Allow using typing_extensions.TypeVar in stubs

[AlexWaygood]

No description provided.

[AlexWaygood]

noqa'd in python/typeshed#11880

[github-actions]

This change has no effect on typeshed. 🤖🎉

[bzoracler]

While this would have been great, the motivation for using typing_extensions.TypeVar in the PR was actually because of the existing TypeVar node class. I tried to search for how typeshed uses default= and noticed that TypeVars which use it were still being imported from typing (e.g. contextlib.pyi).

I'm guessing typing.TypeVar(default=) only worked because symbols from typing_extensions are now treated as exactly equivalent as those from typing. Not sure what typeshed's policy on this actually is; type-checkers certainly don't behave the same:

• mypy: No errors in Python 3.10
• Pyright: complains that you need Python 3.13

[AlexWaygood]

Right, I'm not sure why neither mypy nor pyright is emitting an error in contextlib.pyi on Python <3.13, even though we're passing the default= parameter to typing.TypeVar when it doesn't have that parameter. That seems like a bug in mypy and pyright, and is probably due to the fact that both type checkers extensively special-case TypeVar. I think we shouldn't rely on that bug; we should probably use typing_extensions.TypeVar when we want to provide a default value for the TypeVar.

[JelleZijlstra]

I was also curious why we hadn't run into this before. It might actually be reasonable for type checkers to allow default= on typing.TypeVar in stubs before Python 3.13 (I assume that's the choice pyright made), but in flake8-pyi we should allow typing_extensions.TypeVar so users can use it to avoid confusion with the runtime behavior.