3.10.0.1 breaks CI in importlib_metadata

[jaraco]

In python/importlib_metadata#350, I observed that the tests on Python 3.10 started failing. Looking at a diff of the dependencies, I can see that the issue is brought about by the introduction of the recently released typing-extensions==3.10.0.1.

Error message is:

    class _ConcatenateGenericAlias(list):
    
        # Trick Generic into looking into this for __parameters__.
        if PEP_560:
>           __class__ = _GenericAlias
E           NameError: name '_GenericAlias' is not defined

.tox/python/lib/python3.10/site-packages/typing_extensions.py:2394: NameError

I suspect this issue will have widespread impact on environments running under Python 3.10.

[JelleZijlstra]

Thanks for bringing this up! I can release a new version as soon as this is fixed, but we should probably get CI back up first (#864).

[Fidget-Spinner]

@jaraco thanks for highlighting this to us and sorry about that. I'll try to fix this ASAP.

[sobolevn]

Fixed in #869

[Fidget-Spinner]

Yep. Thanks again @sobolevn!

[JelleZijlstra]

The CI PR might take a while longer so I'm going to try running tests locally myself and then release 3.10.0.2 with the fix.

[srittau]

@JelleZijlstra Feel free to merge the CI PR. I think it's "good enough" and can be cleaned up later

[JelleZijlstra]

We need to wait for the CLA to land.

[srittau]

We need to wait for the CLA to land.

Ah, ok.

[Fidget-Spinner]

I'm awfully confused why plain _GenericAlias passed in 3.9 but failed in 3.10 (also I could've sworn that I tested it locally on 3.10 in that implementation and it passed 😞.

Edit: woops nevermind, I see Jelle found out why in #871. -- I was wrong here, please ignore this! That's another unrelated but similar-looking issue.

[mgorny]

Note that there were some last-minutes changes in typing in Python 3.10.0rc1, so it might have been that if you were testing on b4.

[AMDmi3]

Judging by my runs, it's similarly broken everywhere from 3.10.0-b2 to 3.10.0-rc1.

[Fidget-Spinner]

Nevermind, it seems that I'd made a mistake -- I just tested this on 3.10a7 and it didn't pass there either (I think I might've tested just 3.5-3.9). I'm really sorry everyone.

Hopefully our new CI with 3.10 runs will catch such things in the future. I'll make sure to run 3.10 tests locally too.

[JelleZijlstra]

Update: I just released 3.10.0.2, which passes tests under 3.10.0rc1.