deps(python): bump the python group in /backend with 16 updates

Closed
DE
3d ago
+347-326

Bumps the python group in /backend with 16 updates:

PackageFromTo
anyio4.15.14.16.0
dnspython2.9.02.9.1
grpcio1.84.11.85.1
oauthlib4.0.04.0.1
pyasn10.6.40.6.5
pydantic2.14.12.14.2
python-dateutil3.0.04.0.0
python-multipart0.0.320.0.33
wrapt2.5.12.5.2
pytest9.1.19.2.0
pytest-asyncio1.4.11.4.1.post1
pytest-env1.8.11.8.4
mypy2.4.02.4.1
ruff0.17.20.17.3
grimp3.173.18
pipdeptree4.2.94.2.13

Updates anyio from 4.15.1 to 4.16.0

Release notes

Sourced from anyio's releases.

4.16.0

New features

  • Added BlockingPortal.call_async() for asynchronously calling into an event loop with optional abandonment on cancellation (#1342; PR by @​davidbrochart)
  • Added SOCK_SEQPACKET support to connect_unix() and create_unix_listener() (#1349; PR by @​theunkn0wn1)

Other changes

  • Changed UDP sockets on the asyncio backend to make send() wait until the datagram has been passed to the operating system (#1294; PR by @​graingert)

