Release 0.960 planning

[JukkaL]

I'm planning to make a 0.960 release around May 24 (tentative timeline).

Please post here any issues you'd like to see fixed in the release, or PRs you'd like to have merged. I've already cut the release branch, but we can still cherry-pick low-risk features and bug fixes until May 20. After May 20 only critical bug fixes and documentation updates should be cherry-picked.

[huguesb]

#12616
#12700
#12707

[ethanhs]

I have noticed 3 regressions on Windows (maybe related to #12615 as well), namely testCompileMypyc, testErrorOutput, and testNamespacePkgWStubs. I plan on looking at these more tomorrow.

[JukkaL]

It would be good to fix all the tests failing on master (on all platforms), or at least understand why they are failing.

[huguesb]

It would be good to fix all the tests failing on master (on all platforms), or at least understand why they are failing.

My rough sense of the failures of namespace pkg tests I have experienced on mac is that they're a symptom of racy filesystem operations. I unfortunately do not have the bandwidth or context to investigate in more details.

[JukkaL]

I ran some experiments on macOS and the test failures seem to be related to race conditions. Running tests sequentially seems to work. Some races also reproduce on Linux (again only when running tests in parallel). There may be nothing really platform specific in these.

Anyway, the test failures don't block the release, but it would be great to get to the bottom of this, since test flakes are annoying and can mask real problems.

[JukkaL]

I've cherry-picked these commits:

• Fix namedtuple crash in unannotated function #12804
• Fix crash on type alias definition inside dataclass declaration #12792
• Avoid crashing on invalid python executables #12812
• speedup typechecking of nested if expressions #12700
• checkexpr: cache type of container literals when possible #12707
• Typeshed cherry-pick: Ignore mypy errors in Python 2 builtins/typping #12826

[JukkaL]

Another cherry-pick:

• FindModuleCache: optionally leverage BuildSourceSet #12616

[bluenote10]

This regression is a big showstopper for us, and I hope I can provide a fix before the release:

• Regression: Stub generator no longer generates __members__ for enum class #12717

PR here:

• Bring back type annotation support of dunder methods in stub generator (fix #12717) #12828

[JelleZijlstra]

I merged #12828, can it still be cherry-picked?

[hauntsaninja]

#12832 is a very safe cherry pick. I also just encountered a mypy user who was confused about mypy's support for PEP 612 based on hitting this error message.

[JukkaL]

Cherry-picked these commits:

• Bring back type annotation support of dunder methods in stub generator (fix #12717) #12828
• Friendlier errors for PEP 612 #12832

[ethanhs]

Also, I plan on merging #11143 soon so it will go in the next release. Since it is a significant change to how module searching works, we may want to suggest people try pre-release mypy and report issues in the blog post for 0.960.

[rouge8]

Are there changes to plugin APIs? I noticed both mypy-nonfloat-decimal and an internal plugin stopped working when I test on mypy 128661c.

[hauntsaninja]

Thanks for testing before the release! Could you open an issue with more details? #9881 added to the plugin API, maybe #12762 has some effect?

[rouge8]

Opened #12861

[JukkaL]

Mypy 0.960 is out: https://mypy-lang.blogspot.com/2022/05/mypy-0960-released.html

[AlexWaygood]

#12868 is a crash report that bisects to

• Fix forward references and generic inheritance in attrs classes #12772

[JelleZijlstra]

#12877 reports a new error in 0.960 that I wasn't expecting, but the new error seems correct.

#12876 is another reported regression around dataclasses.

[AlexWaygood]

This is another dataclasses regression (though it's quite similar to some preexisting bugs relating to variable scope):

• false-positive name-defined regression with dataclasses and shadowing #12907

[JukkaL]

Released 0.961 with some fixes (#12908 and #12872).

#12907 is still open, but I think that it is less serious and we don't need to prioritize a fix right now.

I won't write a blog post since this is a small release.