode writes the "\x1b[?2004h" escape sequence into stdout which causes test failures in applications that don’t support it. It can still be explicitly enabled by calling readline.parse_and_bind("set enable-bracketed-paste on") . Patch by Dustin Rodrigues.

bpo-42808 : Simple calls to type(object) are now faster due to the vectorcall calling convention. Patch by Dennis Sweeney.

bpo-42217 : Make the compiler merges same co_code and co_linetable objects in a module like already did for co_consts.

bpo-41972 : Substring search functions such as str1 in str2 and str2.find(str1) now sometimes use the “Two-Way” string comparison algorithm to avoid quadratic behavior on long strings.

bpo-42128 : Implement PEP 634 (structural pattern matching). Patch by Brandt Bucher.

bpo-40692 : In the concurrent.futures.ProcessPoolExecutor , validate that multiprocess.synchronize() is available on a given platform and rely on that check in the concurrent.futures test suite so we can run tests that are unrelated to ProcessPoolExecutor on those platforms.

bpo-38302 : If object.__ipow__() returns NotImplemented , the operator will correctly fall back to object.__pow__() and object.__rpow__() as expected.

bpo-43316 : The python -m gzip command line application now properly fails when detecting an unsupported extension. It exits with a non-zero exit code and prints an error message to stderr.

bpo-43317 : Set the chunk size for the gzip module main function to io.DEFAULT_BUFFER_SIZE. This is slightly faster than the 1024 bytes constant that was used previously.

bpo-43146 : Handle None in single-arg versions of print_exception() and format_exception() .

bpo-43260 : Fix TextIOWrapper can not flush internal buffer forever after very large text is written.

bpo-43258 : Prevent needless allocation of sqlite3 aggregate function context when no rows match an aggregate query. Patch by Erlend E. Aasland.

bpo-43251 : Improve sqlite3 error handling: sqlite3_column_name() failures now result in MemoryError . Patch by Erlend E. Aasland.

bpo-40956 : Fix segfault in sqlite3.Connection.backup() if no argument was provided. The regression was introduced by PR 23838. Patch by Erlend E. Aasland.

bpo-43172 : The readline module now passes its tests when built directly against libedit. Existing irreconcilable API differences remain in readline.get_begidx() and readline.get_endidx() behavior based on libreadline vs libedit use.

bpo-43163 : Fix a bug in codeop that was causing it to not ask for more input when multi-line snippets have unclosed parentheses. Patch by Pablo Galindo

bpo-43162 : deprecate unsupported ability to access enum members as attributes of other enum members

bpo-43146 : Fix recent regression in None argument handling in traceback module functions.

bpo-43102 : The namedtuple __new__ method had its __builtins__ set to None instead of an actual dictionary. This created problems for introspection tools.

bpo-43106 : Added O_EVTONLY , O_FSYNC , O_SYMLINK and O_NOFOLLOW_ANY for macOS. Patch by Donghee Na.

bpo-42960 : Adds resource.RLIMIT_KQUEUES constant from FreeBSD to the resource module.

bpo-42151 : Make the pure Python implementation of xml.etree.ElementTree behave the same as the C implementation ( _elementree ) regarding default attribute values (by not setting specified_attributes=1 ).

bpo-29753 : In ctypes, now packed bitfields are calculated properly and the first item of packed bitfields is now shrank correctly.

bpo-27646 : Clarify that ‘yield from <expr>’ works with any iterable, not just iterators.

bpo-36346 : Update some deprecated unicode APIs which are documented as “will be removed in 4.0” to “3.12”. See PEP 623 for detail.

bpo-43288 : Fix test_importlib to correctly skip Unicode file tests if the filesystem does not support them.

bpo-43174 : Windows build now uses /utf-8 compiler option.

bpo-43103 : Add a new configure --without-static-libpython option to not build the libpythonMAJOR.MINOR.a static library and not install the python.o object file.

bpo-13501 : The configure script can now use libedit instead of readline with the command line option --with-readline=editline .

bpo-42603 : Make configure script use pkg-config to detect the location of Tcl/Tk headers and libraries, used to build tkinter.

On macOS, a Tcl/Tk configuration provided by pkg-config will be preferred over Tcl/Tk frameworks installed in /{System/,}Library/Frameworks . If both exist and the latter is preferred, the appropriate --with-tcltk-* configuration options need to be explicitly set.

bpo-39448 : Add the “regen-frozen” makefile target that regenerates the code for the frozen __hello__ module.

bpo-43155 : PyCMethod_New() is now present in python3.lib .

bpo-41837 : Update macOS installer build to use OpenSSL 1.1.1j.

bpo-43283 : Document why printing to IDLE’s Shell is often slower than printing to a system terminal and that it can be made faster by pre-formatting a single string before printing.

bpo-43278 : Always put compiler and system information on the first line of the REPL welcome message.

bpo-43270 : Remove the private _PyErr_OCCURRED() macro: use the public PyErr_Occurred() function instead.

bpo-35134 : Move odictobject.h, parser_interface.h, picklebufobject.h, pydebug.h, and pyfpe.h into the cpython/ directory. They must not be included directly, as they are already included by Python.h: Include Files .

bpo-35134 : Move pyarena.h, pyctype.h, and pytime.h into the cpython/ directory. They must not be included directly, as they are already included by Python.h: Include Files .

bpo-40170 : PyExceptionClass_Name() is now always declared as a function, in order to hide implementation details. The macro accessed PyTypeObject.tp_name directly. Patch by Erlend E. Aasland.

bpo-43239 : The PyCFunction_New() function is now exported in the ABI when compiled with -fvisibility=hidden .

bpo-40170 : PyIter_Check() is now always declared as a function, in order to hide implementation details. The macro accessed PyTypeObject.tp_iternext directly. Patch by Erlend E. Aasland.

bpo-40170 : Convert PyDescr_IsData() macro to a function to hide implementation details: The macro accessed PyTypeObject.tp_descr_set directly. Patch by Erlend E. Aasland.

bpo-43181 : Convert PyObject_TypeCheck() macro to a static inline function. Patch by Erlend E. Aasland.

Python 3.10.0 alpha 5 ¶

Release date: 2021-02-02

bpo-42938 : Avoid static buffers when computing the repr of ctypes.c_double and ctypes.c_longdouble values.

bpo-42990 : Refactor the PyEval_ family of functions.

An new function _PyEval_Vector is added to simplify calls to Python from C.

_PyEval_EvalCodeWithName is removed

PyEval_EvalCodeEx is retained as part of the API, but is not used internally

bpo-38631 : Replace Py_FatalError() calls in the compiler with regular SystemError exceptions. Patch by Victor Stinner.

bpo-42997 : Improve error message for missing “:” before blocks. Patch by Pablo Galindo.

bpo-43017 : Improve error message in the parser when using un-parenthesised tuples in comprehensions. Patch by Pablo Galindo.

bpo-42986 : Fix parser crash when reporting syntax errors in f-string with newlines. Patch by Pablo Galindo.

bpo-40176 : Syntax errors for unterminated string literals now point to the start of the string instead of reporting EOF/EOL.

bpo-42927 : The inline cache for LOAD_ATTR now also optimizes access to attributes defined by __slots__ . This makes reading such attribute up to 30% faster.

bpo-42864 : Improve error messages in the parser when parentheses are not closed. Patch by Pablo Galindo.

bpo-42924 : Fix bytearray repetition incorrectly copying data from the start of the buffer, even if the data is offset within the buffer (e.g. after reassigning a slice at the start of the bytearray to a shorter byte string).

bpo-42882 : Fix the _PyUnicode_FromId() function (_Py_IDENTIFIER(var) API) when Py_Initialize() / Py_Finalize() is called multiple times: preserve _PyRuntime.unicode_ids.next_index value.

bpo-42827 : Fix a crash when working out the error line of a SyntaxError in some multi-line expressions.

bpo-42823 : frame.f_lineno is correct even if frame.f_trace is set to True

bpo-37324 : Remove deprecated aliases to Collections Abstract Base Classes from the collections module.

bpo-41994 : Fixed possible leak in import when sys.modules is not a dict .

bpo-27772 : In string formatting, preceding the width field by '0' no longer affects the default alignment for strings.

bpo-43108 : Fixed a reference leak in the curses module. Patch by Pablo Galindo

bpo-43077 : Update the bundled pip to 21.0.1 and setuptools to 52.0.0.

bpo-41282 : Deprecate distutils in documentation and add warning on import.

bpo-43014 : Improve performance of tokenize by 20-30%. Patch by Anthony Sottile.

bpo-42323 : Fix math.nextafter() for NaN on AIX.

bpo-42955 : Add sys.stdlib_module_names , containing the list of the standard library module names. Patch by Victor Stinner.

bpo-42944 : Fix random.Random.sample when counts argument is not None .

bpo-42934 : Use TracebackException ’s new compact param in TestResult to reduce time and memory consumed by traceback formatting.

bpo-42931 : Add randbytes() to random.__all__ .

bpo-38250 : [Enum] Flags consisting of a single bit are now considered canonical, and will be the only flags returned from listing and iterating over a Flag class or a Flag member. Multi-bit flags are considered aliases; they will be returned from lookups and operations that result in their value. Iteration for both Flag and Flag members is in definition order.

bpo-42877 : Added the compact parameter to the constructor of traceback.TracebackException to reduce time and memory for use cases that only need to call TracebackException.format() and TracebackException.format_exception_only() .

bpo-42923 : The Py_FatalError() function and the faulthandler module now dump the list of extension modules on a fatal error.

bpo-42848 : Removed recursion from TracebackException to allow it to handle long exception chains.

bpo-42901 : [Enum] move member creation from EnumMeta.__new__ to _proto_member.__set_name__ , allowing members to be created and visible in __init_subclass__ .

bpo-42780 : Fix os.set_inheritable() for O_PATH file descriptors on Linux.

bpo-42866 : Fix a reference leak in the getcodec() function of CJK codecs. Patch by Victor Stinner.

bpo-42846 : Convert the 6 CJK codec extension modules (_codecs_cn, _codecs_hk, _codecs_iso2022, _codecs_jp, _codecs_kr and _codecs_tw) to the multiphase initialization API ( PEP 489 ). Patch by Victor Stinner.

bpo-42851 : remove __init_subclass__ support for Enum members

bpo-42834 : Make internal caches of the _json module compatible with subinterpreters.

bpo-41748 : Fix HTMLParser parsing rules for element attributes containing commas with spaces. Patch by Karl Dubost.

bpo-40810 : Require SQLite 3.7.15 or newer. Patch by Erlend E. Aasland.

bpo-1635741 : Convert the _multibytecodec extension module (CJK codecs) to multi-phase initialization ( PEP 489 ). Patch by Erlend E. Aasland.

bpo-42802 : The distutils bdist_wininst command deprecated in Python 3.8 has been removed. The distutils bdist_wheel command is now recommended to distribute binary packages on Windows.

bpo-24464 : The undocumented built-in function sqlite3.enable_shared_cache is now deprecated, scheduled for removal in Python 3.12. Its use is strongly discouraged by the SQLite3 documentation. Patch by Erlend E. Aasland.

bpo-42384 : Make pdb populate sys.path[0] exactly the same as regular python execution.

bpo-42383 : Fix pdb: previously pdb would fail to restart the debugging target if it was specified using a relative path and the current directory changed.

bpo-42005 : Fix CLI of cProfile and profile to catch BrokenPipeError .

bpo-41604 : Don’t decrement the reference count of the previous user_ptr when set_panel_userptr fails.

bpo-41149 : Allow executing callables that have a boolean value of False when passed to Threading.thread as the target. Patch contributed by Barney Stratford.

bpo-38307 : Add an ‘end_lineno’ attribute to the Class and Function objects that appear in the tree returned by pyclbr functions. This and the existing ‘lineno’ attribute define the extent of class and def statements. Patch by Aviral Srivastava.

bpo-39273 : The BUTTON5_* constants are now exposed in the curses module if available.

bpo-33289 : Correct call to tkinter.colorchooser to return RGB triplet of ints instead of floats. Patch by Cheryl Sabella.

bpo-40304 : Fix doc for type(name, bases, dict). Patch by Boris Verkhovskiy and Éric Araujo.

bpo-42811 : Updated importlib.util.resolve_name() doc to use __spec__.parent instead of __package__. (Thanks Yair Frid.)

bpo-40823 : Use unittest.TestLoader().loadTestsFromTestCase() instead of unittest.makeSuite() in sqlite3 tests. Patch by Erlend E. Aasland.

bpo-40810 : In sqlite3 , fix CheckTraceCallbackContent for SQLite pre 3.7.15.

bpo-43031 : Pass --timeout=$(TESTTIMEOUT) option to the default profile task ./python -m test --pgo command.

bpo-36143 : make regen-all now also runs regen-keyword . Patch by Victor Stinner.

bpo-42874 : Removed the grep -q and -E flags in the tzpath validation section of the configure script to better accommodate users of some platforms (specifically Solaris 10).

bpo-31904 : Add library search path by wr-cc in add_cross_compiling_paths() for VxWorks.

bpo-42856 : Add --with-wheel-pkg-dir=PATH option to the ./configure script. If specified, the ensurepip module looks for setuptools and pip wheel packages in this directory: if both are present, these wheel packages are used instead of ensurepip bundled wheel packages.

Some Linux distribution packaging policies recommend against bundling dependencies. For example, Fedora installs wheel packages in the /usr/share/python-wheels/ directory and don’t install the ensurepip._bundled package.

bpo-41837 : Updated Windows installer to include OpenSSL 1.1.1i

bpo-42584 : Upgrade Windows installer to use SQLite 3.34.0.

bpo-42504 : Ensure that the value of sysconfig.get_config_var(‘MACOSX_DEPLOYMENT_TARGET’) is always a string, even in when the value is parsable as an integer.

bpo-43008 : Make IDLE invoke sys.excepthook() in normal, 2-process mode. Patch by Ken Hilton.

bpo-33065 : Fix problem debugging user classes with __repr__ method.

bpo-23544 : Disable Debug=>Stack Viewer when user code is running or Debugger is active, to prevent hang or crash. Patch by Zackery Spytz.

bpo-32631 : Finish zzdummy example extension module: make menu entries work; add docstrings and tests with 100% coverage.

bpo-42979 : When Python is built in debug mode (with C assertions), calling a type slot like sq_length ( __len__() in Python) now fails with a fatal error if the slot succeeded with an exception set, or failed with no exception set. The error message contains the slot, the type name, and the current exception (if an exception is set). Patch by Victor Stinner.

bpo-43030 : Fixed a compiler warning in Py_UNICODE_ISSPACE() on platforms with signed wchar_t .

Python 3.10.0 alpha 4 ¶

Release date: 2021-01-04

bpo-42814 : Fix undefined behavior in Objects/genericaliasobject.c .

bpo-42806 : Fix the column offsets for f-strings ast nodes surrounded by parentheses and for nodes that spawn multiple lines. Patch by Pablo Galindo.

bpo-40631 : Fix regression where a single parenthesized starred expression was a valid assignment target.

bpo-27794 : Improve the error message for failed writes/deletes to property objects. When possible, the attribute name is now shown. Patch provided by Yurii Karabas.

bpo-42745 : Make the type attribute lookup cache per-interpreter. Patch by Victor Stinner.

bpo-42246 : Jumps to jumps are not eliminated when it would break PEP 626.

bpo-42246 : Make sure that the f_lasti and f_lineno attributes of a frame are set correctly when an exception is raised or re-raised. Required for PEP 626.

bpo-32381 : The coding cookie (ex: # coding: latin1 ) is now ignored in the command passed to the -c command line option. Patch by Victor Stinner.

bpo-30858 : Improve error location in expressions that contain assignments. Patch by Pablo Galindo and Lysandros Nikolaou.

bpo-42615 : Remove jump commands made redundant by the deletion of unreachable bytecode blocks

bpo-42639 : Make the atexit module state per-interpreter. It is now safe have more than one atexit module instance. Patch by Donghee Na and Victor Stinner.

bpo-32381 : Fix encoding name when running a .pyc file on Windows: PyRun_SimpleFileExFlags() now uses the correct encoding to decode the filename.

bpo-42195 : The __args__ of the parameterized generics for typing.Callable and collections.abc.Callable are now consistent. The __args__ for collections.abc.Callable are now flattened while typing.Callable ’s have not changed. To allow this change, types.GenericAlias can now be subclassed and collections.abc.Callable ’s __class_getitem__ will now return a subclass of types.GenericAlias . Tests for typing were also updated to not subclass things like Callable[..., T] as that is not a valid base class. Finally, both Callable s no longer validate their argtypes , in Callable[[argtypes], resulttype] to prepare for PEP 612 . Patch by Ken Jin.

bpo-40137 : Convert functools module to use PyType_FromModuleAndSpec() .

bpo-40077 : Convert array to use heap types, and establish module state for these.

bpo-42008 : Fix _random.Random() seeding.

bpo-1635741 : Port the pyexpat extension module to multi-phase initialization ( PEP 489 ).

bpo-40521 : Make the Unicode dictionary of interned strings compatible with subinterpreters. Patch by Victor Stinner.

bpo-39465 : Make _PyUnicode_FromId() function compatible with subinterpreters. Each interpreter now has an array of identifier objects (interned strings decoded from UTF-8). Patch by Victor Stinner.

bpo-42257 : Handle empty string in variable executable in platform.libc_ver()

bpo-42772 : randrange() now raises a TypeError when step is specified without a stop argument. Formerly, it silently ignored the step argument.

bpo-42759 : Fixed equality comparison of tkinter.Variable and tkinter.font.Font . Objects which belong to different Tcl interpreters are now always different, even if they have the same name.

bpo-42756 : Configure LMTP Unix-domain socket to use socket global default timeout when a timeout is not explicitly provided.

bpo-23328 : Allow / character in username, password fields on _PROXY envars.