Bug fixes

  • Fixed documentation implying that use_uvloop=True falls back to the standard event loop when uvloop (winloop on Windows) is not installed, and showed how to skip the corresponding pytest backend variant when the dependency is missing (#645; PR by @​GruffElixir)
  • Fixed get_current_task() on the asyncio backend raising an opaque TypeError: cannot create weak reference to 'NoneType' object instead of a clear error when called from a callback scheduled via loop.call_soon_threadsafe() (e.g. from within from_thread.run_sync()), where asyncio.current_task() is legitimately None (#773; PR by @​AmirF194)
  • Fixed CancelScope on the asyncio backend swallowing a native Task.cancel() that was delivered in the same event loop iteration as the scope's own cancellation (such as a move_on_after() deadline expiring), which left the host task with a pending cancellation request that never raised CancelledError (#1214)
  • Fixed UDP socket closing hanging if a datagram send was still in flight (#1237; PR by @​graingert)
  • Fixed concurrent aclose_forcefully() calls on asyncio socket streams returning before the underlying socket was closed (#1273; PR by @​graingert)
  • Fixed aclose() on sockets not checkpointing, and thus failing to raise a cancellation exception when called inside a cancelled scope (affected UNIX socket streams and UNIX datagram sockets on asyncio, and all sockets on Trio) (#1288; PR by @​graingert)
  • Fixed aclose() on asyncio TCP socket listeners not checkpointing, and leaking the listening socket if cancelled while another task was waiting in accept() (#1289; PR by @​graingert)
  • Fixed accept() on asyncio UNIX socket listeners hanging forever, and leaving the listening socket registered with the event loop, when the listener was closed by another task (#1289; PR by @​graingert)
  • Fixed send() on asyncio TCP and UDP sockets returning normally, despite the data being discarded, when the socket was closed by another task (or, for TCP, the connection was lost) while send() was waiting for the operating system to accept the data (#1289; PR by @​graingert)
  • Fixed SocketStream.send() on the asyncio backend handing its data to a paused transport after a previous send() was cancelled (#1299; PR by @​graingert)
  • Fixed sleep() treating a negative delay (including -inf) inconsistently across backends (asyncio returned immediately; Trio raised ValueError) by raising ValueError on all backends (#1306; PR by @​BetterAndBetterII)
  • Fixed create_tcp_listener() occasionally failing with a message like Could not create 2 listeners with a consistent port when an ephemeral port is requested and IPv6 is enabled and the dual-stack path is not available or a specific local host name was given (#1313)
  • Fixed AsyncFile not shielding against cancellation while closing (#1314)
  • Fixed TemporaryDirectory.cleanup() reacting to cancellation prematurely when the host task was already cancelled, causing the actual cleanup operation to be skipped (#1316; PR by @​Yasser-Ameur)
  • Fixed a NaN deadline being accepted on asyncio while Trio raised ValueError, which silently disabled the timeout (fail_after(math.nan) returned instead of raising TimeoutError); CancelScope now rejects it on both backends (#1317; PR by @​dylanpulver)
  • Fixed FileReadStream and FileWriteStream leaving files open when cancelled during context manager exit, including unflushed writes (#1318; PR by @​Kuang-xianxin)
  • Fixed Condition.notify() and Condition.notify_all() failing when the underlying lock was acquired directly rather than through the condition (#1319; PR by @​GruffElixir)
  • Fixed importing AnyIO with loaders that do not provide __file__, such as PyOxidizer, raising TypeError instead of falling back to eager imports (#1322; PR by @​skulitom)
  • Fixed text receive streams silently discarding incomplete characters at end-of-file instead of applying the configured decoding error handler (#1339; PR by @​Aurelucien)
  • Fixed Semaphore(fast_acquire=True) losing its fast-acquire behavior when instantiated outside a running event loop (#1348; PR by @​feiiiiii5)
  • Fixed BlockingPortalProvider permanently failing with an AttributeError on every subsequent use if the first attempt to start its blocking portal failed (#1350; PR by @​Yasser-Ameur)
  • Fixed create_udp_socket() silently ignoring the local_port argument when local_host was omitted, making the socket bind to a random ephemeral port instead of the requested one (#1351; PR by @​januththedev)
  • Fixed worker threads on the asyncio backend never returning to the idle pool (and thus never being reused or pruned) when a to_thread.run_sync() call was cancelled after it had been queued for the worker but before the worker thread had picked it up (#1352; PR by @​Zac-HD)
  • Fixed pytest plugin running Hypothesis async tests only on the first parametrized backend instead of each backend (#1353; PR by @​DYNOSuprovo)
  • Fixed TypeError when subclassing an existing TypedAttributeSet subclass, and this subclass contains type annotations (#1358; PR by @​pasmud)
  • Fixed Lock.locked() (and Condition.locked()) raising NoEventLoopError when the lock was created outside a running event loop (#1362; PR by @​00200200)
  • Fixed connect_tcp() leaking a connected socket when an enclosing cancel scope was cancelled before the connection attempt task group exited (#1364; PR by @​charan-rathore)
  • Fixed create_connected_udp_socket() silently ignoring local_port when local_host was not given: the local address was left unset, so the OS picked an ephemeral port instead of the requested one (#1365; PR by @​feiiiiii5)
  • Fixed BlockingPortalProvider not passing the exception raised in its with block on to the start_blocking_portal() context manager it wraps (#1366; PR by @​feiiiiii5)
  • Fixed anyio.itertools.islice() not consuming the first start items when stop <= start (#1372; PR by @​Rodrigo-Palma)
  • Fixed BufferedByteReceiveStream.receive_exactly() and receive_until() returning buffered data after explicit closure instead of raising ClosedResourceError (#1378; PR by @​kokotatan)
  • Fixed a TLS stream sending records the peer could no longer decrypt after a cancelled send() lost an already-encrypted record; the stream now raises BrokenResourceError on further use instead (#1385; PR by @​Kayvan-Zahiri)
  • Fixed close() on memory object streams not waking up tasks blocked in send() or receive() on the same stream (#1391)
  • Fixed anyio.itertools.product() acquiring and consuming unused input iterables when repeat=0 (#1394; PR by @​sricharanreddycheruku)
  • Fixed SpooledTemporaryFile rolling over to disk on writes when max_size is zero (the default) (#1397; PR by @​fhgffy)
  • Fixed Trio UDP sockets remaining open when initialization failed or was cancelled (#1400; PR by @​adenzhou1350)
  • Fixed SpooledTemporaryFile.rollover() moving the current file position to the end of the file (#1401; PR by @​fhgffy)
  • Fixed cancellation during temporary directory creation leaving the directory behind while its context manager is retained (#1403; PR by @​adenzhou1350)

... (truncated)

