What’s New In Python 3.9

Editor

Łukasz Langa

This article explains the new features in Python 3.9, compared to 3.8. Python 3.9 was released on October 5, 2020.

For full details, see the changelog.

See also

PEP 596 - Python 3.9 Release Schedule

Summary – Release highlights

New syntax features:

  • PEP 584, union operators added to dict;

  • PEP 585, type hinting generics in standard collections;

  • PEP 614, relaxed grammar restrictions on decorators.

New built-in features:

  • PEP 616, string methods to remove prefixes and suffixes.

New features in the standard library:

  • PEP 593, flexible function and variable annotations;

  • os.pidfd_open() added that allows process management without races and signals.

Interpreter improvements:

  • PEP 573, fast access to module state from methods of C extension types;

  • PEP 617, CPython now uses a new parser based on PEG;

  • a number of Python builtins (range, tuple, set, frozenset, list, dict) are now sped up using PEP 590 vectorcall;

  • garbage collection does not block on resurrected objects;

  • a number of Python modules (_abc, audioop, _bz2, _codecs, _contextvars, _crypt, _functools, _json, _locale, math, operator, resource, time, _weakref) now use multiphase initialization as defined by PEP 489;

  • a number of standard library modules (audioop, ast, grp, _hashlib, pwd, _posixsubprocess, random, select, struct, termios, zlib) are now using the stable ABI defined by PEP 384.

New library modules:

  • PEP 615, the IANA Time Zone Database is now present in the standard library in the zoneinfo module;

  • an implementation of a topological sort of a graph is now provided in the new graphlib module.

Release process changes:

  • PEP 602, CPython adopts an annual release cycle.

You should check for DeprecationWarning in your code

When Python 2.7 was still supported, a lot of functionality in Python 3 was kept for backward compatibility with Python 2.7. With the end of Python 2 support, these backward compatibility layers have been removed, or will be removed soon. Most of them emitted a DeprecationWarning warning for several years. For example, using collections.Mapping instead of collections.abc.Mapping emits a DeprecationWarning since Python 3.3, released in 2012.

Test your application with the -W default command-line option to see DeprecationWarning and PendingDeprecationWarning, or even with -W error to treat them as errors. Warnings Filter can be used to ignore warnings from third-party code.

Python 3.9 is the last version providing those Python 2 backward compatibility layers, to give more time to Python projects maintainers to organize the removal of the Python 2 support and add support for Python 3.9.

Aliases to Abstract Base Classes in the collections module, like collections.Mapping alias to collections.abc.Mapping, are kept for one last release for backward compatibility. They will be removed from Python 3.10.

More generally, try to run your tests in the Python Development Mode which helps to prepare your code to make it compatible with the next Python version.

Note: a number of pre-existing deprecations were removed in this version of Python as well. Consult the Removed section.

New Features

Dictionary Merge & Update Operators

Merge (|) and update (|=) operators have been added to the built-in dict class. Those complement the existing dict.update and {**d1, **d2} methods of merging dictionaries.

Example:

>>> x = {"key1": "value1 from x", "key2": "value2 from x"}
>>> y = {"key2": "value2 from y", "key3": "value3 from y"}
>>> x | y
{'key1': 'value1 from x', 'key2': 'value2 from y', 'key3': 'value3 from y'}
>>> y | x
{'key2': 'value2 from x', 'key3': 'value3 from y', 'key1': 'value1 from x'}

See PEP 584 for a full description. (Contributed by Brandt Bucher in bpo-36144.)

New String Methods to Remove Prefixes and Suffixes

str.removeprefix(prefix) and str.removesuffix(suffix) have been added to easily remove an unneeded prefix or a suffix from a string. Corresponding bytes, bytearray, and collections.UserString methods have also been added. See PEP 616 for a full description. (Contributed by Dennis Sweeney in bpo-39939.)

Type Hinting Generics in Standard Collections

In type annotations you can now use built-in collection types such as list and dict as generic types instead of importing the corresponding capitalized types (e.g. List or Dict) from typing. Some other types in the standard library are also now generic, for example queue.Queue.

Example:

def greet_all(names: list[str]) -> None:
    for name in names:
        print("Hello", name)

See PEP 585 for more details. (Contributed by Guido van Rossum, Ethan Smith, and Batuhan Taşkaya in bpo-39481.)

New Parser

Python 3.9 uses a new parser, based on PEG instead of LL(1). The new parser’s performance is roughly comparable to that of the old parser, but the PEG formalism is more flexible than LL(1) when it comes to designing new language features. We’ll start using this flexibility in Python 3.10 and later.

The ast module uses the new parser and produces the same AST as the old parser.

In Python 3.10, the old parser will be deleted and so will all functionality that depends on it (primarily the parser module, which has long been deprecated). In Python 3.9 only, you can switch back to the LL(1) parser using a command line switch (-X oldparser) or an environment variable (PYTHONOLDPARSER=1).

See PEP 617 for more details. (Contributed by Guido van Rossum, Pablo Galindo and Lysandros Nikolaou in bpo-40334.)

Other Language Changes

  • __import__() now raises ImportError instead of ValueError, which used to occur when a relative import went past its top-level package. (Contributed by Ngalim Siregar in bpo-37444.)

  • Python now gets the absolute path of the script filename specified on the command line (ex: python3 script.py): the __file__ attribute of the __main__ module became an absolute path, rather than a relative path. These paths now remain valid after the current directory is changed by os.chdir(). As a side effect, the traceback also displays the absolute path for __main__ module frames in this case. (Contributed by Victor Stinner in bpo-20443.)

  • In the Python Development Mode and in debug build, the encoding and errors arguments are now checked for string encoding and decoding operations. Examples: open(), str.encode() and bytes.decode().

    By default, for best performance, the errors argument is only checked at the first encoding/decoding error and the encoding argument is sometimes ignored for empty strings. (Contributed by Victor Stinner in bpo-37388.)

  • "".replace("", s, n) now returns s instead of an empty string for all non-zero n. It is now consistent with "".replace("", s). There are similar changes for bytes and bytearray objects. (Contributed by Serhiy Storchaka in bpo-28029.)