bpo-42740 : typing.get_args() and typing.get_origin() now support PEP 604 union types and PEP 612 additions to Callable .

bpo-42655 : subprocess extra_groups is now correctly passed into setgroups() system call.

bpo-42727 : EnumMeta.__prepare__ now accepts **kwds to properly support __init_subclass__

bpo-38308 : Add optional weights to statistics.harmonic_mean() .

bpo-42721 : When simple query dialogs ( tkinter.simpledialog ), message boxes ( tkinter.messagebox ) or color choose dialog ( tkinter.colorchooser ) are created without arguments master and parent , and the default root window is not yet created, and NoDefaultRoot() was not called, a new temporal hidden root window will be created automatically. It will not be set as the default root window and will be destroyed right after closing the dialog window. It will help to use these simple dialog windows in programs which do not need other GUI.

bpo-25246 : Optimized collections.deque.remove() .

bpo-35728 : Added a root parameter to tkinter.font.nametofont() .

bpo-15303 : tkinter supports now widgets with boolean value False.

bpo-42681 : Fixed range checks for color and pair numbers in curses .

bpo-42685 : Improved placing of simple query windows in Tkinter (such as tkinter.simpledialog.askinteger() ). They are now centered at the center of the parent window if it is specified and shown, otherwise at the center of the screen.

bpo-9694 : Argparse help no longer uses the confusing phrase, “optional arguments”. It uses “options” instead.

bpo-1635741 : Port the _thread extension module to the multiphase initialization API ( PEP 489 ) and convert its static types to heap types.

bpo-37961 : Fix crash in tracemalloc.Traceback.__repr__() (regressed in Python 3.9).

bpo-42630 : tkinter functions and constructors which need a default root window raise now RuntimeError with descriptive message instead of obscure AttributeError or NameError if it is not created yet or cannot be created automatically.

bpo-42639 : atexit._run_exitfuncs() now logs callback exceptions using sys.unraisablehook , rather than logging them directly into sys.stderr and raise the last exception.

bpo-42644 : logging.disable will now validate the types and value of its parameter. It also now accepts strings representing the levels (as does logging.setLevel ) instead of only the numerical values.

bpo-42639 : At Python exit, if a callback registered with atexit.register() fails, its exception is now logged. Previously, only some exceptions were logged, and the last exception was always silently ignored.

bpo-36541 : Fixed lib2to3.pgen2 to be able to parse PEP-570 positional only argument syntax.

bpo-42382 : In importlib.metadata : - EntryPoint objects now expose a .dist object referencing the Distribution when constructed from a Distribution . - Add support for package discovery under package normalization rules. - The object returned by metadata() now has a formally defined protocol called PackageMetadata with declared support for the .get_all() method. - Synced with importlib_metadata 3.3.

bpo-41877 : A check is added against misspellings of autospect, auto_spec and set_spec being passed as arguments to patch, patch.object and create_autospec.

bpo-39717 : [tarfile] update nested exception raising to use from None or from e

bpo-41877 : AttributeError for suspected misspellings of assertions on mocks are now pointing out that the cause are misspelled assertions and also what to do if the misspelling is actually an intended attribute name. The unittest.mock document is also updated to reflect the current set of recognised misspellings.

bpo-41559 : Implemented PEP 612 : added ParamSpec and Concatenate to typing . Patch by Ken Jin.

bpo-42385 : StrEnum: fix _generate_next_value_ to return a str

bpo-31904 : Define THREAD_STACK_SIZE for VxWorks.

bpo-34750 : [Enum] _EnumDict.update() is now supported.

bpo-42517 : Enum: private names do not become members / do not generate errors – they remain normal attributes

bpo-42678 : Enum : call __init_subclass__ after members have been added

bpo-28964 : ast.literal_eval() adds line number information (if available) in error message for malformed nodes.

bpo-42470 : random.sample() no longer warns on a sequence which is also a set.

bpo-31904 : posixpath.expanduser() returns the input path unchanged if user home directory is None on VxWorks.

bpo-42388 : Fix subprocess.check_output(…, input=None) behavior when text=True to be consistent with that of the documentation and universal_newlines=True.

bpo-34463 : Fixed discrepancy between traceback and the interpreter in formatting of SyntaxError with lineno not set ( traceback was changed to match interpreter).

bpo-42393 : Raise OverflowError instead of silent truncation in socket.ntohs() and socket.htons() . Silent truncation was deprecated in Python 3.7. Patch by Erlend E. Aasland

bpo-42222 : Harmonized random.randrange() argument handling to match range() .

The integer test and conversion in randrange() now uses operator.index() .

Non-integer arguments to randrange() are deprecated.

The ValueError is deprecated in favor of a TypeError .

It now runs a little faster than before.

(Contributed by Raymond Hettinger and Serhiy Storchaka.)

bpo-42163 : Restore compatibility for uname_result around deepcopy and _replace.

bpo-42090 : zipfile.Path.joinpath now accepts arbitrary arguments, same as pathlib.Path.joinpath .

bpo-1635741 : Port the _csv module to the multi-phase initialization API ( PEP 489 ).

bpo-42059 : typing.TypedDict types created using the alternative call-style syntax now correctly respect the total keyword argument when setting their __required_keys__ and __optional_keys__ class attributes.

bpo-41960 : Add globalns and localns parameters to the inspect.signature() and inspect.Signature.from_callable() .

bpo-41907 : fix format() behavior for IntFlag

bpo-41891 : Ensure asyncio.wait_for waits for task completion

bpo-24792 : Fixed bug where zipimporter sometimes reports an incorrect cause of import errors.

bpo-31904 : Fix site and sysconfig modules for VxWorks RTOS which has no home directories.

bpo-41462 : Add os.set_blocking() support for VxWorks RTOS.

bpo-40219 : Lowered tkinter.ttk.LabeledScale dummy widget to prevent hiding part of the content label.

bpo-37193 : Fixed memory leak in socketserver.ThreadingMixIn introduced in Python 3.7.

bpo-39068 : Fix initialization race condition in a85encode() and b85encode() in base64 . Patch by Brandon Stansbury.

bpo-17140 : Add documentation for the multiprocessing.pool.ThreadPool class.

bpo-34398 : Prominently feature listings from the glossary in documentation search results. Patch by Ammar Askar.

bpo-42794 : Update test_nntplib to use official group name of news.aioe.org for testing. Patch by Donghee Na.

bpo-31904 : Skip some asyncio tests on VxWorks.

bpo-42641 : Enhance test_select.test_select() : it now takes 500 milliseconds rather than 10 seconds. Use Python rather than a shell to make the test more portable.

bpo-31904 : Skip some tests in _test_all_chown_common() on VxWorks.

bpo-42199 : Fix bytecode helper assertNotInBytecode.

bpo-41443 : Add more attribute checking in test_posix.py

bpo-31904 : Disable os.popen and impacted tests on VxWorks

bpo-41439 : Port test_ssl and test_uuid to VxWorks RTOS.

bpo-42692 : Fix __builtin_available check on older compilers. Patch by Joshua Root.

bpo-27640 : Added --disable-test-modules option to the configure script: don’t build nor install test modules. Patch by Xavier de Gaye, Thomas Petazzoni and Peixing Xin.

bpo-42604 : Now all platforms use a value for the “EXT_SUFFIX” build variable derived from SOABI (for instance in freeBSD, “EXT_SUFFIX” is now “.cpython-310d.so” instead of “.so”). Previously only Linux, Mac and VxWorks were using a value for “EXT_SUFFIX” that included “SOABI”.

bpo-42598 : Fix implicit function declarations in configure which could have resulted in incorrect configuration checks. Patch contributed by Joshua Root.

bpo-31904 : Enable libpython3.so for VxWorks.

bpo-29076 : Add fish shell support to macOS installer.

bpo-42361 : Update macOS installer build to use Tcl/Tk 8.6.11 (rc2, expected to be final release).

bpo-41837 : Update macOS installer build to use OpenSSL 1.1.1i.

bpo-42584 : Update macOS installer to use SQLite 3.34.0.

bpo-42726 : Fixed Python 3 compatibility issue with gdb/libpython.py handling of attribute dictionaries.

bpo-42613 : Fix freeze.py tool to use the prope config and library directories. Patch by Victor Stinner.

bpo-42591 : Export the Py_FrozenMain() function: fix a Python 3.9.0 regression. Python 3.9 uses -fvisibility=hidden and the function was not exported explicitly and so not exported.

bpo-32381 : Remove the private _Py_fopen() function which is no longer needed. Use _Py_wfopen() or _Py_fopen_obj() instead. Patch by Victor Stinner.

bpo-1635741 : Port resource extension module to module state

bpo-42111 : Update the xxlimited module to be a better example of how to use the limited C API.

bpo-40052 : Fix an alignment build warning/error in function PyVectorcall_Function() . Patch by Andreas Schneider, Antoine Pitrou and Petr Viktorin.

Python 3.10.0 alpha 3 ¶

Release date: 2020-12-07

bpo-40791 : Add volatile to the accumulator variable in hmac.compare_digest , making constant-time-defeating optimizations less likely.

bpo-42576 : types.GenericAlias will now raise a TypeError when attempting to initialize with a keyword argument. Previously, this would cause the interpreter to crash if the interpreter was compiled with debug symbols. This does not affect interpreters compiled for release. Patch by Ken Jin.

bpo-42536 : Several built-in and standard library types now ensure that their internal result tuples are always tracked by the garbage collector :

collections.OrderedDict.items()

dict.items()

enumerate()

functools.reduce()

itertools.combinations()

itertools.combinations_with_replacement()

itertools.permutations()

itertools.product()

itertools.zip_longest()

Previously, they could have become untracked by a prior garbage collection. Patch by Brandt Bucher.

bpo-42500 : Improve handling of exceptions near recursion limit. Converts a number of Fatal Errors in RecursionErrors.

bpo-42246 : PEP 626: After a return, the f_lineno attribute of a frame is always the last line executed.

bpo-42435 : Speed up comparison of bytes objects with non-bytes objects when option -b is specified. Speed up comparison of bytarray objects with non-buffer object.

bpo-1635741 : Port the _warnings extension module to the multi-phase initialization API ( PEP 489 ). Patch by Victor Stinner.

bpo-41686 : On Windows, the SIGINT event, _PyOS_SigintEvent() , is now created even if Python is configured to not install signal handlers (if PyConfig.install_signal_handlers equals to 0, or Py_InitializeEx(0) ).

bpo-42381 : Allow assignment expressions in set literals and set comprehensions as per PEP 572. Patch by Pablo Galindo.

bpo-42202 : Change function parameters annotations internal representation to tuple of strings. Patch provided by Yurii Karabas.

bpo-42374 : Fix a regression introduced by the new parser, where an unparenthesized walrus operator was not allowed within generator expressions.

bpo-42316 : Allow an unparenthesized walrus in subscript indexes.

bpo-42349 : Make sure that the compiler front-end produces a well-formed control flow graph. Be more aggressive in the compiler back-end, as it is now safe to do so.

bpo-42296 : On Windows, fix a regression in signal handling which prevented to interrupt a program using CTRL+C. The signal handler can be run in a thread different than the Python thread, in which case the test deciding if the thread can handle signals is wrong.

bpo-42332 : types.GenericAlias objects can now be the targets of weakrefs.

bpo-42282 : Optimise constant subexpressions that appear as part of named expressions (previously the AST optimiser did not descend into named expressions). Patch by Nick Coghlan.

bpo-42266 : Fixed a bug with the LOAD_ATTR opcode cache that was not respecting monkey-patching a class-level attribute to make it a descriptor. Patch by Pablo Galindo.

bpo-40077 : Convert queue to use heap types.

bpo-42246 : Improved accuracy of line tracing events and f_lineno attribute of Frame objects. See PEP 626 for details.

bpo-40077 : Convert mmap to use heap types.

bpo-42233 : Allow GenericAlias objects to use union type expressions . This allows expressions like list[int] | dict[float, str] where previously a TypeError would have been thrown. This also fixes union type expressions not de-duplicating GenericAlias objects. (Contributed by Ken Jin in bpo-42233 .)

bpo-26131 : The import system triggers a ImportWarning when it falls back to using load_module() .

bpo-5054 : CGIHTTPRequestHandler.run_cgi() HTTP_ACCEPT improperly parsed. Replace the special purpose getallmatchingheaders with generic get_all method and add relevant tests.

Original Patch by Martin Panter. Modified by Senthil Kumaran.

bpo-42562 : Fix issue when dis failed to parse function that has no line numbers. Patch provided by Yurii Karabas.

bpo-17735 : inspect.findsource() now raises OSError instead of IndexError when co_lineno of a code object is greater than the file length. This can happen, for example, when a file is edited after it was imported. PR by Irit Katriel.

bpo-42116 : Fix handling of trailing comments by inspect.getsource() .

bpo-42532 : Remove unexpected call of __bool__ when passing a spec_arg argument to a Mock.

bpo-38200 : Added itertools.pairwise()

bpo-41818 : Fix test_master_read() so that it succeeds on all platforms that either raise OSError or return b”” upon reading from master.

bpo-42487 : ChainMap.__iter__ no longer calls __getitem__ on underlying maps

bpo-42482 : TracebackException no longer holds a reference to the exception’s traceback object. Consequently, instances of TracebackException for equivalent but non-equal exceptions now compare as equal.

bpo-41818 : Make test_openpty() avoid unexpected success due to number of rows and/or number of columns being == 0.

bpo-42392 : Remove loop parameter from asyncio.subprocess and asyncio.tasks functions. Patch provided by Yurii Karabas.

bpo-42392 : Remove loop parameter from asyncio.open_connection and asyncio.start_server functions. Patch provided by Yurii Karabas.

bpo-28468 : Add platform.freedesktop_os_release() function to parse freedesktop.org os-release files.

bpo-42299 : Removed the formatter module, which was deprecated in Python 3.4. It is somewhat obsolete, little used, and not tested. It was originally scheduled to be removed in Python 3.6, but such removals were delayed until after Python 2.7 EOL. Existing users should copy whatever classes they use into their code. Patch by Donghee Na and and Terry J. Reedy.

bpo-26131 : Deprecate zipimport.zipimporter.load_module() in favour of exec_module().

bpo-41818 : Updated tests for the pty library. test_basic() has been changed to test_openpty(); this additionally checks if slave termios and slave winsize are being set properly by pty.openpty(). In order to add support for FreeBSD, NetBSD, OpenBSD, and Darwin, this also adds test_master_read(), which demonstrates that pty.spawn() should not depend on an OSError to exit from its copy loop.

bpo-42392 : Remove loop parameter from __init__ in all asyncio.locks and asyncio.Queue classes. Patch provided by Yurii Karabas.

bpo-15450 : Make filecmp.dircmp respect subclassing. Now the filecmp.dircmp.subdirs behaves as expected when subclassing dircmp.

bpo-42413 : The exception socket.timeout is now an alias of TimeoutError .

bpo-31904 : Support signal module on VxWorks.

bpo-42406 : We fixed an issue in pickle.whichmodule in which importing multiprocessing could change the how pickle identifies which module an object belongs to, potentially breaking the unpickling of those objects.

bpo-42403 : Simplify the importlib external bootstrap code: importlib._bootstrap_external now uses regular imports to import builtin modules. When it is imported, the builtin __import__() function is already fully working and so can be used to import builtin modules like sys . Patch by Victor Stinner.

bpo-1635741 : Convert _sre module types to heap types (PEP 384). Patch by Erlend E. Aasland.

bpo-42375 : subprocess module update for DragonFlyBSD support.

bpo-41713 : Port the _signal extension module to the multi-phase initialization API ( PEP 489 ). Patch by Victor Stinner and Mohamed Koubaa.

bpo-37205 : time.time() , time.perf_counter() and time.monotonic() functions can no longer fail with a Python fatal error, instead raise a regular Python exception on failure.

bpo-42328 : Fixed tkinter.ttk.Style.map() . The function accepts now the representation of the default state as empty sequence (as returned by Style.map() ). The structure of the result is now the same on all platform and does not depend on the value of wantobjects .

bpo-42345 : Fix various issues with typing.Literal parameter handling (flatten, deduplicate, use type to cache key). Patch provided by Yurii Karabas.

bpo-37205 : time.perf_counter() on Windows and time.monotonic() on macOS are now system-wide. Previously, they used an offset computed at startup to reduce the precision loss caused by the float type. Use time.perf_counter_ns() and time.monotonic_ns() added in Python 3.7 to avoid this precision loss.

bpo-42318 : Fixed support of non-BMP characters in tkinter on macOS.

bpo-42350 : Fix the threading.Thread class at fork: do nothing if the thread is already stopped (ex: fork called at Python exit). Previously, an error was logged in the child process.

bpo-42333 : Port _ssl extension module to heap types.

bpo-42014 : The onerror callback from shutil.rmtree now receives correct function when os.open fails.

bpo-42237 : Fix os.sendfile() on illumos.

bpo-42308 : Add threading.__excepthook__ to allow retrieving the original value of threading.excepthook() in case it is set to a broken or a different value. Patch by Mario Corchero.

bpo-42131 : Implement PEP 451/spec methods on zipimport.zipimporter: find_spec(), create_module(), and exec_module().

This also allows for the documented deprecation of find_loader(), find_module(), and load_module().

bpo-41877 : Mock objects which are not unsafe will now raise an AttributeError if an attribute with the prefix asert, aseert, or assrt is accessed, in addition to this already happening for the prefixes assert or assret.

bpo-42264 : sqlite3.OptimizedUnicode has been undocumented and obsolete since Python 3.3, when it was made an alias to str . It is now deprecated, scheduled for removal in Python 3.12.

bpo-42251 : Added threading.gettrace() and threading.getprofile() to retrieve the functions set by threading.settrace() and threading.setprofile() respectively. Patch by Mario Corchero.

bpo-42249 : Fixed writing binary Plist files larger than 4 GiB.

bpo-42236 : On Unix, the os.device_encoding() function now returns 'UTF-8' rather than the device encoding if the Python UTF-8 Mode is enabled.

bpo-41754 : webbrowser: Ignore NotADirectoryError when calling xdg-settings .

bpo-42183 : Fix a stack overflow error for asyncio Task or Future repr().