Commits
  • 0ddd02d Released version 4.16.0
  • 576892e Fixed certain pending tasks not woken up by the closing of memory object stre...
  • 02ff4f8 Added BlockingPortal.call_async() (#1343)
  • 1af29a4 Moved documentation code snippets into separate, checked files (#1445)
  • f01c041 [pre-commit.ci] pre-commit autoupdate (#1448)
  • 95a0143 Test with blockbuster and uvloop on Python 3.15 too
  • 4d51154 Modernized the memory object stream tests
  • 1e4b448 Fixed the as_connectable() example in the networking docs (#1442)
  • cd21b36 Fixed TypedAttributeSet not collecting annotations from superclasses (#1358)
  • b7f8c68 [pre-commit.ci] pre-commit autoupdate (#1431)
  • Additional commits viewable in compare view

Updates dnspython from 2.9.0 to 2.9.1

Release notes

Sourced from dnspython's releases.

dnspython 2.9.1

See What's New for details.

The minimum supported version of Python is 3.11.

My thanks to the many people who have contributed to this release. Also thanks to my co-maintainers: Nicki Křížek, Petr Špaček, and Brian Wellington.

This release was published to PyPI using Trusted Publishing.

Changelog

Sourced from dnspython's changelog.

2.9.1

  • LOC text parsing now converts the altitude, size, horizontal precision, and vertical precision fields to centimeters exactly, and requires them to be decimal numbers of meters with at most two fractional digits, as RFC 1876 section 3 specifies. They were converted with float(), so 9.29m became 928 cm rather than 929 cm and the text and wire forms of the record disagreed, forms such as 1e3m, 1_000m, +10m, nanm, and infm were accepted, extra fractional digits were silently dropped, and an altitude outside the representable range was accepted and then failed in to_wire() with a non-DNS exception. dns.exception.SyntaxError is now raised for all of these.

  • The resolver now assigns a new random query id to every query.

  • For transports that require a 0 query id, set it in a way that does not break TSIG.

  • dns.message.QueryMessage.resolve_chaining() now caps the minimum TTL of a negative response with no SOA record at 300 seconds (dns.message.NEGATIVE_TTL_WITHOUT_SOA). Previously the TTL was left at its initial value of 2^32 - 1, so a single NXDOMAIN or NODATA response with no SOA, e.g. a spoofed one, stayed in the resolver cache for the life of the process.

  • DNSSEC validation now rejects an RRSIG whose type covered or class does not match the RRset, as RFC 4035 section 5.3.1 requires. This was not checked, so a revoked DNSKEY, which may only validate the DNSKEY RRset, still validated any RRset signed with its private key if the RRSIG claimed to cover DNSKEY.

  • dns.tokenizer.Tokenizer now ends a line at a carriage return, or at a carriage return followed by a line feed, unless the carriage return is in a quoted string. The carriage return of CRLF line ends used to stay in the last token of each line and in the text of a comment, so e.g. dns.zone.from_text() rejected a zone with CRLF line ends.

  • DNSSEC validation no longer uses a DNSKEY with the REVOKE flag set to validate anything other than the DNSKEY RRset. RFC 5011 section 2.1 says a revoked key must not be used for any other purpose, but it was a candidate key like any other, so as long as the revoked key was still published, data signed with its private key, e.g. by whoever compromised it, validated.

  • dns.ttl.from_text() now rejects a BIND-style TTL that ends in a 0 without a unit, e.g. 1h0, as it already did for other numbers, e.g. 1h5, and as BIND does. The check for a trailing number tested its value instead of whether there was one, so 1h0 was read as one hour.

  • dns.dnssec.sign_zone() now adds the apex NSEC record when the origin is

... (truncated)

Commits
  • dddd69c black lint
  • 185cc38 release prep
  • cf22838 Randomize the id of every message the resolver sends. (#1395)
  • e0fce7f use a short negative TTL for a negative answer with no SOA in resolve_chainin...
  • ad0803c check the RRSIG type covered and class against the RRset (#1388)
  • 8c38a04 Strictly apply RFC 3597 compression rules when decompressing. (#1383)
  • ab5ade8 end a line at a carriage return in the tokenizer (#1387)
  • d9d259d only use a revoked DNSKEY to validate the DNSKEY RRset (#1385)
  • 6d436b8 reject BIND-style TTLs that end in a 0 without a unit (#1384)
  • d8fcf2d bound $GENERATE width and parse text labels in linear time (#1382)
  • Additional commits viewable in compare view

Updates grpcio from 1.84.1 to 1.85.1

Release notes

Sourced from grpcio's releases.