  • Any valid expression can now be used as a decorator. Previously, the grammar was much more restrictive. See PEP 614 for details. (Contributed by Brandt Bucher in bpo-39702.)

  • 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). (Contributed by Serhiy Storchaka in bpo-40257.)

  • Parallel running of aclose() / asend() / athrow() is now prohibited, and ag_running now reflects the actual running status of the async generator. (Contributed by Yury Selivanov in bpo-30773.)

  • 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. (Contributed by Serhiy Storchaka in bpo-40824.)

  • Unparenthesized lambda expressions can no longer be the expression part in an if clause in comprehensions and generator expressions. See bpo-41848 and bpo-43755 for details.

New Modules

zoneinfo

The zoneinfo module brings support for the IANA time zone database to the standard library. It adds zoneinfo.ZoneInfo, a concrete datetime.tzinfo implementation backed by the system’s time zone data.

Example:

>>> from zoneinfo import ZoneInfo
>>> from datetime import datetime, timedelta

>>> # Daylight saving time
>>> dt = datetime(2020, 10, 31, 12, tzinfo=ZoneInfo("America/Los_Angeles"))
>>> print(dt)
2020-10-31 12:00:00-07:00
>>> dt.tzname()
'PDT'

>>> # Standard time
>>> dt += timedelta(days=7)
>>> print(dt)
2020-11-07 12:00:00-08:00
>>> print(dt.tzname())
PST

As a fall-back source of data for platforms that don’t ship the IANA database, the tzdata module was released as a first-party package – distributed via PyPI and maintained by the CPython core team.

See also

PEP 615 – Support for the IANA Time Zone Database in the Standard Library

PEP written and implemented by Paul Ganssle

graphlib

A new module, graphlib, was added that contains the graphlib.TopologicalSorter class to offer functionality to perform topological sorting of graphs. (Contributed by Pablo Galindo, Tim Peters and Larry Hastings in bpo-17005.)

Improved Modules

ast

Added the indent option to dump() which allows it to produce a multiline indented output. (Contributed by Serhiy Storchaka in bpo-37995.)

Added ast.unparse() as a function in the ast module that can be used to unparse an ast.AST object and produce a string with code that would produce an equivalent ast.AST object when parsed. (Contributed by Pablo Galindo and Batuhan Taskaya in bpo-38870.)

Added docstrings to AST nodes that contains the ASDL signature used to construct that node. (Contributed by Batuhan Taskaya in bpo-39638.)

asyncio

Due to significant security concerns, the reuse_address parameter of asyncio.loop.create_datagram_endpoint() is no longer supported. This is because of the behavior of the socket option SO_REUSEADDR in UDP. For more details, see the documentation for loop.create_datagram_endpoint(). (Contributed by Kyle Stanley, Antoine Pitrou, and Yury Selivanov in bpo-37228.)

Added a new coroutine shutdown_default_executor() that schedules a shutdown for the default executor that waits on the ThreadPoolExecutor to finish closing. Also, asyncio.run() has been updated to use the new coroutine. (Contributed by Kyle Stanley in bpo-34037.)

Added asyncio.PidfdChildWatcher, a Linux-specific child watcher implementation that polls process file descriptors. (bpo-38692)

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. (Contributed by Kyle Stanley and Yury Selivanov in bpo-32309.)

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. (Contributed by Elvis Pranskevichus in bpo-32751.)

asyncio now raises TyperError when calling incompatible methods with an ssl.SSLSocket socket. (Contributed by Ido Michael in bpo-37404.)

compileall

Added new possibility to use hardlinks for duplicated .pyc files: hardlink_dupes parameter and –hardlink-dupes command line option. (Contributed by Lumír ‘Frenzy’ Balhar in bpo-40495.)

Added new options for path manipulation in resulting .pyc files: stripdir, prependdir, limit_sl_dest parameters and -s, -p, -e command line options. Added the possibility to specify the option for an optimization level multiple times. (Contributed by Lumír ‘Frenzy’ Balhar in bpo-38112.)

concurrent.futures

Added a new cancel_futures parameter to concurrent.futures.Executor.shutdown() that cancels all pending futures which have not started running, instead of waiting for them to complete before shutting down the executor. (Contributed by Kyle Stanley in bpo-39349.)

Removed daemon threads from ThreadPoolExecutor and ProcessPoolExecutor. This improves compatibility with subinterpreters and predictability in their shutdown processes. (Contributed by Kyle Stanley in bpo-39812.)

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. (Contributed by Kyle Stanley in bpo-39207.)

curses

Added curses.get_escdelay(), curses.set_escdelay(), curses.get_tabsize(), and curses.set_tabsize() functions. (Contributed by Anthony Sottile in bpo-38312.)

datetime

The isocalendar() of datetime.date and isocalendar() of datetime.datetime methods now returns a namedtuple() instead of a tuple. (Contributed by Dong-hee Na in bpo-24416.)

distutils

The upload command now creates SHA2-256 and Blake2b-256 hash digests. It skips MD5 on platforms that block MD5 digest. (Contributed by Christian Heimes in bpo-40698.)

fcntl

Added constants F_OFD_GETLK, F_OFD_SETLK and F_OFD_SETLKW. (Contributed by Dong-hee Na in bpo-38602.)

ftplib

FTP and FTP_TLS now raise a ValueError if the given timeout for their constructor is zero to prevent the creation of a non-blocking socket. (Contributed by Dong-hee Na in bpo-39259.)

gc

When the garbage collector makes a collection in which some objects resurrect (they are reachable from outside the isolated cycles after the finalizers have been executed), do not block the collection of all objects that are still unreachable. (Contributed by Pablo Galindo and Tim Peters in bpo-38379.)

Added a new function gc.is_finalized() to check if an object has been finalized by the garbage collector. (Contributed by Pablo Galindo in bpo-39322.)

hashlib

The hashlib module can now use SHA3 hashes and SHAKE XOF from OpenSSL when available. (Contributed by Christian Heimes in bpo-37630.)

