Release 0.930 planning

[JukkaL]

I'm planning another release fairly soon after 0.920 (#11158) is out to catch up with the GitHub master branch. If everything goes as planned, I'll cut the release branch around Tue 14 Dec. Please post here any PRs that you'd like to include in the release and that are not yet in master (or in the release branch after I've created it).

Note: Please do not sync typeshed until I've created the release branch. There are already some typeshed regressions that likely need to be resolved. If you want to include some specific typeshed changes, I can potentially cherry-pick them.

[JukkaL]

Regression that is a release blocker: #11699 (_order_ in enum)

[JukkaL]

Another regression: #11701 (Literal[True, False] false positives)

[JukkaL]

Another regression that might be a release blocker (but it might be very rare and not super important to fix): #11702 (enum inheritance false positive)

[JukkaL]

Yet another regression that looks like a release blocker: #11704 (need type annotation with contextmanager)

[JukkaL]

One more regression: #11705 ("and" type inference with strict optional disabled)

[hauntsaninja]

Currently unmerged PRs I'd like included:
#11681 (low risk, fixes a crash)
#11708 (low risk, removes an optimisation that resulted in a regression reported multiple times)
#11709 (fixes #11701 as mentioned above)
If possible, enough of #10191 that we no longer crash on match statements

Curious if @sobolevn has thoughts on any PRs to prioritise

[sobolevn]

Direct blockers:

• Enums with annotations and no values are fine to be subclassed #11579
• Special cases Enum and some special props to be non-final, refs #11699 #11713

Also a couple of bug fixes that can potentially be merged:

• Now overloads with ambiguous self are handled properly #11366
• Fixes type inference for generic calls in if expr #11128
• Adds __hash__ method to @dataclasses.dataclass, refs #11463 #11496
• Now Type[None].__bool__ is a thing, refs #11539 #11551 and Now type(None) is more consistent, refs #11550 #11552

[Zac-HD]

#11718 keeps biting me, and should be pretty easy to merge.

[JukkaL]

Two planned reverts to avoid regressions: #11741 (Python 2 division), #11743 (Python 2 raise), unless there's a forward fix very soon.

Some cherry-picks from typeshed to avoid regressions: #11739

[sobolevn]

unless there's a forward fix very soon.

Sorry, I won't be able to submit a fix before the weekend. If this is fine 🙂
Or you can always keep this out of scope of the release, these two feature are not very important.

[JukkaL]

Sorry, I won't be able to submit a fix before the weekend

No worries, I reverted the PRs now so that they won't block the release.

[cdce8p]

Don't know if there is still some time left but I believe #10712 would be a good feature to have.

[hauntsaninja]

I'd like this typeshed PR to be included: python/typeshed#6591 (fixes a regression shipped with 0.920)

[JukkaL]

I've created a release branch for 0.930 based on 22ab84a. If everything goes well, the release would be out Wed 22 Dec. If there is a delay, we'll probably have to postpone until January because of the holiday season.

[hauntsaninja]

Hooray, here's a PR for the aforementioned typeshed fix: #11775

[JelleZijlstra]

I'd suggest including python/typeshed#6620 and the corresponding mypy fix #11781, which fix the crash in #11768.

[hauntsaninja]

PR to the release branch for curses issues: #11785

[JukkaL]

Remaining issues / PRs that still need some updates:

• pre-release v0.920 incorrectly identifies enum as final #11578
  • PR: Enums with annotations and no values are fine to be subclassed #11579
• Type inference regression with "and" and strict optional disabled #11705
  • PR: Stricter None handling with --no-strict-optional, refs #11705 #11717

[JelleZijlstra]

We should also address #11774.

[AlexWaygood]

If it's not too late, I'd like to make a plug for python/typeshed#6634 to be cherry-picked. It doesn't fix a regression, but it does fix a bug with the existing stub that was reported as an issue to typeshed, creating usability issues with contextlib in 3.10.