The overflow occurs under some circumstances when a Task or Future recursively returns itself.

bpo-42140 : Improve asyncio.wait function to create the futures set just one time.

bpo-42133 : Update various modules in the stdlib to fall back on __spec__.loader when __loader__ isn’t defined on a module.

bpo-26131 : The load_module() methods found in importlib now trigger a DeprecationWarning .

bpo-39825 : Windows: Change sysconfig.get_config_var('EXT_SUFFIX') to the expected full platform_tag.extension format. Previously it was hard-coded to .pyd , now it is compatible with distutils.sysconfig and will result in something like .cp38-win_amd64.pyd . This brings windows into conformance with the other platforms.

bpo-26389 : The traceback.format_exception() , traceback.format_exception_only() , and traceback.print_exception() functions can now take an exception object as a positional-only argument.

bpo-41889 : Enum: fix regression involving inheriting a multiply inherited enum

bpo-41861 : Convert sqlite3 to use heap types (PEP 384). Patch by Erlend E. Aasland.

bpo-40624 : Added support for the XPath != operator in xml.etree

bpo-28850 : Fix pprint.PrettyPrinter.format() overrides being ignored for contents of small containers. The pprint._safe_repr() function was removed.

bpo-41625 : Expose the splice() as os.splice() in the os module. Patch by Pablo Galindo

bpo-34215 : Clarify the error message for asyncio.IncompleteReadError when expected is None .

bpo-41543 : Add async context manager support for contextlib.nullcontext.

bpo-21041 : pathlib.PurePath.parents now supports negative indexing. Patch contributed by Yaroslav Pankovych.

bpo-41332 : Added missing connect_accepted_socket() method to asyncio.AbstractEventLoop .

bpo-12800 : Extracting a symlink from a tarball should succeed and overwrite the symlink if it already exists. The fix is to remove the existing file or symlink before extraction. Based on patch by Chris AtLee, Jeffrey Kintscher, and Senthil Kumaran.

bpo-40968 : urllib.request and http.client now send http/1.1 ALPN extension during TLS handshake when no custom context is supplied.

bpo-41001 : Add os.eventfd() to provide a low level interface for Linux’s event notification file descriptor.

bpo-40816 : Add AsyncContextDecorator to contextlib to support async context manager as a decorator.

bpo-40550 : Fix time-of-check/time-of-action issue in subprocess.Popen.send_signal.

bpo-39411 : Add an is_async identifier to pyclbr ’s Function objects. Patch by Batuhan Taskaya

bpo-35498 : Add slice support to pathlib.PurePath.parents .

bpo-42238 : Tentative to deprecate make suspicious by first removing it from the CI and documentation builds, but keeping it around for manual uses.

bpo-42153 : Fix the URL for the IMAP protocol documents.

bpo-41028 : Language and version switchers, previously maintained in every cpython branches, are now handled by docsbuild-script.

bpo-41473 : Re-enable test_gdb on gdb 9.2 and newer: https://bugzilla.redhat.com/show_bug.cgi?id=1866884 bug is fixed in gdb 10.1.

bpo-42553 : Fix test_asyncio.test_call_later() race condition: don’t measure asyncio performance in the call_later() unit test. The test failed randomly on the CI.

bpo-31904 : Fix test_netrc on VxWorks: create temporary directories using temp_cwd().

bpo-31904 : skip test_getaddrinfo_ipv6_scopeid_symbolic and test_getnameinfo_ipv6_scopeid_symbolic on VxWorks

bpo-31904 : skip test_test of test_mailcap on VxWorks

bpo-31904 : add shell requirement for test_pipes

bpo-31904 : skip some tests related to fifo on VxWorks

bpo-31904 : Fix test_doctest.py failures for VxWorks.

bpo-40754 : Include _testinternalcapi module in Windows installer for test suite

bpo-41561 : test_ssl: skip test_min_max_version_mismatch when TLS 1.0 is not available

bpo-31904 : Fix os module failures for VxWorks RTOS.

bpo-31904 : Fix fifo test cases for VxWorks RTOS.

bpo-31904 : remove libnet dependency from detect_socket() for VxWorks

bpo-42398 : Fix a race condition in “make regen-all” when make -jN option is used to run jobs in parallel. The clinic.py script now only use atomic write to write files. Moveover, generated files are now left unchanged if the content does not change, to not change the file modification time.

bpo-41617 : Fix building pycore_bitutils.h internal header on old clang version without __builtin_bswap16() (ex: Xcode 4.6.3 on Mac OS X 10.7). Patch by Joshua Root and Victor Stinner.

bpo-38823 : It is no longer possible to build the _ctypes extension module without wchar_t type: remove CTYPES_UNICODE macro. Anyway, the wchar_t type is required to build Python. Patch by Victor Stinner.

bpo-42087 : Support was removed for AIX 5.3 and below. See bpo-40680 .

bpo-40998 : Addressed three compiler warnings found by undefined behavior sanitizer (ubsan).

bpo-42120 : Remove macro definition of copysign (to _copysign ) in headers.

bpo-38506 : The Windows launcher now properly handles Python 3.10 when listing installed Python versions.

bpo-42504 : Fix build on macOS Big Sur when MACOSX_DEPLOYMENT_TARGET=11

bpo-41116 : Ensure distutils.unixxcompiler.find_library_file can find system provided libraries on macOS 11.

bpo-41100 : Add support for macOS 11 and Apple Silicon systems.

It is now possible to build “Universal 2” binaries using “–enable-universalsdk –with-universal-archs=universal2”.

Binaries build on later macOS versions can be deployed back to older versions (tested up to macOS 10.9), when using the correct deployment target. This is tested using Xcode 11 and later.

bpo-42232 : Added Darwin specific madvise options to mmap module.

bpo-38443 : The --enable-universalsdk and --with-universal-archs options for the configure script now check that the specified architectures can be used.

bpo-42508 : Keep IDLE running on macOS. Remove obsolete workaround that prevented running files with shortcuts when using new universal2 installers built on macOS 11.

bpo-42426 : Fix reporting offset of the RE error in searchengine.

bpo-42415 : Get docstrings for IDLE calltips more often by using inspect.getdoc.

bpo-42212 : The smelly.py script now also checks the Python dynamic library and extension modules, not only the Python static library. Make also the script more verbose: explain what it does.

bpo-36310 : Allow Tools/i18n/pygettext.py to detect calls to gettext in f-strings.

bpo-42423 : The PyType_FromSpecWithBases() and PyType_FromModuleAndSpec() functions now accept a single class as the bases argument.

bpo-1635741 : Port select extension module to multiphase initialization ( PEP 489 ).

bpo-1635741 : Port _posixsubprocess extension module to multiphase initialization ( PEP 489 ).

bpo-1635741 : Port _posixshmem extension module to multiphase initialization ( PEP 489 )

bpo-1635741 : Port _struct extension module to multiphase initialization ( PEP 489 )

bpo-1635741 : Port spwd extension module to multiphase initialization ( PEP 489 )

bpo-1635741 : Port gc extension module to multiphase initialization ( PEP 489 )

bpo-1635741 : Port _queue extension module to multiphase initialization ( PEP 489 )

bpo-39573 : Convert Py_TYPE() and Py_SIZE() back to macros to allow using them as an l-value. Many third party C extension modules rely on the ability of using Py_TYPE() and Py_SIZE() to set an object type and size: Py_TYPE(obj) = type; and Py_SIZE(obj) = size; .

bpo-1635741 : Port symtable extension module to multiphase initialization ( PEP 489 )

bpo-1635741 : Port grp and pwd extension modules to multiphase initialization ( PEP 489 )

bpo-1635741 : Port _random extension module to multiphase initialization ( PEP 489 )

bpo-1635741 : Port _hashlib extension module to multiphase initialization ( PEP 489 )

bpo-41713 : Removed the undocumented PyOS_InitInterrupts() function. Initializing Python already implicitly installs signal handlers: see PyConfig.install_signal_handlers . Patch by Victor Stinner.

bpo-40170 : The Py_TRASHCAN_BEGIN macro no longer accesses PyTypeObject attributes, but now can get the condition by calling the new private _PyTrash_cond() function which hides implementation details.

bpo-42260 : Py_GetPath() , Py_GetPrefix() , Py_GetExecPrefix() , Py_GetProgramFullPath() , Py_GetPythonHome() and Py_GetProgramName() functions now return NULL if called before Py_Initialize() (before Python is initialized). Use the new Python Initialization Configuration API to get the Python Path Configuration. . Patch by Victor Stinner.

bpo-42260 : The PyConfig_Read() function now only parses PyConfig.argv arguments once: PyConfig.parse_argv is set to 2 after arguments are parsed. Since Python arguments are strippped from PyConfig.argv , parsing arguments twice would parse the application options as Python options.

bpo-42262 : Added Py_NewRef() and Py_XNewRef() functions to increment the reference count of an object and return the object. Patch by Victor Stinner.

bpo-42260 : When Py_Initialize() is called twice, the second call now updates more sys attributes for the configuration, rather than only sys.argv . Patch by Victor Stinner.

bpo-41832 : The PyType_FromModuleAndSpec() function now accepts NULL tp_doc slot.

bpo-1635741 : Added PyModule_AddObjectRef() function: similar to PyModule_AddObject() but don’t steal a reference to the value on success. Patch by Victor Stinner.

bpo-42171 : The METH_FASTCALL calling convention is added to the limited API. The functions PyModule_AddType() , PyType_FromModuleAndSpec() , PyType_GetModule() and PyType_GetModuleState() are added to the limited API on Windows.

bpo-42085 : Add dedicated entry to PyAsyncMethods for sending values

bpo-41073 : PyType_GetSlot() can now accept static types.

bpo-30459 : PyList_SET_ITEM() , PyTuple_SET_ITEM() and PyCell_SET() macros can no longer be used as l-value or r-value. For example, x = PyList_SET_ITEM(a, b, c) and PyList_SET_ITEM(a, b, c) = x now fail with a compiler error. It prevents bugs like if (PyList_SET_ITEM (a, b, c) < 0) ... test. Patch by Zackery Spytz and Victor Stinner.

Python 3.10.0 alpha 2 ¶

Release date: 2020-11-03

bpo-42103 : Prevented potential DoS attack via CPU and RAM exhaustion when processing malformed Apple Property List files in binary format.

bpo-42051 : The plistlib module no longer accepts entity declarations in XML plist files to avoid XML vulnerabilities. This should not affect users as entity declarations are not used in regular plist files.

bpo-42236 : If the nl_langinfo(CODESET) function returns an empty string, Python now uses UTF-8 as the filesystem encoding. Patch by Victor Stinner.

bpo-42218 : Fixed a bug in the PEG parser that was causing crashes in debug mode. Now errors are checked in left-recursive rules to avoid cases where such errors do not get handled in time and appear as long-distance crashes in other places.

bpo-42214 : Fixed a possible crash in the PEG parser when checking for the ‘!=’ token in the barry_as_flufl rule. Patch by Pablo Galindo.

bpo-42206 : Propagate and raise the errors caused by PyAST_Validate() in the parser.

bpo-41796 : The ast module internal state is now per interpreter. Patch by Victor Stinner.

bpo-42143 : Fix handling of errors during creation of PyFunctionObject , which resulted in operations on uninitialized memory. Patch by Yonatan Goldschmidt.

bpo-41659 : Fix a bug in the parser, where a curly brace following a primary didn’t fail immediately. This led to invalid expressions like a {b} to throw a SyntaxError with a wrong offset, or invalid expressions ending with a curly brace like a { to not fail immediately in the REPL.

bpo-42150 : Fix possible buffer overflow in the new parser when checking for continuation lines. Patch by Pablo Galindo.

bpo-42123 : Run the parser two times. On the first run, disable all the rules that only generate better error messages to gain performance. If there’s a parse failure, run the parser a second time with those enabled.

bpo-42093 : The LOAD_ATTR instruction now uses new “per opcode cache” mechanism and it is about 36% faster now. Patch by Pablo Galindo and Yury Selivanov.

bpo-42030 : Support for the legacy AIX-specific shared library loading support has been removed. All versions of AIX since 4.3 have supported and defaulted to using the common Unix mechanism instead.

bpo-41984 : The garbage collector now tracks all user-defined classes. Patch by Brandt Bucher.

bpo-41993 : Fixed potential issues with removing not completely initialized module from sys.modules when import fails.

bpo-41979 : Star-unpacking is now allowed for with item’s targets in the PEG parser.

bpo-41974 : Removed special methods __int__ , __float__ , __floordiv__ , __mod__ , __divmod__ , __rfloordiv__ , __rmod__ and __rdivmod__ of the complex class. They always raised a TypeError .

bpo-41902 : Micro optimization when compute sq_item and mp_subscript of range . Patch by Donghee Na.

bpo-41894 : When loading a native module and a load failure occurs, prevent a possible UnicodeDecodeError when not running in a UTF-8 locale by decoding the load error message using the current locale’s encoding.

bpo-41902 : Micro optimization for range.index if step is 1. Patch by Donghee Na.

bpo-41435 : Add sys._current_exceptions() function to retrieve a dictionary mapping each thread’s identifier to the topmost exception currently active in that thread at the time the function is called.

bpo-38605 : Enable from __future__ import annotations ( PEP 563 ) by default. The values found in __annotations__ dicts are now strings, e.g. {"x": "int"} instead of {"x": int} .

bpo-35455 : On Solaris, thread_time() is now implemented with gethrvtime() because clock_gettime(CLOCK_THREAD_CPUTIME_ID) is not always available. Patch by Jakub Kulik.

bpo-42233 : The repr() of typing types containing Generic Alias Types previously did not show the parameterized types in the GenericAlias . They have now been changed to do so.

bpo-29566 : binhex.binhex() consistently writes macOS 9 line endings.

bpo-26789 : The logging.FileHandler class now keeps a reference to the builtin open() function to be able to open or reopen the file during Python finalization. Fix errors like: NameError: name 'open' is not defined . Patch by Victor Stinner.

bpo-42157 : Removed the unicodedata.ucnhash_CAPI attribute which was an internal PyCapsule object. The related private _PyUnicode_Name_CAPI structure was moved to the internal C API. Patch by Victor Stinner.

bpo-42157 : Convert the unicodedata extension module to the multiphase initialization API ( PEP 489 ) and convert the unicodedata.UCD static type to a heap type. Patch by Mohamed Koubaa and Victor Stinner.

bpo-42146 : Fix memory leak in subprocess.Popen() in case an uid (gid) specified in user ( group , extra_groups ) overflows uid_t ( gid_t ).

bpo-42103 : InvalidFileException and RecursionError are now the only errors caused by loading malformed binary Plist file (previously ValueError and TypeError could be raised in some specific cases).

bpo-41490 : In importlib.resources , .path method is more aggressive about releasing handles to zipfile objects early, enabling use-cases like certifi to leave the context open but delete the underlying zip file.

bpo-41052 : Pickling heap types implemented in C with protocols 0 and 1 raises now an error instead of producing incorrect data.

bpo-42089 : In importlib.metadata.PackageNotFoundError , make reference to the package metadata being missing to improve the user experience.

bpo-41491 : plistlib: fix parsing XML plists with hexadecimal integer values

bpo-42065 : Fix an incorrectly formatted error from _codecs.charmap_decode() when called with a mapped value outside the range of valid Unicode code points. PR by Max Bernstein.

bpo-41966 : Fix pickling pure Python datetime.time subclasses. Patch by Dean Inwood.

bpo-19270 : sched.scheduler.cancel() will now cancel the correct event, if two events with same priority are scheduled for the same time. Patch by Bar Harel.

bpo-28660 : textwrap.wrap() now attempts to break long words after hyphens when break_long_words=True and break_on_hyphens=True .

bpo-35823 : Use vfork() instead of fork() for subprocess.Popen() on Linux to improve performance in cases where it is deemed safe.

bpo-42043 : Add support for zipfile.Path inheritance. zipfile.Path.is_file() now returns False for non-existent names. zipfile.Path objects now expose a .filename attribute and rely on that to resolve .name and .parent when the Path object is at the root of the zipfile.

bpo-42021 : Fix possible ref leaks in sqlite3 module init.

bpo-39101 : Fixed tests using IsolatedAsyncioTestCase from hanging on BaseExceptions.

bpo-41976 : Fixed a bug that was causing ctypes.util.find_library() to return None when triying to locate a library in an environment when gcc>=9 is available and ldconfig is not. Patch by Pablo Galindo

bpo-41943 : Fix bug where TestCase.assertLogs doesn’t correctly filter messages by level.

bpo-41923 : Implement PEP 613 , introducing typing.TypeAlias annotation.

bpo-41905 : A new function in abc: update_abstractmethods to re-calculate an abstract class’s abstract status. In addition, dataclass has been changed to call this function.

bpo-23706 : Added newline parameter to pathlib.Path.write_text() .

bpo-41876 : Tkinter font class repr uses font name

bpo-41831 : str() for the type attribute of the tkinter.Event object always returns now the numeric code returned by Tk instead of the name of the event type.

bpo-39337 : encodings.normalize_encoding() now ignores non-ASCII characters.

bpo-41747 : Ensure all methods that generated from dataclasses.dataclass() objects now have the proper __qualname__ attribute referring to the class they belong to. Patch by Batuhan Taskaya.

bpo-30681 : Handle exceptions caused by unparsable date headers when using email “default” policy. Patch by Tim Bell, Georges Toth

bpo-41586 : Add F_SETPIPE_SZ and F_GETPIPE_SZ to fcntl module. Allow setting pipesize on subprocess.Popen.

bpo-41229 : Add contextlib.aclosing for deterministic cleanup of async generators which is analogous to contextlib.closing for non-async generators. Patch by Joongi Kim and John Belmonte.

bpo-16396 : Allow ctypes.wintypes to be imported on non-Windows systems.

bpo-4356 : Add a key function to the bisect module.

bpo-40592 : shutil.which() now ignores empty entries in PATHEXT instead of treating them as a match.

bpo-40492 : Fix --outfile for cProfile / profile not writing the output file in the original directory when the program being profiled changes the working directory. PR by Anthony Sottile.

bpo-34204 : The shelve module now uses pickle.DEFAULT_PROTOCOL by default instead of pickle protocol 3 .

bpo-27321 : Fixed KeyError exception when flattening an email to a string attempts to replace a non-existent Content-Transfer-Encoding header.

bpo-38976 : The http.cookiejar module now supports the parsing of cookies in CURL-style cookiejar files through MozillaCookieJar on all platforms. Previously, such cookie entries would be silently ignored when loading a cookiejar with such entries.