Release v1.85.1

This is release gRPC Core 1.85.1 (googol).

For gRPC documentation, see grpc.io. For previous releases, see Releases.

This release contains refinements, improvements, and bug fixes.

Python


  • Fixes Denial of service in Health Watch RPC (#43783)
  • Fixes Memory leak in AsyncIO client-streaming calls with interceptors (#43783)

Core


  • host/:authority handling in the HPACK parser (#43797)

Release v1.85.0

This is release 1.85.0 (googol) of gRPC Core.

For gRPC documentation, see grpc.io. For previous releases, see Releases.

This release contains refinements, improvements, and bug fixes, with highlights listed below.

Core

  • [security] Update gRPC repo roots. (#43717)
  • [experiments] Add map_host_header_to_authority experiment. (#43561)
  • [Security] Add MLKEM1024 as a key exchange option. (#43514)
  • Remove OpenSSL1.0.2 Support. (#43466)
  • [Autosharding-LB][Part-1] Implementing A119. (#43406)
  • [OpenSSL] Add support for OpenSSL 4. (#41932)
  • [A110] Add child channel options. (#43314)

C#

  • [Build] Disambiguate protobuf_version in templates. (#43433)

Python

  • [Python][AIO] Fixed data race in PollerCompletionQueue. (#43429)
  • [Python] Fix SyncServicerContext abort not raising exception. (#43126)
  • [Python] Fixed channelz module leak. (#43196)
  • [Python] Fixed CSDS memory leak. (#43202)
  • [Python] Unpin Cython dependency for source builds. (#43378)

... (truncated)

Commits
  • 558b7e3 [Backport][v1.85.x][export] host/:authority handling in the HPACK parser (#4...
  • 54b5cf6 [Release] Bump version to 1.85.1 (on v1.85.x branch) (#43788)
  • fc9be67 [Backport][v1.85.x][export] Internal change (#43783)
  • a83eb27 update g stands for correctly
  • cc92f57 [Release] Bump version to 1.85.0 (on v1.85.x branch) (#43729)
  • 7f3e5d4 [Backport][v1.85.x][security] Update gRPC repo roots. (#43717)
  • 08689a5 [Backport][v1.85.x][CI][macOS] Add retries to git and pip for flaky macOS job...
  • cc11ee7 [Backport][v1.85.x][export] Validate EDS locality priority against number of ...
  • 0e02b12 [Release] Bump version to 1.85.0-pre1 (on v1.85.x branch) (#43591)
  • e124571 [Release] Bump core version to 58.0.0 for upcoming release (#43582)
  • Additional commits viewable in compare view

Updates oauthlib from 4.0.0 to 4.0.1

Release notes

Sourced from oauthlib's releases.

4.0.1

Security

  • CVE-2026-49272 / GHSA-3jp2-3mrg-2vxc: The device code flow token endpoint did not validate the device_code parameter. A request with a known client_id and a bogus device_code could obtain a bearer token. DeviceCodeGrant.validate_token_request now calls DeviceCodeGrant.validate_device_code(), which rejects missing or unknown device codes (invalid_request / invalid_grant) and delegates the code's authorization status to the new RequestValidator.validate_device_code() hook, per RFC 8628 Section 3.5. (cherry-picked from #954)

Upgrade note for device flow integrators

Custom RequestValidator implementations using the device flow must now implement validate_device_code(client_id, code, request). Return one of the DeviceCodeGrant.DEVICE_CODE_* statuses for known codes, or None for an unknown code. When returning AUTHORIZED, the validator must also populate request.user and request.scopes from the stored authorization.

Changelog

Sourced from oauthlib's changelog.