Builtin hash modules can now be disabled with ./configure --without-builtin-hashlib-hashes or selectively enabled with e.g. ./configure --with-builtin-hashlib-hashes=sha3,blake2 to force use of OpenSSL based implementation. (Contributed by Christian Heimes in bpo-40479)

http

HTTP status codes 103 EARLY_HINTS, 418 IM_A_TEAPOT and 425 TOO_EARLY are added to http.HTTPStatus. (Contributed by Dong-hee Na in bpo-39509 and Ross Rhodes in bpo-39507.)

IDLE and idlelib

Added option to toggle cursor blink off. (Contributed by Zackery Spytz in bpo-4603.)

Escape key now closes IDLE completion windows. (Contributed by Johnny Najera in bpo-38944.)

Added keywords to module name completion list. (Contributed by Terry J. Reedy in bpo-37765.)

New in 3.9 maintenance releases

Make IDLE invoke sys.excepthook() (when started without ‘-n’). User hooks were previously ignored. (Contributed by Ken Hilton in bpo-43008.)

The changes above have been backported to 3.8 maintenance releases.

Rearrange the settings dialog. Split the 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. (Contributed by Terry Jan Reedy in bpo-40468.) Move the indent space setting from the Font tab to the new Windows tab. (Contributed by Mark Roseman and Terry Jan Reedy in bpo-33962.)

Apply syntax highlighting to .pyi files. (Contributed by Alex Waygood and Terry Jan Reedy in bpo-45447.)

imaplib

IMAP4 and IMAP4_SSL now have an optional timeout parameter for their constructors. Also, the open() method now has an optional timeout parameter with this change. The overridden methods of IMAP4_SSL and IMAP4_stream were applied to this change. (Contributed by Dong-hee Na in bpo-38615.)

imaplib.IMAP4.unselect() is added. imaplib.IMAP4.unselect() frees server’s resources associated with the selected mailbox and returns the server to the authenticated state. This command performs the same actions as imaplib.IMAP4.close(), except that no messages are permanently removed from the currently selected mailbox. (Contributed by Dong-hee Na in bpo-40375.)

importlib

To improve consistency with import statements, importlib.util.resolve_name() now raises ImportError instead of ValueError for invalid relative import attempts. (Contributed by Ngalim Siregar in bpo-37444.)

Import loaders which publish immutable module objects can now publish immutable packages in addition to individual modules. (Contributed by Dino Viehland in bpo-39336.)

Added importlib.resources.files() function with support for subdirectories in package data, matching backport in importlib_resources version 1.5. (Contributed by Jason R. Coombs in bpo-39791.)

Refreshed importlib.metadata from importlib_metadata version 1.6.1.

inspect

inspect.BoundArguments.arguments is changed from OrderedDict to regular dict. (Contributed by Inada Naoki in bpo-36350 and bpo-39775.)

ipaddress

ipaddress now supports IPv6 Scoped Addresses (IPv6 address with suffix %<scope_id>).

Scoped IPv6 addresses can be parsed using ipaddress.IPv6Address. If present, scope zone ID is available through the scope_id attribute. (Contributed by Oleksandr Pavliuk in bpo-34788.)

Starting with Python 3.9.5 the ipaddress module no longer accepts any leading zeros in IPv4 address strings. (Contributed by Christian Heimes in bpo-36384).

math

Expanded the math.gcd() function to handle multiple arguments. Formerly, it only supported two arguments. (Contributed by Serhiy Storchaka in bpo-39648.)

Added math.lcm(): return the least common multiple of specified arguments. (Contributed by Mark Dickinson, Ananthakrishnan and Serhiy Storchaka in bpo-39479 and bpo-39648.)

Added math.nextafter(): return the next floating-point value after x towards y. (Contributed by Victor Stinner in bpo-39288.)

Added math.ulp(): return the value of the least significant bit of a float. (Contributed by Victor Stinner in bpo-39310.)

multiprocessing

The multiprocessing.SimpleQueue class has a new close() method to explicitly close the queue. (Contributed by Victor Stinner in bpo-30966.)

nntplib

NNTP and NNTP_SSL now raise a ValueError if the given timeout for their constructor is zero to prevent the creation of a non-blocking socket. (Contributed by Dong-hee Na in bpo-39259.)

os

Added CLD_KILLED and CLD_STOPPED for si_code. (Contributed by Dong-hee Na in bpo-38493.)

Exposed the Linux-specific os.pidfd_open() (bpo-38692) and os.P_PIDFD (bpo-38713) for process management with file descriptors.

The os.unsetenv() function is now also available on Windows. (Contributed by Victor Stinner in bpo-39413.)

The os.putenv() and os.unsetenv() functions are now always available. (Contributed by Victor Stinner in bpo-39395.)

Added os.waitstatus_to_exitcode() function: convert a wait status to an exit code. (Contributed by Victor Stinner in bpo-40094.)

pathlib

Added pathlib.Path.readlink() which acts similarly to os.readlink(). (Contributed by Girts Folkmanis in bpo-30618)

pdb

On Windows now Pdb supports ~/.pdbrc. (Contributed by Tim Hopper and Dan Lidral-Porter in bpo-20523.)

poplib

POP3 and POP3_SSL now raise a ValueError if the given timeout for their constructor is zero to prevent the creation of a non-blocking socket. (Contributed by Dong-hee Na in bpo-39259.)

pprint

pprint can now pretty-print types.SimpleNamespace. (Contributed by Carl Bordum Hansen in bpo-37376.)

pydoc

The documentation string is now shown not only for class, function, method etc, but for any object that has its own __doc__ attribute. (Contributed by Serhiy Storchaka in bpo-40257.)

random

Added a new random.Random.randbytes method: generate random bytes. (Contributed by Victor Stinner in bpo-40286.)

signal

Exposed the Linux-specific signal.pidfd_send_signal() for sending to signals to a process using a file descriptor instead of a pid. (bpo-38712)

smtplib

SMTP and SMTP_SSL now raise a ValueError if the given timeout for their constructor is zero to prevent the creation of a non-blocking socket. (Contributed by Dong-hee Na in bpo-39259.)