Additionally, the HTTP Only attribute is persisted in the object, and will be correctly written to file if the MozillaCookieJar object is subsequently dumped.

bpo-42061 : Document __format__ functionality for IP addresses.

bpo-41910 : Document the default implementation of object.__eq__ .

bpo-42010 : Clarify that subscription expressions are also valid for certain classes and types in the standard library, and for user-defined classes and types if the classmethod __class_getitem__() is provided.

bpo-41805 : Documented generic alias type and types.GenericAlias . Also added an entry in glossary for generic types .

bpo-39693 : Fix tarfile’s extractfile documentation

bpo-39416 : Document some restrictions on the default string representations of numeric classes.

bpo-41739 : Fix test_logging.test_race_between_set_target_and_flush(): the test now waits until all threads complete to avoid leaking running threads.

bpo-41970 : Avoid a test failure in test_lib2to3 if the module has already imported at the time the test executes. Patch by Pablo Galindo.

bpo-41944 : Tests for CJK codecs no longer call eval() on content received via HTTP.

bpo-41306 : Fixed a failure in test_tk.test_widgets.ScaleTest happening when executing the test with Tk 8.6.10.

bpo-38980 : Add -fno-semantic-interposition to both the compile and link line when building with --enable-optimizations . Patch by Victor Stinner and Pablo Galindo.

bpo-38439 : Updates the icons for IDLE in the Windows Store package.

bpo-38252 : Use 8-byte step to detect ASCII sequence in 64-bit Windows build.

bpo-39107 : Update Tcl and Tk to 8.6.10 in Windows installer.

bpo-41557 : Update Windows installer to use SQLite 3.33.0.

bpo-38324 : Avoid Unicode errors when accessing certain locale data on Windows.

bpo-41471 : Ignore invalid prefix lengths in system proxy excludes.

bpo-33987 : Mostly finish using ttk widgets, mainly for editor, settings, and searches. Some patches by Mark Roseman.

bpo-40511 : Typing opening and closing parentheses inside the parentheses of a function call will no longer cause unnecessary “flashing” off and on of an existing open call-tip, e.g. when typed in a string literal.

bpo-38439 : Add a 256×256 pixel IDLE icon to the Windows .ico file. Created by Andrew Clover. Remove the low-color gif variations from the .ico file.

bpo-42157 : The private _PyUnicode_Name_CAPI structure of the PyCapsule API unicodedata.ucnhash_CAPI has been moved to the internal C API. Patch by Victor Stinner.

bpo-42015 : Fix potential crash in deallocating method objects when dynamically allocated PyMethodDef ’s lifetime is managed through the self argument of a PyCFunction .

bpo-40423 : The subprocess module and os.closerange will now use the close_range(low, high, flags) syscall when it is available for more efficient closing of ranges of descriptors.

bpo-41845 : PyObject_GenericGetDict() is available again in the limited API when targeting 3.10 or later.

bpo-40422 : Add _Py_closerange function to provide performant closing of a range of file descriptors.

bpo-41986 : Py_FileSystemDefaultEncodeErrors and Py_UTF8Mode are available again in limited API.

bpo-41756 : Add PyIter_Send function to allow sending value into generator/coroutine/iterator without raising StopIteration exception to signal return.

bpo-41784 : Added PyUnicode_AsUTF8AndSize to the limited C API.

Python 3.10.0 alpha 1 ¶

Release date: 2020-10-05

bpo-41304 : Fixes python3x._pth being ignored on Windows, caused by the fix for bpo-29778 ( CVE-2020-15801 ).

bpo-41162 : Audit hooks are now cleared later during finalization to avoid missing events.

bpo-29778 : Ensure python3.dll is loaded from correct locations when Python is embedded ( CVE-2020-15523 ).

bpo-41004 : The __hash__() methods of ipaddress.IPv4Interface and ipaddress.IPv6Interface incorrectly generated constant hash values of 32 and 128 respectively. This resulted in always causing hash collisions. The fix uses hash() to generate hash values for the tuple of (address, mask length, network address).

bpo-39603 : Prevent http header injection by rejecting control characters in http.client.putrequest(…).

bpo-41909 : Fixed stack overflow in issubclass() and isinstance() when getting the __bases__ attribute leads to infinite recursion.

bpo-41922 : Speed up calls to reversed() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-41873 : Calls to float() are now faster due to the vectorcall calling convention. Patch by Dennis Sweeney.

bpo-41870 : Speed up calls to bool() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-1635741 : Port the _bisect module to the multi-phase initialization API ( PEP 489 ).

bpo-39934 : Correctly count control blocks in ‘except’ in compiler. Ensures that a syntax error, rather a fatal error, occurs for deeply nested, named exception handlers.

bpo-41780 : Fix __dir__() of types.GenericAlias . Patch by Batuhan Taskaya.

bpo-1635741 : Port the _lsprof extension module to multi-phase initialization ( PEP 489 ).

bpo-1635741 : Port the cmath extension module to multi-phase initialization ( PEP 489 ).

bpo-1635741 : Port the _scproxy extension module to multi-phase initialization ( PEP 489 ).

bpo-1635741 : Port the termios extension module to multi-phase initialization ( PEP 489 ).

bpo-1635741 : Convert the _sha256 extension module types to heap types.

bpo-41690 : Fix a possible stack overflow in the parser when parsing functions and classes with a huge amount of arguments. Patch by Pablo Galindo.

bpo-1635741 : Port the _overlapped extension module to multi-phase initialization ( PEP 489 ).

bpo-1635741 : Port the _curses_panel extension module to multi-phase initialization ( PEP 489 ).

bpo-1635741 : Port the _opcode extension module to multi-phase initialization ( PEP 489 ).

bpo-41681 : Fixes the wrong error description in the error raised by using 2 , in format string in f-string and str.format() .

bpo-41675 : The implementation of signal.siginterrupt() now uses sigaction() (if it is available in the system) instead of the deprecated siginterrupt() . Patch by Pablo Galindo.

bpo-41670 : Prevent line trace being skipped on platforms not compiled with USE_COMPUTED_GOTOS . Fixes issue where some lines nested within a try-except block were not being traced on Windows.

bpo-41654 : Fix a crash that occurred when destroying subclasses of MemoryError . Patch by Pablo Galindo.

bpo-1635741 : Port the zlib extension module to multi-phase initialization ( PEP 489 ).

bpo-41631 : The _ast module uses again a global state. Using a module state per module instance is causing subtle practical problems. For example, the Mercurial project replaces the __import__() function to implement lazy import, whereas Python expected that import _ast always return a fully initialized _ast module.

bpo-40077 : Convert _operator to use PyType_FromSpec() .

bpo-1653741 : Port _sha3 to multi-phase init. Convert static types to heap types.

bpo-1635741 : Port the _blake2 extension module to the multi-phase initialization API ( PEP 489 ).

bpo-41533 : Free the stack allocated in va_build_stack if do_mkstack fails and the stack is not a small_stack .

bpo-41531 : Fix a bug that was dropping keys when compiling dict literals with more than 0xFFFF elements. Patch by Pablo Galindo.

bpo-41525 : The output of python --help contains now only ASCII characters.

bpo-1635741 : Port the _sha1 , _sha512 , and _md5 extension modules to multi-phase initialization API ( PEP 489 ).

bpo-41431 : Optimize dict_merge() for copying dict (e.g. dict(d) and {}.update(d) ).

bpo-41428 : Implement PEP 604. This supports (int | str) etc. in place of Union[str, int].

bpo-41340 : Removed fallback implementation for strdup .

bpo-38156 : Handle interrupts that come after EOF correctly in PyOS_StdioReadline .

bpo-41342 : round() with integer argument is now faster (9–60%).

bpo-41334 : Constructors str() , bytes() and bytearray() are now faster (around 30–40% for small objects).

bpo-41295 : Resolve a regression in CPython 3.8.4 where defining “__setattr__” in a multi-inheritance setup and calling up the hierarchy chain could fail if builtins/extension types were involved in the base types.

bpo-41323 : Bytecode optimizations are performed directly on the control flow graph. This will result in slightly more compact code objects in some circumstances.

bpo-41247 : Always cache the running loop holder when running asyncio.set_running_loop .

bpo-41252 : Fix incorrect refcounting in _ssl.c’s _servername_callback() .

bpo-1635741 : Port multiprocessing to multi-phase initialization

bpo-1635741 : Port winapi to multiphase initialization

bpo-41215 : Use non-NULL default values in the PEG parser keyword list to overcome a bug that was preventing Python from being properly compiled when using the XLC compiler. Patch by Pablo Galindo.

bpo-41218 : Python 3.8.3 had a regression where compiling with ast.PyCF_ALLOW_TOP_LEVEL_AWAIT would aggressively mark list comprehension with CO_COROUTINE. Now only list comprehension making use of async/await will tagged as so.

bpo-1635741 : Port faulthandler to multiphase initialization.

bpo-1635741 : Port sha256 to multiphase initialization

bpo-41175 : Guard against a NULL pointer dereference within bytearrayobject triggered by the bytearray() + bytearray() operation.

bpo-41100 : add arm64 to the allowable Mac OS arches in mpdecimal.h

bpo-41094 : Fix decoding errors with audit when open files with non-ASCII names on non-UTF-8 locale.

bpo-39960 : The “hackcheck” that prevents sneaking around a type’s __setattr__() by calling the superclass method was rewritten to allow C implemented heap types.

bpo-41084 : Prefix the error message with ‘f-string: ‘, when parsing an f-string expression which throws a SyntaxError .

bpo-40521 : Empty frozensets are no longer singletons.

bpo-41076 : Pre-feed the parser with the location of the f-string expression, not the f-string itself, which allows us to skip the shifting of the AST node locations after the parsing is completed.

bpo-41056 : Fixes a reference to deallocated stack space during startup when constructing sys.path involving a relative symlink when code was supplied via -c. (discovered via Coverity)

bpo-41061 : Fix incorrect expressions and asserts in hashtable code and tests.

bpo-41052 : Opt out serialization/deserialization for _random.Random

bpo-40939 : Rename PyPegen* functions to PyParser* , so that we can remove the old set of PyParser* functions that were using the old parser, but keep everything backwards-compatible.

bpo-35975 : Stefan Behnel reported that cf_feature_version is used even when PyCF_ONLY_AST is not set. This is against the intention and against the documented behavior, so it’s been fixed.

bpo-40939 : Remove the remaining files from the old parser and the symbol module.

bpo-40077 : Convert _bz2 to use PyType_FromSpec() .

bpo-41006 : The encodings.latin_1 module is no longer imported at startup. Now it is only imported when it is the filesystem encoding or the stdio encoding.

bpo-40636 : zip() now supports PEP 618 ’s strict parameter, which raises a ValueError if the arguments are exhausted at different lengths. Patch by Brandt Bucher.

bpo-1635741 : Port _gdbm to multiphase initialization.

bpo-40985 : Fix a bug that caused the SyntaxError text to be empty when a file ends with a line ending in a line continuation character (i.e. backslash). The error text should contain the text of the last line.

bpo-40958 : Fix a possible buffer overflow in the PEG parser when gathering information for emitting syntax errors. Patch by Pablo Galindo.

bpo-1635741 : Port _dbm to multiphase initialization.

bpo-40957 : Fix refleak in _Py_fopen_obj() when PySys_Audit() fails

bpo-40950 : Add a state to the nis module ( PEP 3121 ) and apply the multiphase initialization. Patch by Donghee Na.

bpo-40947 : The Python Path Configuration now takes PyConfig.platlibdir in account.

bpo-40939 : Remove the old parser, the parser module and all associated support code, command-line options and environment variables. Patch by Pablo Galindo.

bpo-40847 : Fix a bug where a line with only a line continuation character is not considered a blank line at tokenizer level. In such cases, more than a single NEWLINE token was emitted. The old parser was working around the issue, but the new parser threw a SyntaxError for valid input due to this. For example, an empty line following a line continuation character was interpreted as a SyntaxError .

bpo-40890 : Each dictionary view now has a mapping attribute that provides a types.MappingProxyType wrapping the original dictionary. Patch contributed by Dennis Sweeney.

bpo-40889 : Improved the performance of symmetric difference operations on dictionary item views. Patch by Dennis Sweeney.

bpo-40904 : Fix possible segfault in the new PEG parser when parsing f-string containing yield statements with no value ( f"{yield}" ). Patch by Pablo Galindo

bpo-40903 : Fixed a possible segfault in the new PEG parser when producing error messages for invalid assignments of the form p=p= . Patch by Pablo Galindo

bpo-40880 : Fix invalid memory read in the new parser when checking newlines in string literals. Patch by Pablo Galindo.

bpo-40883 : Fix memory leak in when parsing f-strings in the new parser. Patch by Pablo Galindo

bpo-40870 : Raise ValueError when validating custom AST’s where the constants True , False and None are used within a ast.Name node.

bpo-40854 : Allow overriding sys.platlibdir via a new PYTHONPLATLIBDIR environment variable.

bpo-40826 : Fix GIL usage in PyOS_Readline() : lock the GIL to set an exception and pass the Python thread state when checking if there is a pending signal.

bpo-1635741 : Port fcntl to multiphase initialization.

bpo-19468 : Delete unnecessary instance check in importlib.reload(). Patch by Furkan Önder.

bpo-40824 : Unexpected errors in calling the __iter__ method are no longer masked by TypeError in the in operator and functions contains() , indexOf() and countOf() of the operator module.

bpo-40792 : Attributes start , stop and step of the range object now always has exact type int . Previously, they could have been an instance of a subclass of int .

bpo-40780 : Fix a corner case where g-style string formatting of a float failed to remove trailing zeros.

bpo-38964 : When there’s a SyntaxError in the expression part of an fstring, the filename attribute of the SyntaxError gets correctly set to the name of the file the fstring resides in.

bpo-40750 : Support the “-d” debug flag in the new PEG parser. Patch by Pablo Galindo

bpo-40217 : Instances of types created with PyType_FromSpecWithBases() will no longer automatically visit their class object when traversing references in the garbage collector. The user is expected to manually visit the object’s class. Patch by Pablo Galindo.

bpo-39573 : Py_TYPE() is changed to the inline static function. Patch by Donghee Na.

bpo-40696 : Fix a hang that can arise after generator.throw() due to a cycle in the exception context chain.

bpo-40521 : Each interpreter now its has own free lists, singletons and caches:

Free lists: float, tuple, list, dict, frame, context, asynchronous generator, MemoryError.

Singletons: empty tuple, empty bytes string, empty Unicode string, single byte character, single Unicode (latin1) character.

Slice cache.

