Make mypy test suite pass mypy type-check on Windows

[AlexWaygood]

Description

Mypy's test suite fails mypy on Windows. This PR fixes that. (This is my first non-docs contribution to mypy, so let me know if I'm doing anything wrong in this PR!)

Here is the error I get if I try running the mypy test suite on my Windows machine:

(venv) C:\Users\Alex\Desktop\Code dump\mypy>python runtests.py
run self: ['C:\\Users\\Alex\\Desktop\\Code dump\\mypy\\venv\\Scripts\\python.exe', '-m', 'mypy', '--config-file', 'mypy_self_check.ini', '-p', 'mypy']
mypy\test\data.py:41: error: Incompatible types in assignment
(expression has type overloaded function, variable has type overloaded
function)  [assignment]
            join = posixpath.join
                   ^
Found 1 error in 1 file (checked 163 source files)

FAILED: self

This error is due to the following lines in mypy/test/data.py:

mypy/mypy/test/data.py

Lines 38 to 41 in 2676cd8

	if case.suite.native_sep:
	join = os.path.join
	else:
	join = posixpath.join

On linux, os.path.join and posixpath.join have the same signature. However, on Windows, os.path.join is a reexport of ntpath.join, which has a different signature to posixpath.join due to the fact that the first parameter has a different name.

Here is how posixpath.join is defined in typeshed:

@overload
def join(a: StrPath, *paths: StrPath) -> str: ...
@overload
def join(a: BytesPath, *paths: BytesPath) -> bytes: ...

Here is how ntpath.join is defined in typeshed:

@overload
def join(path: StrPath, *paths: StrPath) -> str: ...
@overload
def join(path: BytesPath, *paths: BytesPath) -> bytes: ...

Test plan

Not sure. Guidance on this would be appreciated.

[AlexWaygood]

Excellent, CI errors in the one test I'm trying to fix. Not sure I really understand them.

[AlexWaygood]

This PR isn't a good solution to this problem. I have an idea for a better solution to the problem, which would involve making changes to typeshed instead.

[pranavrajpal]

I think this PR is a good idea, but we can probably simplify the type of join. We only seem to use join with exactly 2 str arguments, so I think declaring the type of join as Callable[[str, str], str] and removing the protocol should make this simpler (and hopefully also fixes the CI errors).

[AlexWaygood]

I think this PR is a good idea, but we can probably simplify the type of join. We only seem to use join with exactly 2 str arguments, so I think declaring the type of join as Callable[[str, str], str] and removing the protocol should make this simpler (and hopefully also fixes the CI errors).

Worth a shot, let's see what happens :)

[AlexWaygood]

I think the only way of creating a regression test for this kind of thing would be to run the self-check on Windows as well as Linux in the CI. Is that a change worth making?

[AlexWaygood]

This PR isn't a good solution to this problem. I have an idea for a better solution to the problem, which would involve making changes to typeshed instead.

FWIW, these are the changes to typeshed, which I think might be a better solution to this problem: python/typeshed#6812. (EDIT: These typeshed changes have now been merged, so the problem that this PR is solving will go away as soon as mypy's copy of typeshed is synced.)

[hauntsaninja]

Thanks for looking into this! I'd accept a PR that runs self check on Windows. Self check is pretty fast + hopefully helps keep contributor experience good for Windows users.

[AlexWaygood]

Thanks for looking into this! I'd accept a PR that runs self check on Windows. Self check is pretty fast + hopefully helps keep contributor experience good for Windows users.

Great! I'll close this and open a new PR for that, then, as soon as typeshed is resynced :)

[AlexWaygood]

Promised PR to run a Windows self-check is here: #11909 :)

Once #11905 is merged, I'll merge mypy/master into that branch, and the new test should hopefully pass, making the PR mergeable.