4.0.1 (2026-11-04):

OAuth2.0 Provider (Security):

  • Security: CVE-2026-49272 / GHSA-3jp2-3mrg-2vxc: The device code flow token endpoint did not validate the device_code parameter. A request with a known client_id and a bogus device_code could obtain a bearer token. DeviceCodeGrant.validate_token_request now calls a new DeviceCodeGrant.validate_device_code() which rejects missing or unknown device codes (400 invalid_request / 400 invalid_grant) and delegates the code's authorization status to a new required RequestValidator.validate_device_code() hook, per RFC 8628 Section 3.5. (cherry-picked from #954, merged as 1c91d64)

    Upgrade note for device flow integrators: custom RequestValidator implementations using the device flow must now implement validate_device_code(client_id, code, request). Return one of the DeviceCodeGrant.DEVICE_CODE_* statuses (AUTHORIZED, PENDING, SLOW_DOWN or EXPIRED) for known codes, or None for an unknown code. When returning AUTHORIZED, the validator must also populate request.user and request.scopes from the stored authorization.

Commits
  • 6e3aa05 Release 4.0.1: fix CVE-2026-49272 (device_code validation in device flow)
  • 77a5f1c Merge pull request #954 from hekhuisk/improve-device-workflow
  • See full diff in compare view

Updates pyasn1 from 0.6.4 to 0.6.5

Release notes

Sourced from pyasn1's releases.

Release 0.6.5

Revision 0.6.5, released 26-10-2026

  • GHSA-mm57-4v3p-rchx: Decoded binary REAL mantissas in linear time instead of repeatedly shifting a growing integer and slicing the remaining bytes (thanks for reporting, manus-use, manus-pi, ucsurf, adriank31)
  • GHSA-2qm8-x737-gmw5: Reassembled constructed BIT STRING fragments in linear time while preserving their decoded bit values, including intermediate fragment padding (thanks for reporting, manus-use)
  • GHSA-9m75-vc3g-4x39: Reassembled constructed OCTET STRING and character-string fragments and indefinite-length ANY content in linear time. Nested constructed OCTET STRING fragments now contribute their decoded octets (thanks for reporting, alimony, manus-pi, adriank31)
  • GHSA-fm9c-5rcf-2q5p: Bounded schemaless container type tracking after two distinct tag sets to prevent quadratic CPU use from colliding long tag IDs (thanks for reporting, droideck)
Changelog

Sourced from pyasn1's changelog.

Revision 0.6.5, released 26-10-2026

  • GHSA-mm57-4v3p-rchx: Decoded binary REAL mantissas in linear time instead of repeatedly shifting a growing integer and slicing the remaining bytes (thanks for reporting, manus-use, manus-pi, ucsurf, adriank31)
  • GHSA-2qm8-x737-gmw5: Reassembled constructed BIT STRING fragments in linear time while preserving their decoded bit values, including intermediate fragment padding (thanks for reporting, manus-use)
  • GHSA-9m75-vc3g-4x39: Reassembled constructed OCTET STRING and character-string fragments and indefinite-length ANY content in linear time. Nested constructed OCTET STRING fragments now contribute their decoded octets (thanks for reporting, alimony, manus-pi, adriank31)
  • GHSA-fm9c-5rcf-2q5p: Bounded schemaless container type tracking after two distinct tag sets to prevent quadratic CPU use from colliding long tag IDs (thanks for reporting, droideck)
Commits
  • 5d16736 Prepare release 0.6.5
  • a36e3a3 Fix quadratic CPU use in schemaless container decoding
  • 4ce0079 Remove quadratic constructed OCTET STRING and ANY reassembly
  • 73ba7b0 Remove quadratic constructed BIT STRING reassembly
  • ce3ec33 Remove quadratic binary REAL mantissa decoding
  • See full diff in compare view

Updates pydantic from 2.14.1 to 2.14.2

Release notes

Sourced from pydantic's releases.

v2.14.2 2026-10-29

v2.14.2 (2026-10-29)

This release addresses 2 security issues, https://github.com/pydantic/pydantic/security/advisories/GHSA-qpmf-fgh3-ff25 and https://github.com/pydantic/pydantic/security/advisories/GHSA-3r8g-cqwr-jr29.