They are no longer shared by all interpreters.

bpo-40679 : Certain TypeError messages about missing or extra arguments now include the function’s qualified name . Patch by Dennis Sweeney.

bpo-29590 : Make the stack trace correct after calling generator.throw() on a generator that has yielded from a yield from .

bpo-4022 : Improve performance of generators by not raising internal StopIteration.

bpo-1635741 : Port mmap to multiphase initialization.

bpo-1635741 : Port _lzma to multiphase initialization.

bpo-37999 : Builtin and extension functions that take integer arguments no longer accept Decimal s, Fraction s and other objects that can be converted to integers only with a loss (e.g. that have the __int__() method but do not have the __index__() method).

bpo-29882 : Add int.bit_count() , counting the number of ones in the binary representation of an integer. Patch by Niklas Fiekas.

bpo-36982 : Use ncurses extended color functions when available to support terminals with 256 colors, and add the new function curses.has_extended_color_support() to indicate whether extended color support is provided by the underlying ncurses library.

bpo-19569 : Add the private macros _Py_COMP_DIAG_PUSH , _Py_COMP_DIAG_IGNORE_DEPR_DECLS , and _Py_COMP_DIAG_POP .

bpo-26680 : The int type now supports the x.is_integer() method for compatibility with float.

bpo-41900 : C14N 2.0 serialisation in xml.etree.ElementTree failed for unprefixed attributes when a default namespace was defined.

bpo-41887 : Strip leading spaces and tabs on ast.literal_eval() . Also document stripping of spaces and tabs for eval() .

bpo-41773 : Note in documentation that random.choices() doesn’t support non-finite weights, raise ValueError when given non-finite weights.

bpo-41840 : Fix a bug in the symtable module that was causing module-scope global variables to not be reported as both local and global. Patch by Pablo Galindo.

bpo-41842 : Add codecs.unregister() function to unregister a codec search function.

bpo-40564 : In zipfile.Path , mutate the passed ZipFile object type instead of making a copy. Prevents issues when both the local copy and the caller’s copy attempt to close the same file handle.

bpo-40670 : More reliable validation of statements in timeit.Timer . It now accepts “empty” statements (only whitespaces and comments) and rejects misindentent statements.

bpo-41833 : The threading.Thread constructor now uses the target name if the target argument is specified but the name argument is omitted.

bpo-41817 : fix tkinter.EventType Enum so all members are strings, and none are tuples

bpo-41810 : types.EllipsisType , types.NotImplementedType and types.NoneType have been reintroduced, providing a new set of types readily interpretable by static type checkers.

bpo-41815 : Fix SQLite3 segfault when backing up closed database. Patch contributed by Peter David McCormick.

bpo-41816 : StrEnum added: it ensures that all members are already strings or string candidates

bpo-41517 : fix bug allowing Enums to be extended via multiple inheritance

bpo-39587 : use the correct mix-in data type when constructing Enums

bpo-41792 : Add is_typeddict function to typing.py to check if a type is a TypedDict class

Previously there was no way to check that without using private API. See the relevant issue in python/typing .

bpo-41789 : Honor object overrides in Enum class creation (specifically, __str__ , __repr__ , __format__ , and __reduce_ex__ ).

bpo-32218 : enum.Flag and enum.IntFlag members are now iterable.

bpo-39651 : Fix a race condition in the call_soon_threadsafe() method of asyncio.ProactorEventLoop : do nothing if the self-pipe socket has been closed.

bpo-1635741 : Port the mashal extension module to the multi-phase initialization API ( PEP 489 ).

bpo-1635741 : Port the _string extension module to the multi-phase initialization API ( PEP 489 ).

bpo-41732 : Added an iterator to memoryview .

bpo-41720 : Fixed turtle.Vec2D.__rmul__() for arguments which are not int or float.

bpo-41696 : Fix handling of debug mode in asyncio.run() . This allows setting PYTHONASYNCIODEBUG or -X dev to enable asyncio debug mode when using asyncio.run() .

bpo-41687 : Fix implementation of sendfile to be compatible with Solaris.

bpo-41662 : No longer override exceptions raised in __len__() of a sequence of parameters in sqlite3 with ProgrammingError .

bpo-39010 : Restarting a ProactorEventLoop on Windows no longer logs spurious ConnectionResetErrors .

bpo-41638 : ProgrammingError message for absent parameter in sqlite3 contains now the name of the parameter instead of its index when parameters are supplied as a dict.

bpo-41662 : Fixed crash when mutate list of parameters during iteration in sqlite3 .

bpo-41513 : Improved the accuracy of math.hypot(). Internally, each step is computed with extra precision so that the result is now almost always correctly rounded.

bpo-41609 : The pdb whatis command correctly reports instance methods as ‘Method’ rather than ‘Function’.

bpo-39994 : Fixed pprint’s handling of dict subclasses that override __repr__.

bpo-32751 : When cancelling the task due to a timeout, asyncio.wait_for() will now wait until the cancellation is complete also in the case when timeout is <= 0, like it does with positive timeouts.

bpo-37658 : asyncio.wait_for() now properly handles races between cancellation of itself and the completion of the wrapped awaitable.

bpo-40782 : Change the method asyncio.AbstractEventLoop.run_in_executor to not be a coroutine.

bpo-41520 : Fix codeop regression that prevented turning compile warnings into errors.

bpo-41528 : turtle uses math module functions to convert degrees to radians and vice versa and to calculate vector norm

bpo-41513 : Minor algorithmic improvement to math.hypot() and math.dist() giving small gains in speed and accuracy.

bpo-41503 : Fixed a race between setTarget and flush in logging.handlers.MemoryHandler.

bpo-41497 : Fix potential UnicodeDecodeError in dis module.

bpo-41467 : On Windows, fix asyncio recv_into() return value when the socket/pipe is closed ( BrokenPipeError ): return 0 rather than an empty byte string ( b'' ).

bpo-41425 : Make tkinter doc example runnable.

bpo-41421 : Make an algebraic simplification to random.paretovariate(). It now is slightly less subject to round-off error and is slightly faster. Inputs that used to cause ZeroDivisionError now cause an OverflowError instead.

bpo-41440 : Add os.cpu_count() support for VxWorks RTOS.

bpo-41316 : Fix the tarfile module to write only basename of TAR file to GZIP compression header.

bpo-41384 : Raise TclError instead of TypeError when an unknown option is passed to tkinter.OptionMenu.

bpo-41317 : Use add_done_callback() in asyncio.loop.sock_accept() to unsubscribe reader early on cancellation.

bpo-41364 : Reduce import overhead of uuid .

bpo-35328 : Set the environment variable VIRTUAL_ENV_PROMPT at venv activation.

bpo-41341 : Recursive evaluation of typing.ForwardRef in get_type_hints .

bpo-41344 : Prevent creating shared_memory.SharedMemory objects with size=0 .

bpo-41333 : collections.OrderedDict.pop() is now 2 times faster.

bpo-41288 : Unpickling invalid NEWOBJ_EX opcode with the C implementation raises now UnpicklingError instead of crashing.

bpo-39017 : Avoid infinite loop when reading specially crafted TAR files using the tarfile module ( CVE-2019-20907 ).

bpo-41273 : Speed up any transport using _ProactorReadPipeTransport by calling recv_into instead of recv , thus not creating a new buffer for each recv call in the transport’s read loop.

bpo-41235 : Fix the error handling in ssl.SSLContext.load_dh_params() .

bpo-41207 : In distutils.spawn, restore expectation that DistutilsExecError is raised when the command is not found.

bpo-29727 : Register array.array as a MutableSequence . Patch by Pablo Galindo.

bpo-39168 : Remove the __new__ method of typing.Generic .

bpo-41194 : Fix a crash in the _ast module: it can no longer be loaded more than once. It now uses a global state rather than a module state.

bpo-41195 : Add read-only ssl.SSLContext.security_level attribute to retrieve the context’s security level.

bpo-41193 : The write_history() atexit function of the readline completer now ignores any OSError to ignore error if the filesystem is read-only, instead of only ignoring FileNotFoundError and PermissionError .

bpo-41182 : selector: use DefaultSelector based upon implementation

bpo-41161 : The decimal module now requires libmpdec-2.5.0. Users of –with-system-libmpdec should update their system library.

bpo-40874 : The decimal module now requires libmpdec-2.5.0.

bpo-41138 : Fixed the trace module CLI for Python source files with non-UTF-8 encoding.

bpo-31082 : Use the term “iterable” in the docstring for functools.reduce() .

bpo-40521 : Remove freelist from collections.deque().

bpo-31938 : Fix default-value signatures of several functions in the select module - by Anthony Sottile.

bpo-41068 : Fixed reading files with non-ASCII names from ZIP archive directly after writing them.

bpo-41058 : pdb.find_function() now correctly determines the source file encoding.

bpo-41056 : Invalid file descriptor values are now prevented from being passed to os.fpathconf. (discovered by Coverity)

bpo-41056 : Fix a NULL pointer dereference within the ssl module during a MemoryError in the keylog callback. (discovered by Coverity)

bpo-41056 : Fixed an instance where a MemoryError within the zoneinfo module might not be reported or not reported at its source. (found by Coverity)

bpo-41048 : mimetypes.read_mime_types() function reads the rule file using UTF-8 encoding, not the locale encoding. Patch by Srinivas Reddy Thatiparthy.

bpo-41043 : Fixed the use of glob() in the stdlib: literal part of the path is now always correctly escaped.

bpo-41025 : Fixed an issue preventing the C implementation of zoneinfo.ZoneInfo from being subclassed.

bpo-35018 : Add the xml.sax.handler.LexicalHandler class that is present in other SAX XML implementations.

bpo-41002 : Improve performance of HTTPResponse.read with a given amount. Patch by Bruce Merry.

bpo-40448 : ensurepip now disables the use of pip cache when installing the bundled versions of pip and setuptools . Patch by Krzysztof Konopko.

bpo-40967 : Removed asyncio.Task.current_task() and asyncio.Task.all_tasks() . Patch contributed by Rémi Lapeyre.

bpo-40924 : Ensure importlib.resources.path returns an extant path for the SourceFileLoader’s resource reader. Avoids the regression identified in master while a long-term solution is devised.

bpo-40955 : Fix a minor memory leak in subprocess module when extra_groups was specified.

bpo-40855 : The standard deviation and variance functions in the statistics module were ignoring their mu and xbar arguments.

bpo-40939 : Use the new PEG parser when generating the stdlib keyword module.

bpo-23427 : Add sys.orig_argv attribute: the list of the original command line arguments passed to the Python executable.

bpo-33689 : Ignore empty or whitespace-only lines in .pth files. This matches the documentated behavior. Before, empty lines caused the site-packages dir to appear multiple times in sys.path. By Ido Michael, contributors Malcolm Smith and Tal Einat.

bpo-40884 : Added a defaults parameter to logging.Formatter , to allow specifying default values for custom fields. Patch by Asaf Alon and Bar Harel.

bpo-40876 : Clarify error message in the csv module.

bpo-39791 : Refresh importlib.metadata from importlib_metadata 1.6.1.

bpo-40807 : Stop codeop._maybe_compile, used by code.InteractiveInterpreter (and IDLE). from emitting each warning three times.

bpo-32604 : Fix reference leak in the select module when the module is imported in a subinterpreter.

bpo-39791 : Built-in loaders (SourceFileLoader and ZipImporter) now supply TraversableResources implementations for ResourceReader , and the fallback function has been removed.

bpo-39314 : rlcompleter.Completer and the standard Python shell now close the parenthesis for functions that take no arguments. Patch contributed by Rémi Lapeyre.

bpo-17005 : The topological sort functionality that was introduced initially in the functools module has been moved to a new graphlib module to better accommodate the new tools and keep the original scope of the functools module. Patch by Pablo Galindo

bpo-40834 : Fix truncate when sending str object with_xxsubinterpreters.channel_send.

bpo-40755 : Add rich comparisons to collections.Counter().

bpo-26407 : Unexpected errors in calling the __iter__ method are no longer masked by TypeError in csv.reader() , csv.writer.writerow() and csv.writer.writerows() .

bpo-39384 : Fixed email.contentmanager to allow set_content() to set a null string.

bpo-40744 : The sqlite3 module uses SQLite API functions that require SQLite v3.7.3 or higher. This patch removes support for older SQLite versions, and explicitly requires SQLite 3.7.3 both at build, compile and runtime. Patch by Sergey Fedoseev and Erlend E. Aasland.

bpo-40777 : Initialize PyDateTime_IsoCalendarDateType.tp_base at run-time to avoid errors on some compilers.

bpo-38488 : Update ensurepip to install pip 20.1.1 and setuptools 47.1.0.

bpo-40792 : The result of operator.index() now always has exact type int . Previously, the result could have been an instance of a subclass of int .

bpo-40767 : webbrowser now properly finds the default browser in pure Wayland systems by checking the WAYLAND_DISPLAY environment variable. Patch contributed by Jérémy Attali.

bpo-40791 : hashlib.compare_digest() uses OpenSSL’s CRYPTO_memcmp() function when OpenSSL is available.

bpo-40795 : ctypes module: If ctypes fails to convert the result of a callback or if a ctypes callback function raises an exception, sys.unraisablehook is now called with an exception set. Previously, the error was logged into stderr by PyErr_Print() .

bpo-16995 : Add base64.b32hexencode() and base64.b32hexdecode() to support the Base32 Encoding with Extended Hex Alphabet.

bpo-30008 : Fix ssl code to be compatible with OpenSSL 1.1.x builds that use no-deprecated and --api=1.1.0 .

bpo-30064 : Fix asyncio loop.sock_* race condition issue

bpo-40759 : Deprecate the symbol module.

bpo-40756 : The second argument (extra) of LoggerAdapter.__init__ now defaults to None.

bpo-37129 : Add a new os.RWF_APPEND flag for os.pwritev() .

bpo-40737 : Fix possible reference leak for sqlite3 initialization.

bpo-40726 : Handle cases where the end_lineno is None on ast.increment_lineno() .

bpo-40698 : distutils upload creates SHA2-256 and Blake2b-256 digests. MD5 digests is skipped if platform blocks MD5.

bpo-40695 : hashlib no longer falls back to builtin hash implementations when OpenSSL provides a hash digest and the algorithm is blocked by security policy.

bpo-9216 : hashlib.new() passed usedforsecurity to OpenSSL EVP constructor _hashlib.new() . test_hashlib and test_smtplib handle strict security policy better.

bpo-40614 : ast.parse() will not parse self documenting expressions in f-strings when passed feature_version is less than (3, 8) .

bpo-40626 : Add h5 file extension as MIME Type application/x-hdf5, as per HDF Group recommendation for HDF5 formatted data files. Patch contributed by Mark Schwab.

bpo-25920 : On macOS, when building Python for macOS 10.4 and older, which wasn’t the case for python.org macOS installer, socket.getaddrinfo() no longer uses an internal lock to prevent race conditions when calling getaddrinfo() which is thread-safe since macOS 10.5. Python 3.9 requires macOS 10.6 or newer. The internal lock caused random hang on fork when another thread was calling socket.getaddrinfo() . The lock was also used on FreeBSD older than 5.3, OpenBSD older than 201311 and NetBSD older than 4.

bpo-40671 : Prepare _hashlib for PEP 489 and use PyModule_AddType() .

bpo-32309 : Added a new coroutine asyncio.to_thread() . It is mainly used for running IO-bound functions in a separate thread to avoid blocking the event loop, and essentially works as a high-level version of run_in_executor() that can directly take keyword arguments.

bpo-36543 : Restored the deprecated xml.etree.cElementTree module.

bpo-40611 : MAP_POPULATE constant has now been added to the list of exported mmap module flags.

bpo-39881 : PEP 554 for use in the test suite. (Patch By Joannah Nanjekye)

bpo-13097 : ctypes now raises an ArgumentError when a callback is invoked with more than 1024 arguments.

bpo-39385 : A new test assertion context-manager, unittest.assertNoLogs() will ensure a given block of code emits no log messages using the logging module. Contributed by Kit Yan Choi.

bpo-23082 : Updated the error message and docs of PurePath.relative_to() to better reflect the function behaviour.

bpo-40318 : Use SQLite3 trace v2 API, if it is available.

bpo-40105 : ZipFile truncates files to avoid corruption when a shorter comment is provided in append (“a”) mode. Patch by Jan Mazur.

bpo-40084 : Fix Enum.__dir__ : dir(Enum.member) now includes attributes as well as methods.

bpo-31122 : ssl.wrap_socket() now raises ssl.SSLEOFError rather than OSError when peer closes connection during TLS negotiation

bpo-39728 : fix default _missing_ so a duplicate ValueError is not set as the __context__ of the original ValueError .

bpo-39244 : Fixed multiprocessing.context.get_all_start_methods to properly return the default method first on macOS.

bpo-39040 : Fix parsing of invalid mime headers parameters by collapsing whitespace between encoded words in a bare-quote-string.

bpo-38731 : Add --quiet option to command-line interface of py_compile . Patch by Gregory Schevchenko.

bpo-35714 : struct.error is now raised if there is a null character in a struct format string.

bpo-38144 : Added the root_dir and dir_fd parameters in glob.glob() .

bpo-26543 : Fix IMAP4.noop() when debug mode is enabled (ex: imaplib.Debug = 3 ).

bpo-12178 : csv.writer() now correctly escapes escapechar when input contains escapechar . Patch by Catalin Iacob, Berker Peksag, and Itay Elbirt.

bpo-36290 : AST nodes are now raising TypeError on conflicting keyword arguments. Patch contributed by Rémi Lapeyre.

bpo-33944 : Added site.py site-packages tracing in verbose mode.

bpo-35078 : Refactor formatweekday, formatmonthname methods in LocaleHTMLCalendar and LocaleTextCalendar classes in calendar module to call the base class methods.This enables customizable CSS classes for LocaleHTMLCalendar. Patch by Srinivas Reddy Thatiparthy

bpo-29620 : assertWarns() no longer raises a RuntimeException when accessing a module’s __warningregistry__ causes importation of a new module, or when a new module is imported in another thread. Patch by Kernc.

bpo-31844 : Remove ParserBase.error() method from the private and undocumented _markupbase module. html.parser.HTMLParser is the only subclass of ParserBase and its error() implementation was deprecated in Python 3.4 and removed in Python 3.5.

bpo-34226 : Fix cgi.parse_multipart without content_length. Patch by Roger Duran

bpo-33660 : Fix pathlib.PosixPath to resolve a relative path located on the root directory properly.

bpo-28557 : Improve the error message for a misbehaving rawio.readinto

bpo-26680 : The d.is_integer() method is added to the Decimal type, for compatibility with other number types.

bpo-26680 : The x.is_integer() method is incorporated into the abstract types of the numeric tower, Real, Rational and Integral, with appropriate default implementations.

bpo-41428 : Add documentation for PEP 604 (Allow writing union types as X | Y ).

bpo-41774 : In Programming FAQ “Sequences (Tuples/Lists)” section, add “How do you remove multiple items from a list”.

bpo-35293 : Fix RemovedInSphinx40Warning when building the documentation. Patch by Donghee Na.

bpo-37149 : Change Shipman tkinter doc link from archive.org to TkDocs. (The doc has been removed from the NMT server.) The new link responds much faster and includes a short explanatory note.

bpo-41726 : Update the refcounts info of PyType_FromModuleAndSpec .

bpo-41624 : Fix the signature of typing.Coroutine .

bpo-40204 : Enable Sphinx 3.2 c_allow_pre_v3 option and disable c_warn_on_allowed_pre_v3 option to make the documentation compatible with Sphinx 2 and Sphinx 3.

bpo-41045 : Add documentation for debug feature of f-strings.

bpo-41314 : Changed the release when from __future__ import annotations becomes the default from 4.0 to 3.10 (following a change in PEP 563).

bpo-40979 : Refactored typing.rst, arranging more than 70 classes, functions, and decorators into new sub-sections.