LMTP constructor now has an optional timeout parameter. (Contributed by Dong-hee Na in bpo-39329.)

socket

The socket module now exports the CAN_RAW_JOIN_FILTERS constant on Linux 4.1 and greater. (Contributed by Stefan Tatschner and Zackery Spytz in bpo-25780.)

The socket module now supports the CAN_J1939 protocol on platforms that support it. (Contributed by Karl Ding in bpo-40291.)

The socket module now has the socket.send_fds() and socket.recv_fds() functions. (Contributed by Joannah Nanjekye, Shinya Okano and Victor Stinner in bpo-28724.)

time

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. (Contributed by Batuhan Taskaya in bpo-40192)

sys

Added a new sys.platlibdir attribute: name of the platform-specific library directory. It is used to build the path of standard library and the paths of installed extension modules. It is equal to "lib" on most platforms. On Fedora and SuSE, it is equal to "lib64" on 64-bit platforms. (Contributed by Jan Matějek, Matěj Cepl, Charalampos Stratakis and Victor Stinner in bpo-1294959.)

Previously, sys.stderr was block-buffered when non-interactive. Now stderr defaults to always being line-buffered. (Contributed by Jendrik Seipp in bpo-13601.)

tracemalloc

Added tracemalloc.reset_peak() to set the peak size of traced memory blocks to the current size, to measure the peak of specific pieces of code. (Contributed by Huon Wilson in bpo-40630.)

typing

PEP 593 introduced an typing.Annotated type to decorate existing types with context-specific metadata and new include_extras parameter to typing.get_type_hints() to access the metadata at runtime. (Contributed by Till Varoquaux and Konstantin Kashin.)

unicodedata

The Unicode database has been updated to version 13.0.0. (bpo-39926).

venv

The activation scripts provided by venv now all specify their prompt customization consistently by always using the value specified by __VENV_PROMPT__. Previously some scripts unconditionally used __VENV_PROMPT__, others only if it happened to be set (which was the default case), and one used __VENV_NAME__ instead. (Contributed by Brett Cannon in bpo-37663.)

xml

White space characters within attributes are now preserved when serializing xml.etree.ElementTree to XML file. EOLNs are no longer normalized to “n”. This is the result of discussion about how to interpret section 2.11 of XML spec. (Contributed by Mefistotelis in bpo-39011.)

Optimizations

  • Optimized the idiom for assignment a temporary variable in comprehensions. Now for y in [expr] in comprehensions is as fast as a simple assignment y = expr. For example:

    sums = [s for s in [0] for x in data for s in [s + x]]

    Unlike the := operator this idiom does not leak a variable to the outer scope.

    (Contributed by Serhiy Storchaka in bpo-32856.)

  • Optimized 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. (Contributed by Victor Stinner in bpo-40010.)

  • Optimized the subprocess module on FreeBSD using closefrom(). (Contributed by Ed Maste, Conrad Meyer, Kyle Evans, Kubilay Kocak and Victor Stinner in bpo-38061.)

  • PyLong_FromDouble() is now up to 1.87x faster for values that fit into long. (Contributed by Sergey Fedoseev in bpo-37986.)

  • A number of Python builtins (range, tuple, set, frozenset, list, dict) are now sped up by using PEP 590 vectorcall protocol. (Contributed by Dong-hee Na, Mark Shannon, Jeroen Demeyer and Petr Viktorin in bpo-37207.)

  • Optimized difference_update() for the case when the other set is much larger than the base set. (Suggested by Evgeny Kapun with code contributed by Michele Orrù in bpo-8425.)

  • Python’s small object allocator (obmalloc.c) now allows (no more than) one empty arena to remain available for immediate reuse, without returning it to the OS. This prevents thrashing in simple loops where an arena could be created and destroyed anew on each iteration. (Contributed by Tim Peters in bpo-37257.)

  • floor division of float operation now has a better performance. Also the message of ZeroDivisionError for this operation is updated. (Contributed by Dong-hee Na in bpo-39434.)

  • Decoding short ASCII strings with UTF-8 and ascii codecs is now about 15% faster. (Contributed by Inada Naoki in bpo-37348.)

Here’s a summary of performance improvements from Python 3.4 through Python 3.9:

Python version                       3.4     3.5     3.6     3.7     3.8    3.9
--------------                       ---     ---     ---     ---     ---    ---

Variable and attribute read access:
    read_local                       7.1     7.1     5.4     5.1     3.9    3.9
    read_nonlocal                    7.1     8.1     5.8     5.4     4.4    4.5
    read_global                     15.5    19.0    14.3    13.6     7.6    7.8
    read_builtin                    21.1    21.6    18.5    19.0     7.5    7.8
    read_classvar_from_class        25.6    26.5    20.7    19.5    18.4   17.9
    read_classvar_from_instance     22.8    23.5    18.8    17.1    16.4   16.9
    read_instancevar                32.4    33.1    28.0    26.3    25.4   25.3
    read_instancevar_slots          27.8    31.3    20.8    20.8    20.2   20.5
    read_namedtuple                 73.8    57.5    45.0    46.8    18.4   18.7
    read_boundmethod                37.6    37.9    29.6    26.9    27.7   41.1

Variable and attribute write access:
    write_local                      8.7     9.3     5.5     5.3     4.3    4.3
    write_nonlocal                  10.5    11.1     5.6     5.5     4.7    4.8
    write_global                    19.7    21.2    18.0    18.0    15.8   16.7
    write_classvar                  92.9    96.0   104.6   102.1    39.2   39.8
    write_instancevar               44.6    45.8    40.0    38.9    35.5   37.4
    write_instancevar_slots         35.6    36.1    27.3    26.6    25.7   25.8

Data structure read access:
    read_list                       24.2    24.5    20.8    20.8    19.0   19.5
    read_deque                      24.7    25.5    20.2    20.6    19.8   20.2
    read_dict                       24.3    25.7    22.3    23.0    21.0   22.4
    read_strdict                    22.6    24.3    19.5    21.2    18.9   21.5

Data structure write access:
    write_list                      27.1    28.5    22.5    21.6    20.0   20.0
    write_deque                     28.7    30.1    22.7    21.8    23.5   21.7
    write_dict                      31.4    33.3    29.3    29.2    24.7   25.4
    write_strdict                   28.4    29.9    27.5    25.2    23.1   24.5

Stack (or queue) operations:
    list_append_pop                 93.4   112.7    75.4    74.2    50.8   50.6
    deque_append_pop                43.5    57.0    49.4    49.2    42.5   44.2
    deque_append_popleft            43.7    57.3    49.7    49.7    42.8   46.4

Timing loop:
    loop_overhead                    0.5     0.6     0.4     0.3     0.3    0.3

These results were generated from the variable access benchmark script at: Tools/scripts/var_access_benchmark.py. The benchmark script displays timings in nanoseconds. The benchmarks were measured on an Intel® Core™ i7-4960HQ processor running the macOS 64-bit builds found at python.org.

