Update mypy to 2.4.0

[srittau]

No description provided.

[srittau]

This was more complicated than expected, but finally ready for a final review.

[AlexWaygood]

Thank you!

[AlexWaygood]

I would still personally lean towards just following the C implementation here, since it's the only one that will be relevant to the vast majority of type-checker users. And I think that's what we've done elsewhere in a few other places. But don't feel strongly; this is fine for now. And as you say, these aren't particularly useful, so it probably just doesn't really matter for users.

[srittau]

I prefer to support only what's supported by both versions.

[AlexWaygood]

not an issue to be fixed in this PR, but we might want to revisit the annotations for set.__and__. They look incorrect -- it doesn't seem like set.__and__ really takes any abstract set for its second argument. It seems like it might also only take sets and frozensets, similar to set.__rand__:

>>> from collections.abc import Set
>>> class Foo(Set):
...     __contains__ = __iter__ = __len__ = lambda *args: ...
...
>>> {1, 2}.__and__(Foo())
NotImplemented

Possibly the same also goes for other set dunders?

[srittau]

I'll actually fix this here, since I've also fixed a few other dunder annotations.

Similar to the other builtins, set checks explicitly for set or frozenset (or derived classes) using PyAnySet_Check:

https://github.com/python/cpython/blob/2eb2af805c156ca52b316bb4ced5e0ab3971d0ae/Objects/setobject.c#L1876-L1889

static PyObject *
set_and(PyObject *self, PyObject *other)
{
    if (!PyAnySet_Check(self) || !PyAnySet_Check(other))
        Py_RETURN_NOTIMPLEMENTED;
    /* ... */
}

[AlexWaygood]

I think the nonreflected frozenset dunders have the same issue:

>>> from collections.abc import Set
>>> class Foo(Set):
...     __contains__ = __iter__ = __len__ = lambda *args: ...
...
>>> frozenset({1, 2}).__and__(Foo())
NotImplemented

[srittau]

Good catch, they would have.

[srittau]

That said, I'm going to revert the last two commits for now and commit. This seems to open a whole new can of worms.

[AlexWaygood]

Sounds good

[github-actions]

Diff from mypy_primer, showing the effect of this PR on open source code:

scipy-stubs (https://github.com/scipy/scipy-stubs)
- tests/misc/test_submodules_empty.pyi:3: note: See https://mypy.readthedocs.io/en/stable/running_mypy.html#missing-imports