bpo-40552 : Fix in tutorial section 4.2. Code snippet is now correct.

bpo-39883 : Make code, examples, and recipes in the Python documentation be licensed under the more permissive BSD0 license in addition to the existing Python 2.0 license.

bpo-37703 : Updated Documentation to comprehensively elaborate on the behaviour of gather.cancel()

bpo-41939 : Fix test_site.test_license_exists_at_url(): call urllib.request.urlcleanup() to reset the global urllib.request._opener . Patch by Victor Stinner.

bpo-41731 : Make test_cmd_line_script pass with option ‘-vv’.

bpo-41602 : Add tests for SIGINT handling in the runpy module.

bpo-41521 : test.support : Rename blacklist parameter of check__all__() to not_exported .

bpo-41477 : Make ctypes optional in test_genericalias.

bpo-41085 : Fix integer overflow in the array.array.index() method on 64-bit Windows for index larger than 2**31 .

bpo-41069 : test.support.TESTFN and the current directory for tests when run via test.regrtest contain now non-ascii characters if possible.

bpo-38377 : On Linux, skip tests using multiprocessing if the current user cannot create a file in /dev/shm/ directory. Add the skip_if_broken_multiprocessing_synchronize() function to the test.support module.

bpo-41009 : Fix use of support.require_{linux|mac|freebsd}_version() decorators as class decorator.

bpo-41003 : Fix test_copyreg when numpy is installed: test.pickletester now saves/restores warnings filters when importing numpy , to ignore filters installed by numpy .

bpo-40964 : Disable remote imaplib tests, host cyrus.andrew.cmu.edu is blocking incoming connections.

bpo-40927 : Fix test_binhex when run twice: it now uses import_fresh_module() to ensure that it raises DeprecationWarning each time.

bpo-17258 : Skip some multiprocessing tests when MD5 hash digest is blocked.

bpo-31904 : Increase LOOPBACK_TIMEOUT to 10 for VxWorks RTOS.

bpo-38169 : Increase code coverage for SharedMemory and ShareableList

bpo-34401 : Make test_gdb properly run on HP-UX. Patch by Michael Osipov.

bpo-38249 : Update Py_UNREACHABLE to use __builtin_unreachable() if only the compiler is able to use it. Patch by Donghee Na.

bpo-41617 : Fix pycore_bitutils.h header file to support old clang versions: __builtin_bswap16() is not available in LLVM clang 3.0.

bpo-40204 : Pin Sphinx version to 2.3.1 in Doc/Makefile .

bpo-36020 : The C99 functions snprintf() and vsnprintf() are now required to build Python.

bpo-40684 : make install now uses the PLATLIBDIR variable for the destination lib-dynload/ directory when ./configure --with-platlibdir is used.

bpo-40683 : Fixed an issue where the zoneinfo module and its tests were not included when Python is installed with make .

bpo-41744 : Fixes automatic import of props file when using the Nuget package.

bpo-41627 : The user site directory for 32-bit now includes a -32 suffix to distinguish it from the 64-bit interpreter’s directory.

bpo-41526 : Fixed layout of final page of the installer by removing the special thanks to Mark Hammond (with his permission).

bpo-41492 : Fixes the description that appears in UAC prompts.

bpo-40948 : Improve post-install message to direct people to the “py” command.

bpo-41412 : The installer will now fail to install on Windows 7 and Windows 8. Further, the UCRT dependency is now always downloaded on demand.

bpo-40741 : Update Windows release to include SQLite 3.32.3.

bpo-41142 : msilib now supports creating CAB files with non-ASCII file path and adding files with non-ASCII file path to them.

bpo-41074 : Fixed support of non-ASCII names in functions msilib.OpenDatabase() and msilib.init_database() and non-ASCII SQL in method msilib.Database.OpenView() .

bpo-41039 : Stable ABI redirection DLL (python3.dll) now uses #pragma comment(linker) for re-exporting.

bpo-40164 : Updates Windows OpenSSL to 1.1.1g

bpo-39631 : Changes the registered MIME type for .py files on Windows to text/x-python instead of text/plain .

bpo-40677 : Manually define IO_REPARSE_TAG_APPEXECLINK in case some old Windows SDK doesn’t have it.

bpo-37556 : Extend py.exe help to mention overrides via venv, shebang, environmental variables & ini files.

bpo-41557 : Update macOS installer to use SQLite 3.33.0.

bpo-39580 : Avoid opening Finder window if running installer from the command line. Patch contributed by Rick Heil.

bpo-41100 : Fix configure error when building on macOS 11. Note that the current Python release was released shortly after the first developer preview of macOS 11 (Big Sur); there are other known issues with building and running on the developer preview. Big Sur is expected to be fully supported in a future bugfix release of Python 3.8.x and with 3.9.0.

bpo-40741 : Update macOS installer to use SQLite 3.32.3.

bpo-41005 : fixed an XDG settings issue not allowing macos to open browser in webbrowser.py

bpo-40741 : Update macOS installer to use SQLite 3.32.2.

bpo-41775 : Use ‘IDLE Shell’ as shell title

bpo-35764 : Rewrite the Calltips doc section.

bpo-40181 : In calltips, stop reminding that ‘/’ marks the end of positional-only arguments.

bpo-41468 : Improve IDLE run crash error message (which users should never see).

bpo-41373 : Save files loaded with no line ending, as when blank, or different line endings, by setting its line ending to the system default. Fix regression in 3.8.4 and 3.9.0b4.

bpo-41300 : Save files with non-ascii chars. Fix regression released in 3.9.0b4 and 3.8.4.

bpo-37765 : Add keywords to module name completion list. Rewrite Completions section of IDLE doc.

bpo-41152 : The encoding of stdin , stdout and stderr in IDLE is now always UTF-8.

bpo-41144 : Make Open Module open a special module such as os.path.

bpo-39885 : Make context menu Cut and Copy work again when right-clicking within a selection.

bpo-40723 : Make test_idle pass when run after import.

bpo-41936 : Removed undocumented macros Py_ALLOW_RECURSION and Py_END_ALLOW_RECURSION and the recursion_critical field of the PyInterpreterState structure.

bpo-41692 : The PyUnicode_InternImmortal() function is now deprecated and will be removed in Python 3.12: use PyUnicode_InternInPlace() instead. Patch by Victor Stinner.

bpo-41842 : Add PyCodec_Unregister() function to unregister a codec search function.

bpo-41834 : Remove the _Py_CheckRecursionLimit variable: it has been replaced by ceval.recursion_limit of the PyInterpreterState structure. Patch by Victor Stinner.

bpo-41689 : Types created with PyType_FromSpec() now make any signature in their tp_doc slot accessible from __text_signature__ .

bpo-41524 : Fix bug in PyOS_mystrnicmp and PyOS_mystricmp that incremented pointers beyond the end of a string.

bpo-41324 : Add a minimal decimal capsule API. The API supports fast conversions between Decimals up to 38 digits and their triple representation as a C struct.

bpo-30155 : Add PyDateTime_DATE_GET_TZINFO() and PyDateTime_TIME_GET_TZINFO() macros for accessing the tzinfo attributes of datetime.datetime and datetime.time objects.

bpo-40170 : Revert PyType_HasFeature() change: it reads again directly the PyTypeObject.tp_flags member when the limited C API is not used, rather than always calling PyType_GetFlags() which hides implementation details.

bpo-41123 : Remove PyUnicode_AsUnicodeCopy .

bpo-41123 : Removed PyLong_FromUnicode() .

bpo-41123 : Removed PyUnicode_GetMax() .

bpo-41123 : Removed Py_UNICODE_str* functions manipulating Py_UNICODE* strings.

bpo-41103 : PyObject_AsCharBuffer() , PyObject_AsReadBuffer() , PyObject_CheckReadBuffer() , and PyObject_AsWriteBuffer() are removed. Please migrate to new buffer protocol; PyObject_GetBuffer() and PyBuffer_Release() .

bpo-36346 : Raises DeprecationWarning for PyUnicode_FromUnicode(NULL, size) and PyUnicode_FromStringAndSize(NULL, size) with size > 0 .

bpo-36346 : Mark Py_UNICODE_COPY , Py_UNICODE_FILL , PyUnicode_WSTR_LENGTH , PyUnicode_FromUnicode , PyUnicode_AsUnicode , and PyUnicode_AsUnicodeAndSize as deprecated in C. Remove Py_UNICODE_MATCH which was deprecated and broken since Python 3.3.

bpo-40989 : The PyObject_INIT() and PyObject_INIT_VAR() macros become aliases to, respectively, PyObject_Init() and PyObject_InitVar() functions.

bpo-36020 : On Windows, #include "pyerrors.h" no longer defines snprintf and vsnprintf macros.

bpo-40943 : The PY_SSIZE_T_CLEAN macro must now be defined to use PyArg_ParseTuple() and Py_BuildValue() formats which use # : es# , et# , s# , u# , y# , z# , U# and Z# . See Parsing arguments and building values and the PEP 353 .

bpo-40910 : Export explicitly the Py_GetArgcArgv() function to the C API and document the function. Previously, it was exported implicitly which no longer works since Python is built with -fvisibility=hidden .

bpo-40724 : Allow defining buffer slots in type specs.

bpo-40679 : Fix a _PyEval_EvalCode() crash if qualname argument is NULL.

bpo-40839 : Calling PyDict_GetItem() without GIL held had been allowed for historical reason. It is no longer allowed.

bpo-40826 : PyOS_InterruptOccurred() now fails with a fatal error if it is called with the GIL released.

bpo-40792 : The result of PyNumber_Index() now always has exact type int . Previously, the result could have been an instance of a subclass of int .

bpo-39573 : Convert Py_REFCNT() and Py_SIZE() macros to static inline functions. They cannot be used as l-value anymore: use Py_SET_REFCNT() and Py_SET_SIZE() to set an object reference count and size. This change is backward incompatible on purpose, to prepare the C API for an opaque PyObject structure.

bpo-40703 : The PyType_FromSpec*() functions no longer overwrite the type’s “__module__” attribute if it is set via “Py_tp_members” or “Py_tp_getset”.

bpo-39583 : Remove superfluous “extern C” declarations from Include/cpython/*.h .

Python 3.9.0 beta 1 ¶

Release date: 2020-05-19

bpo-40501 : uuid no longer uses ctypes to load libuuid or rpcrt4.dll at runtime.

bpo-40663 : Correctly generate annotations where parentheses are omitted but required (e.g: Type[(str, int, *other))] .

bpo-40596 : Fixed str.isidentifier() for non-canonicalized strings containing non-BMP characters on Windows.

bpo-40593 : Improved syntax errors for invalid characters in source code.

bpo-40585 : Fixed a bug when using codeop.compile_command() that was causing exceptions to be swallowed with the new parser. Patch by Pablo Galindo

bpo-40566 : Apply PEP 573 to abc .

bpo-40502 : Initialize n->n_col_offset . (Patch by Joannah Nanjekye)

bpo-40527 : Fix command line argument parsing: no longer write errors multiple times into stderr.

bpo-1635741 : Port errno to multiphase initialization ( PEP 489 ).

bpo-40523 : Add pass-throughs for hash() and reversed() to weakref.proxy objects. Patch by Pablo Galindo.

bpo-1635741 : Port syslog to multiphase initialization ( PEP 489 ).

bpo-40246 : Reporting a specialised error message for invalid string prefixes, which was introduced in bpo-40246 , is being reverted due to backwards compatibility concerns for strings that immediately follow a reserved keyword without whitespace between them. Constructs like bg="#d00" if clear else"#fca" were failing to parse, which is not an acceptable breakage on such short notice.

bpo-40417 : Fix imp module deprecation warning when PyImport_ReloadModule is called. Patch by Robert Rouhani.

bpo-40408 : Fixed support of nested type variables in GenericAlias (e.g. list[list[T]] ).

bpo-1635741 : Port _stat module to multiphase initialization ( PEP 489 ).

bpo-29587 : Enable implicit exception chaining when calling generator.throw() .

bpo-40328 : Add tools for generating mappings headers for CJKCodecs.

bpo-40228 : Setting frame.f_lineno is now robust w.r.t. changes in the source-to-bytecode compiler

bpo-38880 : Added the ability to list interpreters associated with channel ends in the internal subinterpreters module.

bpo-37986 : Improve performance of PyLong_FromDouble() for values that fit into long .

bpo-40662 : Fixed ast.get_source_segment() for ast nodes that have incomplete location information. Patch by Irit Katriel.

bpo-40665 : Convert bisect to use Argument Clinic.

bpo-40536 : Added the available_timezones() function to the zoneinfo module. Patch by Paul Ganssle.

bpo-40645 : The hmac.HMAC exposes internal implementation details. The attributes digest_cons , inner , and outer are deprecated and will be removed in the future.

bpo-40645 : The internal module _hashlib wraps and exposes OpenSSL’s HMAC API. The new code will be used in Python 3.10 after the internal implementation details of the pure Python HMAC module are no longer part of the public API.

bpo-40637 : Builtin hash modules can now be disabled or selectively enabled with configure --with-builtin-hashlib-hashes=sha3,blake1 or --without-builtin-hashlib-hashes .

bpo-37630 : The hashlib module can now use SHA3 hashes and SHAKE XOF from OpenSSL when available.

bpo-40479 : The hashlib now compiles with OpenSSL 3.0.0-alpha2.

bpo-40257 : Revert changes to inspect.getdoc() .

bpo-40607 : When cancelling a task due to timeout, asyncio.wait_for() will now propagate the exception if an error happens during cancellation. Patch by Roman Skurikhin.

bpo-40612 : Fix edge cases in SyntaxError formatting. If the offset is <= 0, no caret is printed. If the offset is > line length, the caret is printed pointing just after the last character.

bpo-40597 : If text content lines are longer than policy.max_line_length, always use a content-encoding to make sure they are wrapped.

bpo-40571 : Added functools.cache() as a simpler, more discoverable way to access the unbounded cache variant of lru_cache(maxsize=None).

bpo-40503 : PEP 615 , the zoneinfo module. Adds support for the IANA time zone database.

bpo-40397 : Removed attributes __args__ and __parameters__ from special generic aliases like typing.List (not subscripted).

bpo-40549 : Convert posixmodule.c (“posix” or “nt” module) to the multiphase initialization (PEP 489).

bpo-31033 : Add a msg argument to Future.cancel() and Task.cancel() .

bpo-40541 : Added an optional counts parameter to random.sample().

bpo-40515 : The ssl and hashlib modules now actively check that OpenSSL is build with thread support. Python 3.7.0 made thread support mandatory and no longer works safely with a no-thread builds.

bpo-31033 : When a asyncio.Task is cancelled, the exception traceback now chains all the way back to where the task was first interrupted.

bpo-40504 : functools.lru_cache() objects can now be the targets of weakrefs.

bpo-40559 : Fix possible memory leak in the C implementation of asyncio.Task .

bpo-40480 : fnmatch.fnmatch() could take exponential time in the presence of multiple * pattern characters. This was repaired by generating more elaborate regular expressions to avoid futile backtracking.

bpo-40495 : compileall is now able to use hardlinks to prevent duplicates in a case when .pyc files for different optimization levels have the same content.

bpo-40457 : The ssl module now support OpenSSL builds without TLS 1.0 and 1.1 methods.

bpo-40355 : Improve error reporting in ast.literal_eval() in the presence of malformed ast.Dict nodes instead of silently ignoring any non-conforming elements. Patch by Curtis Bucher.

bpo-40465 : Deprecated the optional random argument to random.shuffle() .

bpo-40459 : platform.win32_ver() now produces correct ptype strings instead of empty strings.

bpo-39435 : The first argument of pickle.loads() is now positional-only.

bpo-39305 : Update nntplib to merge nntplib.NNTP and nntplib._NNTPBase . Patch by Donghee Na.

bpo-32494 : Update dbm.gnu to use gdbm_count if possible when calling len() . Patch by Donghee Na.

bpo-40453 : Add isolated=True keyword-only parameter to _xxsubinterpreters.create() . An isolated subinterpreter cannot spawn threads, spawn a child process or call os.fork() .

bpo-40286 : Remove _random.Random.randbytes() : the C implementation of randbytes() . Implement the method in Python to ease subclassing: randbytes() now directly reuses getrandbits() .

bpo-40394 : Added default arguments to difflib.SequenceMatcher.find_longest_match() .

bpo-39995 : Fix a race condition in concurrent.futures._ThreadWakeup: access to _ThreadWakeup is now protected with the shutdown lock.

bpo-30966 : Process.shutdown(wait=True) of concurrent.futures now closes explicitly the result queue.

bpo-30966 : Add a new close() method to the SimpleQueue class to explicitly close the queue.

bpo-39966 : Revert bpo-25597 . unittest.mock.MagicMock with wraps’ set uses default return values for magic methods.

bpo-39791 : Added files() function to importlib.resources with support for subdirectories in package data, matching backport in importlib_resources 1.5.

bpo-40375 : imaplib.IMAP4.unselect() is added. Patch by Donghee Na.

bpo-40389 : repr() now returns typing.Optional[T] when called for typing.Union of two types, one of which is NoneType .

bpo-40291 : Add support for CAN_J1939 sockets (available on Linux 5.4+)

bpo-40273 : types.MappingProxyType is now reversible.

bpo-39075 : The repr for types.SimpleNamespace is now insertion ordered rather than alphabetical.

bpo-40192 : On AIX, thread_time() is now implemented with thread_cputime() which has nanosecond resolution, rather than clock_gettime(CLOCK_THREAD_CPUTIME_ID) which has a resolution of 10 milliseconds. Patch by Batuhan Taskaya.

bpo-40025 : Raise TypeError when _generate_next_value_ is defined after members. Patch by Ethan Onstott.

bpo-39058 : In the argparse module, the repr for Namespace() and other argument holders now displayed in the order attributes were added. Formerly, it displayed in alphabetical order even though argument order is preserved the user visible parts of the module.

