6.0.0rc1 pytest.skip mypy error: Attribute function "__call__" ... does not accept self argument

[simonjayhawkins]

• a detailed description of the bug or suggestion
• output of pip list from the virtual environment you are using
• pytest and operating system versions
• minimal example if possible

import pytest


def valid_config():
    return False


def test_function():
    if not valid_config():
        pytest.skip("unsupported configuration")

$ pytest ~/pytest-mypy.py 
=========================================================================== test session starts ============================================================================
platform win32 -- Python 3.8.3, pytest-6.0.0rc1, py-1.9.0, pluggy-0.13.1
rootdir: C:\Users\simon
plugins: hypothesis-5.16.3, asyncio-0.12.0, cov-2.10.0, forked-1.1.2, xdist-1.32.0
collected 1 item

..\pytest-mypy.py s                                                                                                                                                   [100%] 

============================================================================ 1 skipped in 0.12s ============================================================================ 

$ mypy ~/pytest-mypy.py
C:\Users\simon\pytest-mypy.py:10: error: Attribute function "__call__" with type "Callable[[str, DefaultNamedArg(bool, 'allow_module_level')], NoReturn]" does not accept self argument
C:\Users\simon\pytest-mypy.py:10: error: Too many positional arguments for "skip"
C:\Users\simon\pytest-mypy.py:10: error: Argument 1 to "skip" has incompatible type "str"; expected "bool"
Found 3 errors in 1 file (checked 1 source file)

skip is decorated with @_with_exception(Skipped), will look soon but assuming decorator is not preserving signature

[bluetech]

There is some trickery there but I can't reproduce it here, pytest.skip is type-checked correctly.

If you can provide steps for creating a standalone venv which reproduce this (package versions + configs) that'd be great.

[bluetech]

In particular, I suspect it's maybe an older mypy version -- which versions are you using, and if using an older version, can you reproduce with latest mypy?

[simonjayhawkins]

Thanks @bluetech for looking into this. we are pinned to mypy 0.730 at the moment. I'll get back to you on using different versions

[simonjayhawkins]

@bluetech I can confirm that I don't get this error from mypy 0.750 onwards but not sure which mypy commit fixed this. maybe as a result of 'more powerful self types'. What confuses me is that this issue looks a bit like python/mypy#6910 which is still open.

Looking at the pytest code the annotation for __call__ in _WithException is annotated as _F which doesn't account for the self argument, so I think the error message being raised by mypy does make sense.

However, from a pandas standpoint, I'm happy to add #type: ignore to the few pystest.skip uses we have for now since we should be thinking about bumping our version of mypy anyhow.

cc @WillAyd

One thing i've not checked is the Python version. I am using Python 3.8.3 and so Protocol is being used for _WithException whereas for earlier versions, Protocol is an alias to Generic.

[simonjayhawkins]

ahh, just noticed the code comment

# Elaborate hack to work around https://github.com/python/mypy/issues/2087.
# Ideally would just be `exit.Exception = Exit` etc.

that's interesting. in pandas we normally just add an ignore for these cases. maybe python/mypy#2087 (comment) is relevant.

[Zac-HD]

we are pinned to mypy 0.730 at the moment... I can confirm that I don't get this error from mypy 0.750 onwards

As noted in the changelog, our type hints are developed against mypy 0.780 - and I think it's reasonable to require a recent version of mypy to use with the forthcoming major version of pytest.

@The-Compiler @nicoddemus I suggest that we close this without action, save perhaps noting in the changelog that mypy versions before 0.750 are known to be incompatible.

[nicoddemus]

@Zac-HD I agree we should only support versions >=X of mypy, and need to make that explicit somewhere (changelog is fine for now).

WDYT @bluetech?

[simonjayhawkins]

I suggest that we close this without action

i'm OK with that from pandas perspective.