Deprecated

  • The distutils bdist_msi command is now deprecated, use bdist_wheel (wheel packages) instead. (Contributed by Hugo van Kemenade in bpo-39586.)

  • Currently math.factorial() accepts float instances with non-negative integer values (like 5.0). It raises a ValueError for non-integral and negative floats. It is now deprecated. In future Python versions it will raise a TypeError for all floats. (Contributed by Serhiy Storchaka in bpo-37315.)

  • The parser and symbol modules are deprecated and will be removed in future versions of Python. For the majority of use cases, users can leverage the Abstract Syntax Tree (AST) generation and compilation stage, using the ast module.

  • The Public C API functions PyParser_SimpleParseStringFlags(), PyParser_SimpleParseStringFlagsFilename(), PyParser_SimpleParseFileFlags() and PyNode_Compile() are deprecated and will be removed in Python 3.10 together with the old parser.

  • Using NotImplemented in a boolean context has been deprecated, as it is almost exclusively the result of incorrect rich comparator implementations. It will be made a TypeError in a future version of Python. (Contributed by Josh Rosenberg in bpo-35712.)

  • The random module currently accepts any hashable type as a possible seed value. Unfortunately, some of those types are not guaranteed to have a deterministic hash value. After Python 3.9, the module will restrict its seeds to None, int, float, str, bytes, and bytearray.

  • Opening the GzipFile file for writing without specifying the mode argument is deprecated. In future Python versions it will always be opened for reading by default. Specify the mode argument for opening it for writing and silencing a warning. (Contributed by Serhiy Storchaka in bpo-28286.)

  • Deprecated the split() method of _tkinter.TkappType in favour of the splitlist() method which has more consistent and predicable behavior. (Contributed by Serhiy Storchaka in bpo-38371.)

  • The explicit passing of coroutine objects to asyncio.wait() has been deprecated and will be removed in version 3.11. (Contributed by Yury Selivanov and Kyle Stanley in bpo-34790.)

  • binhex4 and hexbin4 standards are now deprecated. The binhex module and the following binascii functions are now deprecated:

    • b2a_hqx(), a2b_hqx()

    • rlecode_hqx(), rledecode_hqx()

    (Contributed by Victor Stinner in bpo-39353.)

  • ast classes slice, Index and ExtSlice are considered deprecated and will be removed in future Python versions. value itself should be used instead of Index(value). Tuple(slices, Load()) should be used instead of ExtSlice(slices). (Contributed by Serhiy Storchaka in bpo-34822.)

  • ast classes Suite, Param, AugLoad and AugStore are considered deprecated and will be removed in future Python versions. They were not generated by the parser and not accepted by the code generator in Python 3. (Contributed by Batuhan Taskaya in bpo-39639 and bpo-39969 and Serhiy Storchaka in bpo-39988.)

  • The PyEval_InitThreads() and PyEval_ThreadsInitialized() functions are now deprecated and will be removed in Python 3.11. Calling PyEval_InitThreads() now does nothing. The GIL is initialized by Py_Initialize() since Python 3.7. (Contributed by Victor Stinner in bpo-39877.)

  • Passing None as the first argument to the shlex.split() function has been deprecated. (Contributed by Zackery Spytz in bpo-33262.)

  • smtpd.MailmanProxy() is now deprecated as it is unusable without an external module, mailman. (Contributed by Samuel Colvin in bpo-35800.)

  • The lib2to3 module now emits a PendingDeprecationWarning. Python 3.9 switched to a PEG parser (see PEP 617), and Python 3.10 may include new language syntax that is not parsable by lib2to3’s LL(1) parser. The lib2to3 module may be removed from the standard library in a future Python version. Consider third-party alternatives such as LibCST or parso. (Contributed by Carl Meyer in bpo-40360.)

  • The random parameter of random.shuffle() has been deprecated. (Contributed by Raymond Hettinger in bpo-40465)

Removed

  • The erroneous version at unittest.mock.__version__ has been removed.

  • nntplib.NNTP: xpath() and xgtitle() methods have been removed. These methods are deprecated since Python 3.3. Generally, these extensions are not supported or not enabled by NNTP server administrators. For xgtitle(), please use nntplib.NNTP.descriptions() or nntplib.NNTP.description() instead. (Contributed by Dong-hee Na in bpo-39366.)

  • array.array: tostring() and fromstring() methods have been removed. They were aliases to tobytes() and frombytes(), deprecated since Python 3.2. (Contributed by Victor Stinner in bpo-38916.)

  • The undocumented sys.callstats() function has been removed. Since Python 3.7, it was deprecated and always returned None. It required a special build option CALL_PROFILE which was already removed in Python 3.7. (Contributed by Victor Stinner in bpo-37414.)

  • The sys.getcheckinterval() and sys.setcheckinterval() functions have been removed. They were deprecated since Python 3.2. Use sys.getswitchinterval() and sys.setswitchinterval() instead. (Contributed by Victor Stinner in bpo-37392.)

  • The C function PyImport_Cleanup() has been removed. It was documented as: “Empty the module table. For internal use only.” (Contributed by Victor Stinner in bpo-36710.)

  • _dummy_thread and dummy_threading modules have been removed. These modules were deprecated since Python 3.7 which requires threading support. (Contributed by Victor Stinner in bpo-37312.)