bpo-24416 : The isocalendar() methods of datetime.date and datetime.datetime now return a named tuple instead of a tuple .

bpo-34790 : Add version of removal for explicit passing of coros to asyncio.wait() ’s documentation

bpo-40561 : Provide docstrings for webbrowser open functions.

bpo-40499 : Mention that asyncio.wait() requires a non-empty set of awaitables.

bpo-39705 : Tutorial example for sorted() in the Loop Techniques section is given a better explanation. Also a new example is included to explain sorted()’s basic behavior.

bpo-39435 : Fix an incorrect signature for pickle.loads() in the docs

bpo-40055 : distutils.tests now saves/restores warnings filters to leave them unchanged. Importing tests imports docutils which imports pkg_resources which adds a warnings filter.

bpo-40436 : test_gdb and test.pythoninfo now check gdb command exit code.

bpo-40653 : Move _dirnameW out of HAVE_SYMLINK to fix a potential compiling issue.

bpo-40514 : Add --with-experimental-isolated-subinterpreters build option to configure : better isolate subinterpreters, experimental build mode.

bpo-40650 : Include winsock2.h in pytime.c for timeval.

bpo-40458 : Increase reserved stack space to prevent overflow crash on Windows.

bpo-39148 : Add IPv6 support to asyncio datagram endpoints in ProactorEventLoop. Change the raised exception for unknown address families to ValueError as it’s not coming from Windows API.

bpo-34956 : When building Python on macOS from source, _tkinter now links with non-system Tcl and Tk frameworks if they are installed in /Library/Frameworks , as had been the case on older releases of macOS. If a macOS SDK is explicitly configured, by using --enable-universalsdk= or -isysroot , only the SDK itself is searched. The default behavior can still be overridden with --with-tcltk-includes and --with-tcltk-libs .

bpo-35569 : Expose RFC 3542 IPv6 socket options.

bpo-40479 : Update multissltest helper to test with latest OpenSSL 1.0.2, 1.1.0, 1.1.1, and 3.0.0-alpha.

bpo-40431 : Fix a syntax typo in turtledemo that now raises a SyntaxError .

bpo-40163 : Fix multissltest tool. OpenSSL has changed download URL for old releases. The multissltest tool now tries to download from current and old download URLs.

bpo-39465 : Remove the _PyUnicode_ClearStaticStrings() function from the C API.

bpo-38787 : Add PyCFunction_CheckExact() macro for exact type checks now that we allow subtypes of PyCFunction, as well as PyCMethod_CheckExact() and PyCMethod_Check() for the new PyCMethod subtype.

bpo-40545 : Declare _PyErr_GetTopmostException() with PyAPI_FUNC() to properly export the function in the C API. The function remains private ( _Py ) prefix.

bpo-40412 : Nullify inittab_copy during finalization, preventing future interpreter initializations in an embedded situation from crashing. Patch by Gregory Szorc.

bpo-40429 : The PyThreadState_GetFrame() function now returns a strong reference to the frame.

bpo-40428 : Remove the following functions from the C API. Call PyGC_Collect() explicitly to free all free lists.

PyAsyncGen_ClearFreeLists()

PyContext_ClearFreeList()

PyDict_ClearFreeList()

PyFloat_ClearFreeList()

PyFrame_ClearFreeList()

PyList_ClearFreeList()

PySet_ClearFreeList()

PyTuple_ClearFreeList()

bpo-40421 : New PyFrame_GetBack() function: get the frame next outer frame.

bpo-40421 : New PyFrame_GetCode() function: return a borrowed reference to the frame code.

bpo-40217 : Ensure that instances of types created with PyType_FromSpecWithBases() will visit its class object when traversing references in the garbage collector (implemented as an extension of the provided tp_traverse ). Patch by Pablo Galindo.

bpo-38787 : Module C state is now accessible from C-defined heap type methods ( PEP 573 ). Patch by Marcel Plch and Petr Viktorin.

Python 3.9.0 alpha 6 ¶

Release date: 2020-04-27

bpo-40121 : Fixes audit events raised on creating a new socket.

bpo-39073 : Disallow CR or LF in email.headerregistry.Address arguments to guard against header injection attacks.

bpo-39503 : CVE-2020-8492 : The AbstractBasicAuthHandler class of the urllib.request module uses an inefficient regular expression which can be exploited by an attacker to cause a denial of service. Fix the regex to prevent the catastrophic backtracking. Vulnerability reported by Ben Caller and Matt Schwager.

bpo-40313 : Improve the performance of bytes.hex().

bpo-40334 : Switch to a new parser, based on PEG. For more details see PEP 617. To temporarily switch back to the old parser, use -X oldparser or PYTHONOLDPARSER=1 . In Python 3.10 we will remove the old parser completely, including the parser module (already deprecated) and anything that depends on it.

bpo-40267 : Fix the tokenizer to display the correct error message, when there is a SyntaxError on the last input character and no newline follows. It used to be unexpected EOF while parsing , while it should be invalid syntax .

bpo-39522 : Correctly unparse explicit u prefix for strings when postponed evaluation for annotations activated. Patch by Batuhan Taskaya.

bpo-40246 : Report a specialized error message, invalid string prefix , when the tokenizer encounters a string with an invalid prefix.

bpo-40082 : Fix the signal handler: it now always uses the main interpreter, rather than trying to get the current Python thread state.

bpo-37388 : str.encode() and str.decode() no longer check the encoding and errors in development mode or in debug mode during Python finalization. The codecs machinery can no longer work on very late calls to str.encode() and str.decode().

bpo-40077 : Fix possible refleaks in _json , memo of PyScannerObject should be traversed.

bpo-37207 : Speed up calls to dict() by using the PEP 590 vectorcall calling convention.

bpo-40141 : Add column and line information to ast.keyword nodes. Patch by Pablo Galindo.

bpo-1635741 : Port resource to multiphase initialization ( PEP 489 ).

bpo-1635741 : Port math to multiphase initialization ( PEP 489 ).

bpo-1635741 : Port _uuid module to multiphase initialization ( PEP 489 ).

bpo-40077 : Convert json module to use PyType_FromSpec() .

bpo-40067 : Improve the error message for multiple star expressions in an assignment. Patch by Furkan Onder

bpo-1635741 : Port _functools module to multiphase initialization (PEP 489). Patch by Paulo Henrique Silva.

bpo-1635741 : Port operator module to multiphase initialization (PEP 489). Patch by Paulo Henrique Silva.

bpo-20526 : Fix PyThreadState_Clear() . PyThreadState.frame is a borrowed reference, not a strong reference: PyThreadState_Clear() must not call Py_CLEAR(tstate->frame) .

bpo-1635741 : Port time module to multiphase initialization ( PEP 489 ). Patch by Paulo Henrique Silva.

bpo-1635741 : Port _weakref extension module to multiphase initialization ( PEP 489 ).

bpo-40020 : Fix a leak and subsequent crash in parsetok.c caused by realloc misuse on a rare codepath.

bpo-39939 : Added str.removeprefix and str.removesuffix methods and corresponding bytes, bytearray, and collections.UserString methods to remove affixes from a string if present. See PEP 616 for a full description. Patch by Dennis Sweeney.

bpo-39481 : Implement PEP 585. This supports list[int], tuple[str, …] etc.

bpo-32894 : Support unparsing of infinity numbers in postponed annotations. Patch by Batuhan Taşkaya.

bpo-37207 : Speed up calls to list() by using the PEP 590 vectorcall calling convention. Patch by Mark Shannon.

bpo-40398 : typing.get_args() now always returns an empty tuple for special generic aliases.

bpo-40396 : Functions typing.get_origin() , typing.get_args() and typing.get_type_hints() support now generic aliases like list[int] .

bpo-38061 : Optimize the subprocess module on FreeBSD using closefrom() . A single close(fd) syscall is cheap, but when sysconf(_SC_OPEN_MAX) is high, the loop calling close(fd) on each file descriptor can take several milliseconds.

The workaround on FreeBSD to improve performance was to load and mount the fdescfs kernel module, but this is not enabled by default.

Initial patch by Ed Maste (emaste), Conrad Meyer (cem), Kyle Evans (kevans) and Kubilay Kocak (koobs): https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=242274

bpo-38061 : On FreeBSD, os.closerange(fd_low, fd_high) now calls closefrom(fd_low) if fd_high is greater than or equal to sysconf(_SC_OPEN_MAX) .

bpo-40360 : The lib2to3 module is pending deprecation due to PEP 617 .

bpo-40138 : Fix the Windows implementation of os.waitpid() for exit code larger than INT_MAX >> 8 . The exit status is now interpreted as an unsigned number.

bpo-39942 : Set “__main__” as the default module name when “__name__” is missing in typing.TypeVar . Patch by Weipeng Hong.

bpo-40275 : The logging package is now imported lazily in unittest only when the assertLogs() assertion is used.

bpo-40275 : The asyncio package is now imported lazily in unittest only when the IsolatedAsyncioTestCase class is used.

bpo-40330 : In ShareableList.__setitem__() , check the size of a new string item after encoding it to utf-8, not before.

bpo-40148 : Added pathlib.Path.with_stem() to create a new Path with the stem replaced.

bpo-40325 : Deprecated support for set objects in random.sample().

bpo-40257 : Improved help for the typing module. Docstrings are now shown for all special forms and special generic aliases (like Union and List ). Using help() with generic alias like List[int] will show the help for the correspondent concrete type ( list in this case).

bpo-40257 : inspect.getdoc() no longer returns docstring inherited from the type of the object or from parent class if it is a class if it is not defined in the object itself. In pydoc the documentation string is now shown not only for class, function, method etc, but for any object that has its own __doc__ attribute.

bpo-40287 : Fixed SpooledTemporaryFile.seek() to return the position.

bpo-40290 : Added zscore() to statistics.NormalDist().

bpo-40282 : Allow random.getrandbits(0) to succeed and to return 0.

bpo-40286 : Add random.randbytes() function and random.Random.randbytes() method to generate random bytes.

bpo-40277 : collections.namedtuple() now provides a human-readable repr for its field accessors.

bpo-40270 : The included copy of sqlite3 on Windows is now compiled with the json extension. This allows the use of functions such as json_object .

bpo-29255 : Wait in KqueueSelector.select when no fds are registered

bpo-40260 : Ensure modulefinder uses io.open_code() and respects coding comments.

bpo-40234 : Allow again to spawn daemon threads in subinterpreters (revert change which denied them).

bpo-39207 : Workers in ProcessPoolExecutor are now spawned on demand, only when there are no available idle workers to reuse. This optimizes startup overhead and reduces the amount of lost CPU time to idle workers. Patch by Kyle Stanley.

bpo-40091 : Fix a hang at fork in the logging module: the new private _at_fork_reinit() method is now used to reinitialize locks at fork in the child process.

bpo-40149 : Implement traverse and clear slots in _abc._abc_data type.

bpo-40208 : Remove deprecated symtable.SymbolTable.has_exec() .

bpo-40196 : Fix a bug in the symtable module that was causing incorrectly report global variables as local. Patch by Pablo Galindo.

bpo-40190 : Add support for _SC_AIX_REALMEM to posix.sysconf() .

bpo-40182 : Removed the _field_types attribute of the typing.NamedTuple class.

bpo-36517 : Multiple inheritance with typing.NamedTuple now raises an error instead of silently ignoring other types.

bpo-40126 : Fixed reverting multiple patches in unittest.mock. Patcher’s __exit__() is now never called if its __enter__() is failed. Returning true from __exit__() silences now the exception.

bpo-40094 : CGIHTTPRequestHandler of http.server now logs the CGI script exit code, rather than the CGI script exit status of os.waitpid(). For example, if the script is killed by signal 11, it now logs: “CGI script exit code -11.”

bpo-40108 : Improve the error message when triying to import a module using runpy and incorrectly using the “.py” extension at the end of the module name. Patch by Pablo Galindo.

bpo-40094 : Add os.waitstatus_to_exitcode() function: convert a wait status to an exit code.

bpo-40089 : Fix threading._after_fork(): if fork was not called by a thread spawned by threading.Thread, threading._after_fork() now creates a _MainThread instance for _main_thread, instead of a _DummyThread instance.

bpo-40089 : Add a private _at_fork_reinit() method to _thread.Lock , _thread.RLock , threading.RLock and threading.Condition classes: reinitialize the lock at fork in the child process, reset the lock to the unlocked state. Rename also the private _reset_internal_locks() method of threading.Event to _at_fork_reinit() .

bpo-25780 : Expose CAN_RAW_JOIN_FILTERS in the socket module.

bpo-39503 : AbstractBasicAuthHandler of urllib.request now parses all WWW-Authenticate HTTP headers and accepts multiple challenges per header: use the realm of the first Basic challenge.

bpo-39812 : Removed daemon threads from concurrent.futures by adding an internal threading._register_atexit() , which calls registered functions prior to joining all non-daemon threads. This allows for compatibility with subinterpreters, which don’t support daemon threads.

bpo-40050 : Fix importlib._bootstrap_external : avoid creating a new winreg builtin module if it’s already available in sys.modules , and remove redundant imports.

bpo-40014 : Fix os.getgrouplist() : if getgrouplist() function fails because the group list is too small, retry with a larger group list. On failure, the glibc implementation of getgrouplist() sets ngroups to the total number of groups. For other implementations, double the group list size.

bpo-40017 : Add time.CLOCK_TAI constant if the operating system support it.

bpo-40016 : In re docstring, clarify the relationship between inline and argument compile flags.

bpo-39953 : Update internal table of OpenSSL error codes in the ssl module.

bpo-36144 : Added PEP 584 operators to weakref.WeakValueDictionary .

bpo-36144 : Added PEP 584 operators to weakref.WeakKeyDictionary .

bpo-38891 : Fix linear runtime behaviour of the __getitem__ and __setitem__ methods in multiprocessing.shared_memory.ShareableList . This avoids quadratic performance when iterating a ShareableList . Patch by Thomas Krennwallner.

bpo-39682 : Remove undocumented support for closing a pathlib.Path object via its context manager. The context manager magic methods remain, but they are now a no-op, making Path objects immutable.

bpo-36144 : Added PEP 584 operators ( | and |= ) to collections.ChainMap .

bpo-39011 : Normalization of line endings in ElementTree attributes was removed, as line endings which were replaced by entity numbers should be preserved in original form.

bpo-38410 : Properly handle sys.audit() failures in sys.set_asyncgen_hooks() .

bpo-36541 : lib2to3 now recognizes named assignment expressions (the walrus operator, := )

bpo-35967 : In platform, delay the invocation of ‘uname -p’ until the processor attribute is requested.

bpo-35113 : inspect.getsource() now returns correct source code for inner class with same name as module level class. Decorators are also returned as part of source of the class. Patch by Karthikeyan Singaravelan.

bpo-33262 : Deprecate passing None as an argument for shlex.split() ’s s parameter. Patch by Zackery Spytz.

bpo-31758 : Prevent crashes when using an uninitialized _elementtree.XMLParser object. Patch by Oren Milman.

bpo-27635 : The pickle documentation incorrectly claimed that __new__ isn’t called by default when unpickling.

bpo-39879 : Updated Data model docs to include dict() insertion order preservation. Patch by Furkan Onder and Samy Lahfa.

bpo-38387 : Document PyDoc_STRVAR macro in the C-API reference.

bpo-13743 : Some methods within xml.dom.minidom.Element class are now better documented.

bpo-31904 : Set expected default encoding in test_c_locale_coercion.py for VxWorks RTOS.

bpo-40162 : Update Travis CI configuration to OpenSSL 1.1.1f.

bpo-40146 : Update OpenSSL to 1.1.1f in Azure Pipelines.

bpo-40094 : Add test.support.wait_process() function.

bpo-40003 : test.bisect_cmd now copies Python command line options like -O or -W . Moreover, emit a warning if test.bisect_cmd is used with -w / --verbose2 option.

bpo-39380 : Add the encoding in ftplib.FTP and ftplib.FTP_TLS to the constructor as keyword-only and change the default from latin-1 to utf-8 to follow RFC 2640 .

bpo-39793 : Use the same domain when testing make_msgid . Patch by Batuhan Taskaya.

bpo-1812 : Fix newline handling in doctest.testfile when loading from a package whose loader has a get_data method. Patch by Peter Donis.

bpo-38360 : Support single-argument form of macOS -isysroot flag.

bpo-40158 : Fix CPython MSBuild Properties in NuGet Package (build/native/python.props)

bpo-38527 : Fix configure check on Solaris for “float word ordering”: sometimes, the correct “grep” command was not being used. Patch by Arnon Yaari.

bpo-40164 : Updates Windows to OpenSSL 1.1.1f

bpo-8901 : Ignore the Windows registry when the -E option is used.

bpo-38329 : python.org macOS installers now update the Current version symlink of /Library/Frameworks/Python.framework/Versions for 3.9 installs. Previously, Current was only updated for Python 2.x installs. This should make it easier to embed Python 3 into other macOS applications.

bpo-40164 : Update macOS installer builds to use OpenSSL 1.1.1g.

bpo-38439 : Add a 256×256 pixel IDLE icon to support more modern environments. Created by Andrew Clover. Delete the unused macOS idle.icns icon file.

bpo-38689 : IDLE will no longer freeze when inspect.signature fails when fetching a calltip.

bpo-40385 : Removed the checkpyc.py tool. Please see compileall without force mode as a potential alternative.

bpo-40179 : Fixed translation of #elif in Argument Clinic.

bpo-40094 : Fix which.py script exit code: it now uses os.waitstatus_to_exitcode() to convert os.system() exit status into an exit code.

bpo-40241 : Move the PyGC_Head structure to the internal C API.

bpo-40170 : Convert PyObject_IS_GC() macro to a function to hide implementation details.

bpo-40241 : Add the functions PyObject_GC_IsTracked() and PyObject_GC_IsFinalized() to the public API to allow to query if Python objects are being currently tracked or have been already finalized by the garbage collector respectively. Patch by Pablo Galindo.

bpo-40170 : The PyObject_NEW() macro becomes an alias to the PyObject_New() macro, and the PyObject_NEW_VAR() macro becomes an alias to the PyObject_NewVar() macro, to hide implementation details. They no longer access directly the PyTypeObject.tp_basicsize member.

bpo-40170 : PyType_HasFeature() now always calls PyType_GetFlags() to hide implementation details. Previously, it accessed directly the PyTypeObject.tp_flags member when the limited C API was not used.

bpo-40170 : Convert the PyObject_GET_WEAKREFS_LISTPTR() macro to a function to hide implementation details: the macro accessed directly to the PyTypeObject.tp_weaklistoffset member.

bpo-40170 : Convert PyObject_CheckBuffer() macro to a function to hide implementation details: the macro accessed directly the PyTypeObject.tp_as_buffer member.

bpo-40170 : Always declare PyIndex_Check() as an opaque function to hide implementation details: remove PyIndex_Check() macro. The macro accessed directly the PyTypeObject.tp_as_number member.

bpo-39947 : Add PyThreadState_GetID() function: get the unique identifier of a Python thread state.

