[ilevkivskyi]

Fixes #9756
Fixes #10160
Fixes #13182
Fixes #14443

This is needed to de-risk #22286

[hauntsaninja]

looks good!

[hauntsaninja]

Suggested change

	x = [C1, C2]
	x = [C1, C2]
	reveal_type(x)
	from typing import TypeVar
	T = TypeVar("T")
	def join(a: T, b: T) -> T: ...
	reveal_type(join(C1, C2))

[ilevkivskyi]

Btw it looks like this fixes 3 existing issues, updated PR description. The homeassistant error are bizarre, need to investigate. Error in sympy means we still can infer a join that is not a supertype of one of the items, may need some more tightening. The rest is likely because we handle argument names in lambdas incorrectly.

[ilevkivskyi]

Errors in pandera are somewhat unfortunate, but are actually consistent. Mypy uses surrounding items in a list to infer type of lambdas. In their case they should have used a list of tuples as an argument to pytest.mark.parameterize(), not a list of lists.

[ilevkivskyi]

Actually something is fishy there, I will dig more deeply.

[ilevkivskyi]

OK, this uncovered inconsistency in ignore_pos_arg_names: it randomly stops working in some cases. For example, if left and right have an argument with the same name in different positions. But this is a rabbit hole where I don't want to go.

I think I will add a workaround for now with saying that meet(Any, X) is Any for the purpose of argument types (it should be the case in general, but this is yet another rabbit hole).

[github-actions]

According to mypy_primer, this change doesn't affect type check results on a corpus of open source code. ✅