  • aifc.openfp() alias to aifc.open(), sunau.openfp() alias to sunau.open(), and wave.openfp() alias to wave.open() have been removed. They were deprecated since Python 3.7. (Contributed by Victor Stinner in bpo-37320.)

  • The isAlive() method of threading.Thread has been removed. It was deprecated since Python 3.8. Use is_alive() instead. (Contributed by Dong-hee Na in bpo-37804.)

  • Methods getchildren() and getiterator() of classes ElementTree and Element in the ElementTree module have been removed. They were deprecated in Python 3.2. Use iter(x) or list(x) instead of x.getchildren() and x.iter() or list(x.iter()) instead of x.getiterator(). (Contributed by Serhiy Storchaka in bpo-36543.)

  • The old plistlib API has been removed, it was deprecated since Python 3.4. Use the load(), loads(), dump(), and dumps() functions. Additionally, the use_builtin_types parameter was removed, standard bytes objects are always used instead. (Contributed by Jon Janzen in bpo-36409.)

  • The C function PyGen_NeedsFinalizing has been removed. It was not documented, tested, or used anywhere within CPython after the implementation of PEP 442. Patch by Joannah Nanjekye. (Contributed by Joannah Nanjekye in bpo-15088)

  • base64.encodestring() and base64.decodestring(), aliases deprecated since Python 3.1, have been removed: use base64.encodebytes() and base64.decodebytes() instead. (Contributed by Victor Stinner in bpo-39351.)

  • fractions.gcd() function has been removed, it was deprecated since Python 3.5 (bpo-22486): use math.gcd() instead. (Contributed by Victor Stinner in bpo-39350.)

  • The buffering parameter of bz2.BZ2File has been removed. Since Python 3.0, it was ignored and using it emitted a DeprecationWarning. Pass an open file object to control how the file is opened. (Contributed by Victor Stinner in bpo-39357.)

  • The encoding parameter of json.loads() has been removed. As of Python 3.1, it was deprecated and ignored; using it has emitted a DeprecationWarning since Python 3.8. (Contributed by Inada Naoki in bpo-39377)

  • with (await asyncio.lock): and with (yield from asyncio.lock): statements are not longer supported, use async with lock instead. The same is correct for asyncio.Condition and asyncio.Semaphore. (Contributed by Andrew Svetlov in bpo-34793.)

  • The sys.getcounts() function, the -X showalloccount command line option and the show_alloc_count field of the C structure PyConfig have been removed. They required a special Python build by defining COUNT_ALLOCS macro. (Contributed by Victor Stinner in bpo-39489.)

  • The _field_types attribute of the typing.NamedTuple class has been removed. It was deprecated since Python 3.8. Use the __annotations__ attribute instead. (Contributed by Serhiy Storchaka in bpo-40182.)

  • The symtable.SymbolTable.has_exec() method has been removed. It was deprecated since 2006, and only returning False when it’s called. (Contributed by Batuhan Taskaya in bpo-40208)

  • The asyncio.Task.current_task() and asyncio.Task.all_tasks() have been removed. They were deprecated since Python 3.7 and you can use asyncio.current_task() and asyncio.all_tasks() instead. (Contributed by Rémi Lapeyre in bpo-40967)

  • The unescape() method in the html.parser.HTMLParser class has been removed (it was deprecated since Python 3.4). html.unescape() should be used for converting character references to the corresponding unicode characters.

Porting to Python 3.9

This section lists previously described changes and other bugfixes that may require changes to your code.

Changes in the Python API

  • __import__() and importlib.util.resolve_name() now raise ImportError where it previously raised ValueError. Callers catching the specific exception type and supporting both Python 3.9 and earlier versions will need to catch both using except (ImportError, ValueError):.

  • The venv activation scripts no longer special-case when __VENV_PROMPT__ is set to "".

  • The select.epoll.unregister() method no longer ignores the EBADF error. (Contributed by Victor Stinner in bpo-39239.)

  • The compresslevel parameter of bz2.BZ2File became keyword-only, since the buffering parameter has been removed. (Contributed by Victor Stinner in bpo-39357.)

  • Simplified AST for subscription. Simple indices will be represented by their value, extended slices will be represented as tuples. Index(value) will return a value itself, ExtSlice(slices) will return Tuple(slices, Load()). (Contributed by Serhiy Storchaka in bpo-34822.)

  • The importlib module now ignores the PYTHONCASEOK environment variable when the -E or -I command line options are being used.

  • The encoding parameter has been added to the classes ftplib.FTP and ftplib.FTP_TLS as a keyword-only parameter, and the default encoding is changed from Latin-1 to UTF-8 to follow RFC 2640.

  • asyncio.loop.shutdown_default_executor() has been added to AbstractEventLoop, meaning alternative event loops that inherit from it should have this method defined. (Contributed by Kyle Stanley in bpo-34037.)

  • The constant values of future flags in the __future__ module is updated in order to prevent collision with compiler flags. Previously PyCF_ALLOW_TOP_LEVEL_AWAIT was clashing with CO_FUTURE_DIVISION. (Contributed by Batuhan Taskaya in bpo-39562)

  • array('u') now uses wchar_t as C type instead of Py_UNICODE. This change doesn’t affect to its behavior because Py_UNICODE is alias of wchar_t since Python 3.3. (Contributed by Inada Naoki in bpo-34538.)

  • The logging.getLogger() API now returns the root logger when passed the name 'root', whereas previously it returned a non-root logger named 'root'. This could affect cases where user code explicitly wants a non-root logger named 'root', or instantiates a logger using logging.getLogger(__name__) in some top-level module called 'root.py'. (Contributed by Vinay Sajip in bpo-37742.)

  • Division handling of PurePath now returns NotImplemented instead of raising a TypeError when passed something other than an instance of str or PurePath. This allows creating compatible classes that don’t inherit from those mentioned types. (Contributed by Roger Aiudi in bpo-34775).

  • Starting with Python 3.9.5 the 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 notatation. glibc implementation of modern inet_pton() does not accept any leading zeros. (Contributed by Christian Heimes in bpo-36384).