Python 3.9.0 alpha 5 ¶

Release date: 2020-03-23

bpo-38576 : Disallow control characters in hostnames in http.client, addressing CVE-2019-18348 . Such potentially malicious header injection URLs now cause a InvalidURL to be raised.

bpo-40010 : Optimize pending calls in multithreaded applications. If a thread different than the main thread schedules a pending call ( Py_AddPendingCall() ), the bytecode evaluation loop is no longer interrupted at each bytecode instruction to check for pending calls which cannot be executed. Only the main thread can execute pending calls.

Previously, the bytecode evaluation loop was interrupted at each instruction until the main thread executes pending calls.

bpo-1635741 : Port _collections module to multiphase initialization ( PEP 489 ).

bpo-40010 : Optimize signal handling in multithreaded applications. If a thread different than the main thread gets a signal, the bytecode evaluation loop is no longer interrupted at each bytecode instruction to check for pending signals which cannot be handled. Only the main thread of the main interpreter can handle signals.

Previously, the bytecode evaluation loop was interrupted at each instruction until the main thread handles signals.

bpo-39984 : If Py_AddPendingCall() is called in a subinterpreter, the function is now scheduled to be called from the subinterpreter, rather than being called from the main interpreter. Each subinterpreter now has its own list of scheduled calls.

bpo-1635741 : Port _heapq module to multiphase initialization.

bpo-1635741 : Port itertools module to multiphase initialization ( PEP 489 ).

bpo-37207 : Speed up calls to frozenset() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-39984 : subinterpreters: Move _PyRuntimeState.ceval.tracing_possible to PyInterpreterState.ceval.tracing_possible : each interpreter now has its own variable.

bpo-37207 : Speed up calls to set() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-1635741 : Port _statistics module to multiphase initialization ( PEP 489 ).

bpo-39968 : Use inline function to replace extension modules’ get_module_state macros.

bpo-39965 : Correctly raise SyntaxError if await is used inside non-async functions and PyCF_ALLOW_TOP_LEVEL_AWAIT is set (like in the asyncio REPL). Patch by Pablo Galindo.

bpo-39562 : Allow executing asynchronous comprehensions on the top level when the PyCF_ALLOW_TOP_LEVEL_AWAIT flag is given. Patch by Batuhan Taskaya.

bpo-37207 : Speed up calls to tuple() by using the PEP 590 vectorcall calling convention. Patch by Donghee Na.

bpo-38373 : Changed list overallocation strategy. It no longer overallocates if the new size is closer to overallocated size than to the old size and adds padding.

bpo-39926 : Update Unicode database to Unicode version 13.0.0.

bpo-19466 : Clear the frames of daemon threads earlier during the Python shutdown to call objects destructors. So “unclosed file” resource warnings are now emitted for daemon threads in a more reliable way.

bpo-38894 : Fix a bug that was causing incomplete results when calling pathlib.Path.glob in the presence of symlinks that point to files where the user does not have read access. Patch by Pablo Galindo and Matt Wozniski.

bpo-39877 : Fix PyEval_RestoreThread() random crash at exit with daemon threads. It now accesses the _PyRuntime variable directly instead of using tstate->interp->runtime , since tstate can be a dangling pointer after Py_Finalize() has been called. Moreover, the daemon thread now exits before trying to take the GIL.

bpo-39871 : Fix a possible SystemError in math.{atan2,copysign,remainder}() when the first argument cannot be converted to a float . Patch by Zackery Spytz.

bpo-39776 : Fix race condition where threads created by PyGILState_Ensure() could get a duplicate id.

This affects consumers of tstate->id like the contextvar caching machinery, which could return invalid cached objects under heavy thread load (observed in embedded scenarios).

bpo-39778 : Fixed a crash due to incorrect handling of weak references in collections.OrderedDict classes. Patch by Pablo Galindo.

bpo-1635741 : Port audioop extension module to multiphase initialization ( PEP 489 ).

bpo-39702 : Relax decorator grammar restrictions to allow any valid expression ( PEP 614 ).

bpo-38091 : Tweak import deadlock detection code to not deadlock itself.

bpo-1635741 : Port _locale extension module to multiphase initialization ( PEP 489 ).

bpo-39087 : Optimize PyUnicode_AsUTF8() and PyUnicode_AsUTF8AndSize() slightly when they need to create internal UTF-8 cache.

bpo-39520 : Fix unparsing of ext slices with no items ( foo[:,] ). Patch by Batuhan Taskaya.

bpo-39220 : Do not optimize annotations if ‘from __future__ import annotations’ is used. Patch by Pablo Galindo.

bpo-35712 : Using NotImplemented in a boolean context has been deprecated. Patch contributed by Josh Rosenberg.

bpo-22490 : Don’t leak environment variable __PYVENV_LAUNCHER__ into the interpreter session on macOS.

bpo-39830 : Add zipfile.Path to __all__ in the zipfile module.

bpo-40000 : Improved error messages for validation of ast.Constant nodes. Patch by Batuhan Taskaya.

bpo-39999 : __module__ of the AST node classes is now set to “ast” instead of “_ast”. Added docstrings for dummy AST node classes and deprecated attributes.

bpo-39991 : uuid.getnode() now skips IPv6 addresses with the same string length than a MAC address (17 characters): only use MAC addresses.

bpo-39988 : Deprecated ast.AugLoad and ast.AugStore node classes because they are no longer used.

bpo-39656 : Ensure bin/python3.# is always present in virtual environments on POSIX platforms - by Anthony Sottile.

bpo-39969 : Deprecated ast.Param node class because it’s no longer used. Patch by Batuhan Taskaya.

bpo-39360 : Ensure all workers exit when finalizing a multiprocessing.Pool implicitly via the module finalization handlers of multiprocessing. This fixes a deadlock situation that can be experienced when the Pool is not properly finalized via the context manager or a call to multiprocessing.Pool.terminate . Patch by Batuhan Taskaya and Pablo Galindo.

bpo-35370 : sys.settrace(), sys.setprofile() and _lsprof.Profiler.enable() now properly report PySys_Audit() error if “sys.setprofile” or “sys.settrace” audit event is denied.

bpo-39936 : AIX: Fix _aix_support module when the subprocess is not available, when building Python from scratch. It now uses new private _bootsubprocess module, rather than having two implementations depending if subprocess is available or not. So _aix_support.aix_platform() result is now the same if subprocess is available or not.

bpo-36144 : collections.OrderedDict now implements | and |= ( PEP 584 ).

bpo-39652 : The column name found in sqlite3.Cursor.description is now truncated on the first ‘[’ only if the PARSE_COLNAMES option is set.

bpo-39915 : Ensure unittest.mock.AsyncMock.await_args_list has call objects in the order of awaited arguments instead of using unittest.mock.Mock.call_args which has the last value of the call. Patch by Karthikeyan Singaravelan.

bpo-36144 : Updated os.environ and os.environb to support PEP 584 ’s merge ( | ) and update ( |= ) operators.

bpo-38662 : The ensurepip module now invokes pip via the runpy module. Hence it is no longer tightly coupled with the internal API of the bundled pip version, allowing easier updates to a newer pip version both internally and for distributors.

bpo-38075 : Fix the random.Random.seed() method when a bool is passed as the seed.

bpo-39916 : More reliable use of os.scandir() in Path.glob() . It no longer emits a ResourceWarning when interrupted.

bpo-39850 : multiprocessing now supports abstract socket addresses (if abstract sockets are supported in the running platform). When creating arbitrary addresses (like when default-constructing multiprocessing.connection.Listener objects) abstract sockets are preferred to avoid the case when the temporary-file-generated address is too large for an AF_UNIX socket address. Patch by Pablo Galindo.

bpo-36287 : ast.dump() no longer outputs optional fields and attributes with default values. The default values for optional fields and attributes of AST nodes are now set as class attributes (e.g. Constant.kind is set to None ).

bpo-39889 : Fixed ast.unparse() for extended slices containing a single element (e.g. a[i:j,] ). Remove redundant tuples when index with a tuple (e.g. a[i, j] ).

bpo-39828 : Fix json.tool to catch BrokenPipeError . Patch by Donghee Na.

bpo-13487 : Avoid a possible “RuntimeError: dictionary changed size during iteration” from inspect.getmodule() when it tried to loop through sys.modules .

bpo-39674 : Revert “ bpo-37330 : open() no longer accept ‘U’ in file mode”. The “U” mode of open() is kept in Python 3.9 to ease transition from Python 2.7, but will be removed in Python 3.10.

bpo-28577 : The hosts method on 32-bit prefix length IPv4Networks and 128-bit prefix IPv6Networks now returns a list containing the single Address instead of an empty list.

bpo-39826 : Add getConnection method to logging HTTPHandler to enable custom connections.

bpo-39763 : Reimplement distutils.spawn.spawn function with the subprocess module.

bpo-39794 : Add –without-decimal-contextvar build option. This enables a thread-local rather than a coroutine local context.

bpo-36144 : collections.defaultdict now implements | ( PEP 584 ).

bpo-39517 : Fix runpy.run_path() when using pathlike objects

bpo-39775 : Change inspect.Signature.parameters back to collections.OrderedDict . This was changed to dict in Python 3.9.0a4.

bpo-39678 : Refactor queue_manager in concurrent.futures.ProcessPoolExecutor to make it easier to maintain.

bpo-39764 : Fix AttributeError when calling get_stack on a PyAsyncGenObject Task

bpo-39769 : The compileall.compile_dir() function’s ddir parameter and the compileall command line flag -d no longer write the wrong pathname to the generated pyc file for submodules beneath the root of the directory tree being compiled. This fixes a regression introduced with Python 3.5.

bpo-36144 : types.MappingProxyType objects now support the merge ( | ) operator from PEP 584 .

bpo-38691 : The importlib module now ignores the PYTHONCASEOK environment variable when the -E or -I command line options are being used.

bpo-39719 : Remove tempfile.SpooledTemporaryFile.softspace() as files no longer have the softspace attribute in Python 3. Patch by Shantanu.

bpo-39667 : Improve pathlib.Path compatibility on zipfile.Path and correct performance degradation as found in zipp 3.0.

bpo-39638 : Keep ASDL signatures in the docstrings for AST nodes. Patch by Batuhan Taskaya

bpo-39639 : Deprecated ast.Suite node class because it’s no longer used. Patch by Batuhan Taskaya.

bpo-39609 : Add thread_name_prefix to default asyncio executor

bpo-39548 : Fix handling of header in urllib.request.AbstractDigestAuthHandler when the optional qop parameter is not present.

bpo-39509 : HTTP status codes 103 EARLY_HINTS and 425 TOO_EARLY are added to http.HTTPStatus . Patch by Donghee Na.

bpo-39507 : Adding HTTP status 418 “I’m a Teapot” to HTTPStatus in http library. Patch by Ross Rhodes.

bpo-39495 : Remove default value from attrs parameter of xml.etree.ElementTree.TreeBuilder.start() for consistency between Python and C implementations.

bpo-38971 : Open issue in the BPO indicated a desire to make the implementation of codecs.open() at parity with io.open(), which implements a try/except to assure file stream gets closed before an exception is raised.

bpo-38641 : Added starred expressions support to return and yield statements for lib2to3 . Patch by Vlad Emelianov.

bpo-37534 : When using minidom module to generate XML documents the ability to add Standalone Document Declaration is added. All the changes are made to generate a document in compliance with Extensible Markup Language (XML) 1.0 (Fifth Edition) W3C Recommendation (available here: https://www.w3.org/TR/xml/#sec-prolog-dtd ).

bpo-34788 : Add support for scoped IPv6 addresses to ipaddress . Patch by Oleksandr Pavliuk.

bpo-34822 : Simplified AST for subscription. Simple indices are now represented by their value, extended slices are represented as tuples. ast classes Index and ExtSlice are considered deprecated and will be removed in future Python versions. In the meantime, Index(value) now returns a value itself, ExtSlice(slices) returns Tuple(slices, Load()) .

bpo-39868 : Updated the Language Reference for PEP 572 .

bpo-13790 : Change ‘string’ to ‘specification’ in format doc.

bpo-17422 : The language reference no longer restricts default class namespaces to dicts only.

bpo-39530 : Fix misleading documentation about mixed-type numeric comparisons.

bpo-39718 : Update token documentation to reflect additions in Python 3.8

bpo-39677 : Changed operand name of MAKE_FUNCTION from argc to flags for module dis

bpo-40019 : test_gdb now skips tests if it detects that gdb failed to read debug information because the Python binary is optimized.

bpo-27807 : test_site.test_startup_imports() is now skipped if a path of sys.path contains a .pth file.

bpo-26067 : Do not fail test_shutil test_chown test when uid or gid of user cannot be resolved to a name.

bpo-39855 : test_subprocess.test_user() now skips the test on an user name if the user name doesn’t exist. For example, skip the test if the user “nobody” doesn’t exist on Linux.

bpo-39761 : Fix build with DTrace but without additional DFLAGS.

bpo-39763 : setup.py now uses a basic implementation of the subprocess module if the subprocess module is not available: before required C extension modules are built.

bpo-1294959 : Add --with-platlibdir option to the configure script: name of the platform-specific library directory, stored in the new sys.platlibdir attribute. It is used to build the path of platform-specific extension modules and the path of the standard library. It is equal to "lib" on most platforms. On Fedora and SuSE, it is equal to "lib64" on 64-bit platforms. Patch by Jan Matějek, Matěj Cepl, Charalampos Stratakis and Victor Stinner.

bpo-39930 : Ensures the required vcruntime140.dll is included in install packages.

bpo-39847 : Avoid hang when computer is hibernated whilst waiting for a mutex (for lock-related objects from threading ) around 49-day uptime.

bpo-38597 : distutils will no longer statically link vcruntime140.dll when a redistributable version is unavailable. All future releases of CPython will include a copy of this DLL to ensure distributed extensions can continue to load.

bpo-38380 : Update Windows builds to use SQLite 3.31.1

bpo-39789 : Update Windows release build machines to Visual Studio 2019 (MSVC 14.2).

bpo-34803 : Package for nuget.org now includes repository reference and bundled icon image.

bpo-38380 : Update macOS builds to use SQLite 3.31.1

bpo-27115 : For ‘Go to Line’, use a Query box subclass with IDLE standard behavior and improved error checking.

bpo-39885 : Since clicking to get an IDLE context menu moves the cursor, any text selection should be and now is cleared.

bpo-39852 : Edit “Go to line” now clears any selection, preventing accidental deletion. It also updates Ln and Col on the status bar.

bpo-39781 : Selecting code context lines no longer causes a jump.

bpo-36184 : Port python-gdb.py to FreeBSD. python-gdb.py now checks for “take_gil” function name to check if a frame tries to acquire the GIL, instead of checking for “pthread_cond_timedwait” which is specific to Linux and can be a different condition than the GIL.

bpo-38080 : Added support to fix getproxies in the lib2to3.fixes.fix_urllib module. Patch by José Roberto Meza Cabrera.

bpo-40024 : Add PyModule_AddType() helper function: add a type to a module. Patch by Donghee Na.

bpo-39946 : Remove _PyRuntime.getframe hook and remove _PyThreadState_GetFrame macro which was an alias to _PyRuntime.getframe . They were only exposed by the internal C API. Remove also PyThreadFrameGetter type.

bpo-39947 : Add PyThreadState_GetFrame() function: get the current frame of a Python thread state.

bpo-37207 : Add _PyArg_NoKwnames helper function. Patch by Donghee Na.

bpo-39947 : Add PyThreadState_GetInterpreter() : get the interpreter of a Python thread state.

bpo-39947 : Add PyInterpreterState_Get() function to the limited C API.

bpo-35370 : If PySys_Audit() fails in PyEval_SetProfile() or PyEval_SetTrace() , log the error as an unraisable exception.

bpo-39947 : Move the static inline function flavor of Py_EnterRecursiveCall() and Py_LeaveRecursiveCall() to the internal C API: they access PyThreadState attributes. The limited C API provides regular functions which hide implementation details.

bpo-39947 : Py_TRASHCAN_BEGIN_CONDITION and Py_TRASHCAN_END macro no longer access PyThreadState attributes, but call new private _PyTrash_begin() and _PyTrash_end() functions which hide implementation details.

bpo-39884 : PyDescr_NewMethod() and PyCFunction_NewEx() now include the method name in the SystemError “bad call flags” error message to ease debug.

bpo-39877 : Deprecated PyEval_InitThreads() and PyEval_ThreadsInitialized() . Calling PyEval_InitThreads() now does nothing.