Packaging

Performance

Fixes

  • Make JSON validation by iteration consistent with Python approach by @​davidhewitt in #13275.

    JSON validation now handles field lookup and extra data the same way as Python validation. In particular, keys that only match an unused alias or field name are now treated as extra data, instead of being silently dropped:

    from pydantic import BaseModel, ConfigDict
    from pydantic.alias_generators import to_camel
    

    class Page(BaseModel):
    model_config = ConfigDict(extra='forbid', alias_generator=to_camel)

    page_size: int = 10
    

    Page.model_validate_json('{"page_size": 50}')

    Before: Page(page_size=10)

    After: ValidationError (extra_forbidden)

    See the pull request for the full list of behavior changes. This change addresses https://github.com/pydantic/pydantic/security/advisories/GHSA-qpmf-fgh3-ff25.

  • Ignore extra data matching a field name in TypedDict and dataclasses with extra=allow by @​Viicos in #14090. Addresses https://github.com/pydantic/pydantic/security/advisories/GHSA-3r8g-cqwr-jr29.

  • Fix unpickling of generic classes defined in separate module by @​Viicos in #13994

  • Support explicit by_alias overriding validate_by_alias in validation JSON schemas by @​Viicos in #13997

  • Fix mypy plugin crash on deferred empty model by @​ilevkivskyi in #14027

  • Fix infinite recursion when validating an invalid default with PydanticUseDefault by @​Viicos in #14028

  • Implement recursion guard for validate_strings() by @​Viicos in #14029

... (truncated)

Changelog

Sourced from pydantic's changelog.

v2.14.2 (2026-10-29)

GitHub release

What's Changed

Packaging

Performance

Fixes

  • Make JSON validation by iteration consistent with Python approach by @​davidhewitt in #13275.

    JSON validation now handles field lookup and extra data the same way as Python validation. In particular, keys that only match an unused alias or field name are now treated as extra data, instead of being silently dropped:

    from pydantic import BaseModel, ConfigDict
    from pydantic.alias_generators import to_camel
    

    class Page(BaseModel):
    model_config = ConfigDict(extra='forbid', alias_generator=to_camel)

    page_size: int = 10
    

    Page.model_validate_json('{"page_size": 50}')

    Before: Page(page_size=10)

    After: ValidationError (extra_forbidden)

    See the pull request for the full list of behavior changes.

  • Fix unpickling of generic classes defined in separate module by @​Viicos in #13994

  • Support explicit by_alias overriding validate_by_alias in validation JSON schemas by @​Viicos in #13997

  • Fix mypy plugin crash on deferred empty model by @​ilevkivskyi in #14027

  • Fix infinite recursion when validating an invalid default with PydanticUseDefault by @​Viicos in #14028

  • Implement recursion guard for validate_strings() by @​Viicos in #14029

  • Support named tuples in Sequence serialization by @​Viicos in #14033

... (truncated)

Commits
  • aa540ac Update changelog for v2.14.2
  • 6cde12f Update changelog for v2.14.2
  • 2db7c25 Fix deep copy of self-referential models
  • d783154 Update changelog for v2.14.2
  • df49b6c Avoid heap allocation for small JSON model field buffers
  • cd93e69 Skip decorator field checks when no field decorators exist
  • 2771e57 Make JSON validation by iteration consistent with Python approach
  • 8102e3f Avoid Python tuple allocation during model fields validation
  • dd139b5 Bump pydantic.v1 namespace to v1.10.27
  • 66155f9 Prepare release v2.14.2
  • Additional commits viewable in compare view

Updates python-dateutil from 3.0.0 to 4.0.0

Release notes

Sourced from python-dateutil's releases.

4.0.0

dateutil 4.0 requires Python 3.11 or later and no longer depends on six. Apart from that, the changes in this release are bug fixes, but note that isoparse is now stricter about the numeric fields it accepts (see below).