  • codecs.lookup() now normalizes the encoding name the same way as encodings.normalize_encoding(), except that codecs.lookup() also converts the name to lower case. For example, "latex+latin1" encoding name is now normalized to "latex_latin1". (Contributed by Jordon Xu in bpo-37751.)

Changes in the C API

  • Instances of heap-allocated types (such as those created with PyType_FromSpec() and similar APIs) hold a reference to their type object since Python 3.8. As indicated in the “Changes in the C API” of Python 3.8, for the vast majority of cases, there should be no side effect but for types that have a custom tp_traverse function, ensure that all custom tp_traverse functions of heap-allocated types visit the object’s type.

    Example:

    int
foo_traverse(foo_struct *self, visitproc visit, void *arg) {
// Rest of the traverse function
#if PY_VERSION_HEX >= 0x03090000
    // This was not needed before Python 3.9 (Python issue 35810 and 40217)
    Py_VISIT(Py_TYPE(self));
#endif
}

    If your traverse function delegates to tp_traverse of its base class (or another type), ensure that Py_TYPE(self) is visited only once. Note that only heap type are expected to visit the type in tp_traverse.

    For example, if your tp_traverse function includes:

    base->tp_traverse(self, visit, arg)

    then add:

    #if PY_VERSION_HEX >= 0x03090000
    // This was not needed before Python 3.9 (bpo-35810 and bpo-40217)
    if (base->tp_flags & Py_TPFLAGS_HEAPTYPE) {
        // a heap type's tp_traverse already visited Py_TYPE(self)
    } else {
        Py_VISIT(Py_TYPE(self));
    }
#else

    (See bpo-35810 and bpo-40217 for more information.)

  • The functions PyEval_CallObject, PyEval_CallFunction, PyEval_CallMethod and PyEval_CallObjectWithKeywords are deprecated. Use PyObject_Call() and its variants instead. (See more details in bpo-29548.)

CPython bytecode changes

  • The LOAD_ASSERTION_ERROR opcode was added for handling the assert statement. Previously, the assert statement would not work correctly if the AssertionError exception was being shadowed. (Contributed by Zackery Spytz in bpo-34880.)

  • The COMPARE_OP opcode was split into four distinct instructions:

    • COMPARE_OP for rich comparisons

    • IS_OP for ‘is’ and ‘is not’ tests

    • CONTAINS_OP for ‘in’ and ‘not in’ tests

    • JUMP_IF_NOT_EXC_MATCH for checking exceptions in ‘try-except’ statements.

    (Contributed by Mark Shannon in bpo-39156.)

Build Changes

  • Added --with-platlibdir option to the configure script: name of the platform-specific library directory, stored in the new sys.platlibdir attribute. See sys.platlibdir attribute for more information. (Contributed by Jan Matějek, Matěj Cepl, Charalampos Stratakis and Victor Stinner in bpo-1294959.)

  • The COUNT_ALLOCS special build macro has been removed. (Contributed by Victor Stinner in bpo-39489.)

  • On non-Windows platforms, the setenv() and unsetenv() functions are now required to build Python. (Contributed by Victor Stinner in bpo-39395.)

  • On non-Windows platforms, creating bdist_wininst installers is now officially unsupported. (See bpo-10945 for more details.)

  • 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. (Contributed by Ned Deily in bpo-34956.)

  • Python can now be built for Windows 10 ARM64. (Contributed by Steve Dower in bpo-33125.)

  • Some individual tests are now skipped when --pgo is used. The tests in question increased the PGO task time significantly and likely didn’t help improve optimization of the final executable. This speeds up the task by a factor of about 15x. Running the full unit test suite is slow. This change may result in a slightly less optimized build since not as many code branches will be executed. If you are willing to wait for the much slower build, the old behavior can be restored using ./configure [..] PROFILE_TASK="-m test --pgo-extended". We make no guarantees as to which PGO task set produces a faster build. Users who care should run their own relevant benchmarks as results can depend on the environment, workload, and compiler tool chain. (See bpo-36044 and bpo-37707 for more details.)

C API Changes

New Features

Porting to Python 3.9

  • PyInterpreterState.eval_frame (PEP 523) now requires a new mandatory tstate parameter (PyThreadState*). (Contributed by Victor Stinner in bpo-38500.)

  • Extension modules: m_traverse, m_clear and m_free functions of PyModuleDef are no longer called if the module state was requested but is not allocated yet. This is the case immediately after the module is created and before the module is executed (Py_mod_exec function). More precisely, these functions are not called if m_size is greater than 0 and the module state (as returned by PyModule_GetState()) is NULL.

    Extension modules without module state (m_size <= 0) are not affected.

  • 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. (Contributed by Victor Stinner in bpo-39984.)

  • The Windows registry is no longer used to initialize sys.path when the -E option is used (if PyConfig.use_environment is set to 0). This is significant when embedding Python on Windows. (Contributed by Zackery Spytz in bpo-8901.)

  • The global variable PyStructSequence_UnnamedField is now a constant and refers to a constant string. (Contributed by Serhiy Storchaka in bpo-38650.)

  • The PyGC_Head structure is now opaque. It is only defined in the internal C API (pycore_gc.h). (Contributed by Victor Stinner in bpo-40241.)

  • The Py_UNICODE_COPY, Py_UNICODE_FILL, PyUnicode_WSTR_LENGTH, PyUnicode_FromUnicode(), PyUnicode_AsUnicode(), _PyUnicode_AsUnicode, and PyUnicode_AsUnicodeAndSize() are marked as deprecated in C. They have been deprecated by PEP 393 since Python 3.3. (Contributed by Inada Naoki in bpo-36346.)

  • The Py_FatalError() function is replaced with a macro which logs automatically the name of the current function, unless the Py_LIMITED_API macro is defined. (Contributed by Victor Stinner in bpo-39882.)

  • The vectorcall protocol now requires that the caller passes only strings as keyword names. (See bpo-37540 for more information.)

  • Implementation details of a number of macros and functions are now hidden:

    (See bpo-40170 for more details.)

