-m disables autouse fixtures for custom Items

[dmtucker]

# conftest.py

import pytest


@pytest.fixture(scope='session', autouse=True)
def thing():
    raise Exception('thing ran!')

def pytest_collect_file(path, parent):
    return FooItem(path, parent)

class FooItem(pytest.Item, pytest.File):

    def __init__(self, *args, **kwargs):
        super().__init__(*args, **kwargs)
        self.add_marker('foo')

    def runtest(self):
        raise Exception(str(self.fspath) + ' ran!')

# test_foo.py

def test_whatever():
    pass

If I run test_foo.py, I get a failure (from the injected FooItem) and an error (from the exception in the thing fixture, which ran because it is autouse=True).

$ venv/bin/pytest test_foo.py
...
==================== 1 failed, 1 warning, 1 error in 0.15s =====================

However, if I just want to run the FooItem (which has its own marker, 'foo'), the autouse fixture is skipped:

$ venv/bin/pytest -m foo test_foo.py
...
================== 1 failed, 1 deselected, 1 warning in 0.17s ==================

• Note that the results show a failure from the FooItem, not an error from the thing fixture.

Here's my venv's pip list:

Package            Version
------------------ -------
attrs              19.3.0 
importlib-metadata 1.5.0  
more-itertools     8.2.0  
packaging          20.1   
pip                19.2.3 
pluggy             0.13.1 
py                 1.8.1  
pyparsing          2.4.6  
pytest             5.3.5  
setuptools         41.2.0 
six                1.14.0 
wcwidth            0.1.8  
zipp               2.1.0  

I'm on macOS 10.13.6.

[dmtucker]

This is very similar to #5715.

[RonnyPfannschmidt]

fixtures are currently not implemented at the level of Item. but at the level of specific Item subclasses

this in turn means, that FooItem won't get fixtures

[dmtucker]

@RonnyPfannschmidt Would it be reasonable for me to inherit from one of those specific subclasses instead of Item directly?

For context, I'm working on adding pytest-xdist support to pytest-mypy... It needs to run mypy once, then the collected MypyItems need to see if their associated file has errors in the output. Currently, the run is done in pytest_runtestloop which is problematic because pytest-xdist re-implements that. (Even if it didn't, as is that would result in multiple mypy runs...)

The strategy I've been working with is described in the "Making session-scoped fixtures execute only once" section of the pytest-xdist README; however, that won't work if only the MypyItems are run (e.g. with pytest -m mypy or if no fixture-supporting tests are found). Also, I can't access tmp_path_factory or worker_id from a hook (#5012).

Any tips or ideas you have would be much appreciated!

[RonnyPfannschmidt]

sub-classing python items could be a way to work around current limitations

my personal opinion is that code analysis like flake8, mypy & co doesn't belong into test execution

[dmtucker]

👌

FWIW, I was able to get something working (realpython/pytest-mypy#73) by combining that xdist recipe (but not in a fixture) with this one: https://hackebrot.github.io/pytest-tricks/shared_directory_xdist/