[milanboers]

Another one 😒 still an issue with #7724 . Problem is MutableMapping is invariant, so the union type will actually never match.
Changed it to Mapping (which is covariant in the value type) now. It doesn't seem to be mutated anywhere, I also passed frozendict to try. Also tried assigning a dict[str, BufferedReader] to it as in the issue, which passes the type check now.

Same problem would occur in _Hooks, so changed that one too. Kept _TextMapping as-is because it requires str (would still fail on a subtype of str).

[AlexWaygood]

Thanks for all your work on this!

[shawnwall]

thank you! i tried digging but am not well versed in the ways of how tests work in typeshed, but shouldn't we try to get some test cases in place regarding these types of files= usages to make sure we are all covered in future?

[Akuli]

Typeshed doesn't do much tests in the usual sense. There has been a lot of discussion about this, but in short, we don't want to duplicate information that is already specified in the stubs.

For standard library stubs, we have mypy-primer, which runs mypy across many open-source projects on each new PR and posts a diff of mypy outputs as a comment. It would work very well for problems like this. Unfortunately it is currently not being used for third-party stubs, see hauntsaninja/mypy_primer#7.

[JelleZijlstra]

Though we do have a few assert_type()-based tests now, and we could start adding more.

[brandonchinn178]

FYI This breaks session.hooks['response'] = ... #7776

[Akuli]

Unfortunately it is currently not being used for third-party stubs

Now it is, and regressions like this are much less likely to happen: hauntsaninja/mypy_primer#31