Removed

  • Excluded PyFPE_START_PROTECT() and PyFPE_END_PROTECT() macros of pyfpe.h from the limited C API. (Contributed by Victor Stinner in bpo-38835.)

  • The tp_print slot of PyTypeObject has been removed. It was used for printing objects to files in Python 2.7 and before. Since Python 3.0, it has been ignored and unused. (Contributed by Jeroen Demeyer in bpo-36974.)

  • Changes in the limited C API (if Py_LIMITED_API macro is defined):

    • Excluded the following functions from the limited C API:

      • PyThreadState_DeleteCurrent() (Contributed by Joannah Nanjekye in bpo-37878.)

      • _Py_CheckRecursionLimit

      • _Py_NewReference()

      • _Py_ForgetReference()

      • _PyTraceMalloc_NewReference()

      • _Py_GetRefTotal()

      • The trashcan mechanism which never worked in the limited C API.

      • PyTrash_UNWIND_LEVEL

      • Py_TRASHCAN_BEGIN_CONDITION

      • Py_TRASHCAN_BEGIN

      • Py_TRASHCAN_END

      • Py_TRASHCAN_SAFE_BEGIN

      • Py_TRASHCAN_SAFE_END

    • Moved following functions and definitions to the internal C API:

      • _PyDebug_PrintTotalRefs()

      • _Py_PrintReferences()

      • _Py_PrintReferenceAddresses()

      • _Py_tracemalloc_config

      • _Py_AddToAllObjects() (specific to Py_TRACE_REFS build)

    (Contributed by Victor Stinner in bpo-38644 and bpo-39542.)

  • Removed _PyRuntime.getframe hook and removed _PyThreadState_GetFrame macro which was an alias to _PyRuntime.getframe. They were only exposed by the internal C API. Removed also PyThreadFrameGetter type. (Contributed by Victor Stinner in bpo-39946.)

  • Removed the following functions from the C API. Call PyGC_Collect() explicitly to clear all free lists. (Contributed by Inada Naoki and Victor Stinner in bpo-37340, bpo-38896 and bpo-40428.)

    • PyAsyncGen_ClearFreeLists()

    • PyContext_ClearFreeList()

    • PyDict_ClearFreeList()

    • PyFloat_ClearFreeList()

    • PyFrame_ClearFreeList()

    • PyList_ClearFreeList()

    • PyMethod_ClearFreeList() and PyCFunction_ClearFreeList(): the free lists of bound method objects have been removed.

    • PySet_ClearFreeList(): the set free list has been removed in Python 3.4.

    • PyTuple_ClearFreeList()

    • PyUnicode_ClearFreeList(): the Unicode free list has been removed in Python 3.3.

  • Removed _PyUnicode_ClearStaticStrings() function. (Contributed by Victor Stinner in bpo-39465.)

  • Removed Py_UNICODE_MATCH. It has been deprecated by PEP 393, and broken since Python 3.3. The PyUnicode_Tailmatch() function can be used instead. (Contributed by Inada Naoki in bpo-36346.)

  • Cleaned header files of interfaces defined but with no implementation. The public API symbols being removed are: _PyBytes_InsertThousandsGroupingLocale, _PyBytes_InsertThousandsGrouping, _Py_InitializeFromArgs, _Py_InitializeFromWideArgs, _PyFloat_Repr, _PyFloat_Digits, _PyFloat_DigitsInit, PyFrame_ExtendStack, _PyAIterWrapper_Type, PyNullImporter_Type, PyCmpWrapper_Type, PySortWrapper_Type, PyNoArgsFunction. (Contributed by Pablo Galindo Salgado in bpo-39372.)

Notable changes in Python 3.9.1

typing

The behavior of typing.Literal was changed to conform with PEP 586 and to match the behavior of static type checkers specified in the PEP.

  1. Literal now de-duplicates parameters.

  2. Equality comparisons between Literal objects are now order independent.

  3. Literal comparisons now respect types. For example, Literal[0] == Literal[False] previously evaluated to True. It is now False. To support this change, the internally used type cache now supports differentiating types.

  4. Literal objects will now raise a TypeError exception during equality comparisons if any of their parameters are not hashable. Note that declaring Literal with mutable parameters will not throw an error:

    >>> from typing import Literal
>>> Literal[{0}]
>>> Literal[{0}] == Literal[{False}]
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: unhashable type: 'set'

(Contributed by Yurii Karabas in bpo-42345.)

macOS 11.0 (Big Sur) and Apple Silicon Mac support

As of 3.9.1, Python now fully supports building and running on macOS 11.0 (Big Sur) and on Apple Silicon Macs (based on the ARM64 architecture). A new universal build variant, universal2, is now available to natively support both ARM64 and Intel 64 in one set of executables. Binaries can also now be built on current versions of macOS to be deployed on a range of older macOS versions (tested to 10.9) while making some newer OS functions and options conditionally available based on the operating system version in use at runtime (“weaklinking”).

(Contributed by Ronald Oussoren and Lawrence D’Anna in bpo-41100.)

Notable changes in Python 3.9.2

collections.abc

collections.abc.Callable generic now flattens type parameters, similar to what typing.Callable currently does. This means that collections.abc.Callable[[int, str], str] will have __args__ of (int, str, str); previously this was ([int, str], str). To allow this change, types.GenericAlias can now be subclassed, and a subclass will be returned when subscripting the collections.abc.Callable type. Code which accesses the arguments via typing.get_args() or __args__ need to account for this change. A DeprecationWarning may be emitted for invalid forms of parameterizing collections.abc.Callable which may have passed silently in Python 3.9.1. This DeprecationWarning will become a TypeError in Python 3.10. (Contributed by Ken Jin in bpo-42195.)

urllib.parse

Earlier Python versions allowed using both ; and & as query parameter separators in urllib.parse.parse_qs() and urllib.parse.parse_qsl(). Due to security concerns, and to conform with newer W3C recommendations, this has been changed to allow only a single separator key, with & as the default. This change also affects cgi.parse() and cgi.parse_multipart() as they use the affected functions internally. For more details, please see their respective documentation. (Contributed by Adam Goldschmidt, Senthil Kumaran and Ken Jin in bpo-42967.)