[sobolevn]

Two docs only changes:

• TypeAlias: Documents explicit type aliases #11800
• Enum: Adds docs about Enum type #11805

[JukkaL]

Cherry-picked these PRs:

• Various small documentation updates #11817
• Allow calling a function with name '_' #11810
• Documents explicit type aliases #11800.

[JukkaL]

The release is now out: http://mypy-lang.blogspot.com/2021/12/mypy-0930-released.html

It includes this regression, but it seems minor enough to not be big deal:

• Type inference regression with "and" and strict optional disabled #11705
  • PR: Stricter None handling with --no-strict-optional, refs #11705 #11717

If/when a fix is ready, I can make a point release soon afterwards (unless we are close to the next release).

[JukkaL]

I'm keeping this issue open for a while in case there are significant regressions that should be fixed in a bugfix release.

[sobolevn]

@JukkaL great work! Thank you! 👍
Congrats on the new release.

I hope to finish #11717 on this weekend.

[wrobell]

@JukkaL #11827 seems to be a regression.

[AlexWaygood]

python/typeshed#6662 fixes a typeshed regression in 0.930.

[pranavrajpal]

I think all occurrences of ** are being swallowed by whatever is converting the markdown to html on the blog post. For example, in the example for ParamSpec support:

from typing import Awaitable, Callable, ParamSpec, TypeVar

P = ParamSpec("P")
R = TypeVar("R")

def add_logging(f: Callable[P, R]) -> Callable[P, Awaitable[R]]:
    async def inner(*args: P.args, kwargs: P.kwargs) -> R:
        await log_to_database()
        return f(*args, kwargs)
    return inner

@add_logging
def takes_int_str(x: int, y: str) -> int:
    return x + 7

await takes_int_str(1, "A")  # Accepted
await takes_int_str("B", 2)  # Rejected

I think both instances of kwargs should be **kwargs.

I think something similar happened in the 0.920 release blog post. #11137 and #11151 both have a ** in their title that seems to disappear when they're listed in the blog post:

Fix crash on dataclasses.field(unpack) (Nikita Sobolev, PR 11137)

Properly type check *CustomType and CustomType function arguments (Nikita Sobolev, PR 11151)

[pranavrajpal]

I think mypyc is completely broken on 0.930 (see mypyc/mypyc#912 for the error message).

That appears to be caused by #11494. That moved scripts/mypyc to mypyc/__main__.py, but it didn't add it to the mypyc blacklist like we do for mypy/__main__.py.

It's kind of weird that none of the CI caught this. I would have thought that the mypyc command line tests would have caused the same error when running with mypy and mypyc compiled with mypyc.

[sobolevn]

My PRs to fix recent regressions:

• Fixes __slots__ regression in 0.930, refs #11821 #11824
• Fixes regression with __module__ and similar non-final Enum names #11823
• NamedTuple now narrows to bool correctly, when __bool__ is defined #11822

Two reverted features from 0.930:

• Use __truediv__ for python2 with __future__ import, refs #11740 #11787
• Handle raise Exception(), None on Python2.7 #11786

[hauntsaninja]

Catching up on issues, here's my summary. We'll need to release 0.931 for at least the mypyc issue alone.
Feel free to edit / let me know if I've missed something and I'll keep this comment up to date.

Critical

• mypyc seems broken (from move mypyc to console_scripts #11494 )

  • Issue:
    KeyError: '__file__'  mypyc/mypyc#912
  • PRs:
    Skip compiling mypyc/__main__.py #11862
    Run mypyc tests against built wheels #11865
    Skip lxml and mypyc external tests against built wheels #11872
    Various fixes in build_wheel #11873
    Avoid grep on Windows wheel build #11879
    (only the first PR is the fix, the rest are trying to get CI to catch this issue going forward)

• exclude regression (from Support --exclude more than once on command line #11329 )

  • Issues:
    Crash/Regression - 0.930 no longer accepts multiline basic strings for exclude patterns in pyproject.toml #11825
    handling of exclude configuration changed in an incompatible way in 0.930 #11830
  • PR:
    Preserves interpreting multi-line strings as a single value for exclude in TOML files #11828
    (PR only fixes pyproject.toml)

