chore: stricter mypy

[henryiii]

I was looking at updating #1873 (and I realized it would be better to wait till after #1912) and I noticed we weren't very sensitive to updating things like util.call -> utils.cmd.call in our tests; I'd expect mypy to be able to detect all these without wasting time to run things. I also noticed this in #2048, where the tests weren't noticing the change to enable. These improvements should get proper mypy "coverage" of our tests. We still don't have 100% typing for arguments, but things like not allowing untyped function calls should require helper functions fully typed.

• chore: improve mypy (setuptools, pytest added, simpler strict setting, dropped unreachable since it depends on platform)
• chore(types): type functions in tests
• chore(types): No partial types in tests

[Czaki]

Looks good

[Czaki]

Does having so many non pinned dependencies do not make pre-commit mypy fragile?

[henryiii]

A little (though pytest isn't too bad). Though generally if something changes we just adapt to the change.

Pre-commit doesn't have a way to pin these. Maybe eventually uv will.

[Czaki]

Pre-commit doesn't have a way to pin these. Maybe eventually uv will.

What did you mean by uv will? pre-commit.ci do not allow to pass any constraints.

The only possible way that I see is to have own CI job to bump constraints.

[henryiii]

I mean if uv adds multiple independent locked environments (and tasks), we could use that to move this check to uv.

[Czaki]

in napari (https://github.com/napari/napari) we use uv to compile a constraints file and call it from tox.

Maybe the same approach could be used here?

[henryiii]

If this becomes too much of a problem, we could probably do something (setting it up with nox would be easy). But for now it's fine, and long term we might end up moving from mypy to red-knot anyway.