Deprecations

  • Dropped support for Python versions older than 3.11 and removed the dependency on six, along with Python 2 compatibility helpers that were never part of the public API, such as dateutil.tz.tzname_in_python2. (gh pr #1625)

Bugfixes

  • parser.isoparse, and the isoparser methods parse_isodate, parse_isotime and parse_tzstr, now require each numeric field to be exactly the expected number of ASCII digits. Strings that were previously accepted, such as 20200728121240 (read as 21:24), 2014- 4-21 and 2_14-04-21, and offsets such as +-1:00, now raise ValueError. tzical applies the same check to TZOFFSETFROM and TZOFFSETTO. Reported by @​janos-gonye (gh issue #1074). Fixed by @​varun3257 (gh pr #1617).
  • Fixed an issue where iterating over a cached rrule or rruleset could hang forever if another iterator over the same rule had already completed the cache, for example when a cached rrule was included twice in an rruleset. Reported by @​bysiber (gh pr #1455). Fixed by @​pganssle (gh pr #1655).
  • Unpickling a tzutc now gives the dateutil.tz.UTC singleton, and unpickling a tzoffset gives the same object as calling tzoffset(name, offset). Previously unpickling bypassed the singleton-generation logic. Reported and fixed by @​masipcat (gh issue #1051, gh pr #1053).
  • relativedelta(yearday=366) now resolves to December 31 in leap years. Previously it gave December 30, the same date as yearday=365. Reported and fixed by @​Master-Norna (gh pr #1583).

Misc

  • The build backend is now Hatchling, and the package metadata is in pyproject.toml. Versions are still derived from Git tags and exposed as dateutil.__version__. This mainly affects people building dateutil from source. (gh pr #1637, #1638)
  • Changes to CI, tests and release tooling that do not affect users of the package: gh pr #1636, #1644, #1645, #1646, #1648, #1649, #1650.
Changelog

Sourced from python-dateutil's changelog.

Version 4.0.0 (2026-11-01)

dateutil 4.0 requires Python 3.11 or later and no longer depends on six. Apart from that, the changes in this release are bug fixes, but note that isoparse is now stricter about the numeric fields it accepts (see below).

Deprecations

  • Dropped support for Python versions older than 3.11 and removed the dependency on six, along with Python 2 compatibility helpers that were never part of the public API, such as dateutil.tz.tzname_in_python2. (gh pr #1625)

Bugfixes

  • parser.isoparse, and the isoparser methods parse_isodate, parse_isotime and parse_tzstr, now require each numeric field to be exactly the expected number of ASCII digits. Strings that were previously accepted, such as 20200728121240 (read as 21:24), 2014- 4-21 and 2_14-04-21, and offsets such as +-1:00, now raise ValueError. tzical applies the same check to TZOFFSETFROM and TZOFFSETTO. Reported by @​janos-gonye (gh issue #1074). Fixed by @​varun3257 (gh pr #1617).
  • Fixed an issue where iterating over a cached rrule or rruleset could hang forever if another iterator over the same rule had already completed the cache, for example when a cached rrule was included twice in an rruleset. Reported by @​bysiber (gh pr #1455). Fixed by @​pganssle (gh pr #1655).
  • Unpickling a tzutc now gives the dateutil.tz.UTC singleton, and unpickling a tzoffset gives the same object as calling tzoffset(name, offset). Previously unpickling bypassed the singleton-generation logic. Reported and fixed by @​masipcat (gh issue #1051, gh pr #1053).
  • relativedelta(yearday=366) now resolves to December 31 in leap years. Previously it gave December 30, the same date as yearday=365. Reported and fixed by @​Master-Norna (gh pr #1583).

Misc

  • The build backend is now Hatchling, and the package metadata is in pyproject.toml. Versions are still derived from Git tags and exposed as dateutil.__version__. This mainly affects people building dateutil from source. (gh pr #1637, #1638)
  • Changes to CI, tests and release tooling that do not affect users of the

... (truncated)

Commits
  • a78cbd8 Prepare release 4.0.0
  • d23ff76 Fix yearday=366 resolving to Dec 30 in leap years
  • 5391455 Return the cached instances when unpickling tzutc and tzoffset
  • 6f13622 Reject non-digit characters in isoparse and tzical numeric fields
  • 6e1864c Release the rrule cache lock on every exit from a refill (#1655)
  • 71d92da Skip coverage collection on PyPy in CI
  • d9cad2f Run the tests in parallel with pytest-xdist (#1649)
  • 433c98b Add news fragment for #1650
  • ea6bc3f Remove stale warning from the tz tox environments
  • 27ecba2 Keep GitHub Actions pins updated with Dependabot
  • Additional commits viewable in compare view

Updates python-multipart from 0.0.32 to 0.0.33

Release notes

Sourced from python-multipart's releases.