Normal

• Fixes __slots__ regression in 0.930, refs #11821 #11824
• Fixes regression with __module__ and similar non-final Enum names #11823
• Fix regression when a contextmanager yields a generic #11870

Nice to have

• [0.931 backport] Fix __reduce__ regression #11866
• Use __truediv__ for python2 with __future__ import, refs #11740 #11787
• Handle raise Exception(), None on Python2.7 #11786
• Fixing ** markdown in blogpost
• Stricter None handling with --no-strict-optional, refs #11705 #11717 (if it's ready)
• NamedTuple now narrows to bool correctly, when __bool__ is defined #11822 (this one isn't a regression)

[JukkaL]

I'm planning to publish a 0.931 bugfix release later this week (current estimate is Wed or Thu). Thank you everybody who contributed fixes to regressions, and thank you @hauntsaninja for the useful summary!

In addition to the PRs mentioned above, this also fixes a regression:

• Revert to treating exclude in .ini as single string #11881

[JukkaL]

Here are the changes I've cherry-picked to the release-0.930 branch:

• Skip compiling mypyc/__main__.py #11862
• Run mypyc tests against built wheels #11865
• Skip lxml and mypyc external tests against built wheels #11872
• Various fixes in build_wheel #11873
• Avoid grep on Windows wheel build #11879
• Preserves interpreting multi-line strings as a single value for exclude in TOML files #11828
• Revert to treating exclude in .ini as single string #11881
• Fixes __slots__ regression in 0.930, refs #11821 #11824
• Fixes regression with __module__ and similar non-final Enum names #11823
• Fix regression when a contextmanager yields a generic #11870
• [0.931 backport] Fix __reduce__ regression #11866

I'm planning to make the 0.931 release with these changes tomorrow, unless something new comes up.

Additional things that might be worth fixing (haven't looked into these in much detail):

• regression: type alias for NoReturn no longer works as a type annotation #11903
• crash with sqlalchemy2 stubs #11902
• version 0.930 regression: INTERNAL ERROR AttributeError: attribute 'arguments' of 'FuncDef' undefined #11899
• Implicit type aliases not recognised with PEP 604 + import cycles #11887

[A5rocks]

#11913 is also a regression it seems.

[hauntsaninja]

#11912 fixes a minor regression in 0.930 (caused by a changed in typeshed stubs)

#11918 and #11863 are not a regression, but would make typeshed folks happy. We indicated we'd be willing to issue point releases for this in python/typeshed#4820 (comment)

[JukkaL]

I cherry-picked these additional fixes:

• Allows to use type[T] in stubs #11863
• Handle NoReturn type aliases #11912
• Fix PEP 585 type aliases in stubs #11918

Since there are still a few regressions without a fix, I will wait until tomorrow (Friday) before making the release. If important regressions are still open tomorrow, I can make a 0.932 release once we have the remaining fixes.

[JukkaL]

Another potential regression:

• mypy 0.930 does not install types #11920

[hauntsaninja]

#11717 was merged, but not backported
I believe right now #11924 is the only open PR that fixes a regression

[sobolevn]

PR with stubtest bugfix: #11931
This would be helpful for typeshed

[JukkaL]

More cherry-picks:

• Stricter None handling with --no-strict-optional, refs #11705 #11717
• Fix crash on ErasedType and covers_at_runtime #11924

[JukkaL]

I'm planning to include this PR in the release:

• Restore removed builtin_type() api method #11932

[JukkaL]

Two final cherry-picks for 0.931:

• stubtest: fix literal type construction #11931
• Restore removed builtin_type() api method #11932

We should be ready for the release soon, once the wheel build completes. I'll need to prepare a short release blog post as well.

[JukkaL]

...and 0.931 is out: http://mypy-lang.blogspot.com/2022/01/mypy-0931-released.html