Version 0.0.33

🛡️ Safer form parsing

This release lets you cap URL-encoded form fields as they stream in, and fixes several upload and decoding edge cases. It also drops Python 3.10 and adds Python 3.15.

pip install python-multipart==0.0.33
  • Limit URL-encoded fields while streaming (#333). The new opt-in MAX_FIELDS and MAX_FIELD_SIZE options for FormParser raise FormParserError before oversized data is buffered. Both default to no limit, and multipart forms are unaffected.
  • Nameless uploads no longer crash when spilled to disk (#343). Uploads without a filename used to raise AttributeError when UPLOAD_KEEP_EXTENSIONS or UPLOAD_KEEP_FILENAME was set.
  • UPLOAD_DIR accepts str with UPLOAD_KEEP_FILENAME (#335). A str upload directory used to fail on the first write to disk.
  • Decoded data is complete before callbacks run (#336, #355). Content-Transfer-Encoding decoders are finalized at the end of each part, and quoted-printable escapes split across writes decode correctly.
  • parse_options_header lowercases content types with parameters (#287). Multipart/Form-Data; boundary=abc now returns multipart/form-data, the same as headers without parameters.

⚡ Faster parsing

  • Faster opening boundary matching (#328). Complete boundaries are matched in one step.
  • Faster parser state lookups (#330). Parser states are module-level enum members, so their repr is now multipart.END instead of <MultipartState.END: 12>.
  • Faster querystring callback dispatch (#316).

🐍 Python support

  • Python 3.10 is no longer supported (#329). Python 3.11 is now the minimum.
  • Python 3.15 is supported (#350).

💜 Thank you

Thanks to @​Edneam, @​sungbin1015, @​carlok, and @​fjvbn2003 for contributing to this release.

Thanks to the individual sponsors who contributed $134.00 to support this work. If python-multipart is useful to you, consider sponsoring me on GitHub.

Full changelog: 0.0.32...0.0.33

Changelog

Sourced from python-multipart's changelog.

0.0.33 (2026-10-31)

  • Drop Python 3.10 support #329.
  • Add Python 3.15 support #350.
  • Add opt-in field-count and field-size limits for URL-encoded forms in FormParser #333.
  • Normalize the content type returned by parse_options_header when the header has parameters #287.
  • Finalize Content-Transfer-Encoding decoding before invoking field and file callbacks #336.
  • Keep only the incomplete escape when quoted-printable data is split across writes #355.
  • Handle nameless uploads when UPLOAD_KEEP_FILENAME or UPLOAD_KEEP_EXTENSIONS is set #343.
  • Accept a str UPLOAD_DIR when UPLOAD_KEEP_FILENAME is set #335.
  • Speed up parser state lookups #330.
  • Speed up opening boundary matching #328.
  • Speed up querystring callback dispatch #316.
Commits
  • 9217f26 Bump version to 0.0.33 (#357)
  • ...

    Description has been truncated


    Summary by cubic

    Bumps the Python dependency group in the backend, covering 16 packages across production, test, lint, and dev groups, including several security fixes and two major-version updates.

    Notable changes

    • oauthlib 4.0.1, pydantic 2.14.2, and pyasn1 0.6.5 fix security advisories (device code flow validation, JSON validation, and DoS issues).
    • pydantic JSON validation now treats keys matching unused aliases as extra data and raises with extra='forbid'.
    • python-dateutil 4.0.0 requires Python 3.11+, drops the six dependency, and makes isoparse stricter about numeric fields.
    • python-multipart 0.0.33 drops Python 3.10 support and adds opt-in field count and size limits for URL-encoded forms.