Parent class marked fixture no longer reached with pytest 7.2

[jaraco]

Starting with the release of pytest 7.2, the distutils tests started failing (pypa/distutils#186).

I read the changelog for 7.2, but I don't see anything relevant.

The issue is likely due to some of these factors:

• the test class is a subclass of another class
• both classes are marked with usefixtures
• The fixture on the subclass is reliant on the prior execution of the fixture in the parent class.

See the upstream bug for:

• a detailed description of the bug or problem you are having
• output of pip list from the virtual environment you are using
• pytest and operating system versions

For consideration:

• minimal example if possible

I haven't put together a minimal example yet as I'm still triaging the issue. Is there a known change with pytest 7.2 that was expected to change this behavior?

[jaraco]

To be sure, these tests deserve a refactor. The tests are in this state because the project is transitioning away from unittest, but some artifacts remain.

[jaraco]

Probably plugin:unittest is not relevant here, because the class hierarchy is no longer based on unittest.TestCase. I mention unittest because it explains why the fixtures reference request.instance as self (an artifact of the expectation when they did rely on TestCase).

[RonnyPfannschmidt]

this may be related to extraction of marks from the mro, (dint happenbefore, was aded recently

[jaraco]

The MRO refactor was #7792 (31df38f). I've confirmed that the distutils fail on 31df38f but pass on 31df38f^1, confirming that change is the culprit.

The changelog says:

When inheriting marks from super-classes, the pytestmark attribute of the sub-class now only contains the marks directly applied to it. Previously, it also contained marks from its super-classes.

That implies that it's intentional that previously reachable marks are now intentionally ignored.

I'm not sure why that wouldn't be a breaking change.

Is there any chance pytest will restore support for this use-case? Is there a way to apply a fixture at a superclass and have it take effect in a subclass?

[RonnyPfannschmidt]

The markers should take effect as all of them should land in the node for the class, I'll have to investigate the order difference

[jaraco]

In that case, let me put together a more minimal repro.

[jaraco]

Good news. A minimal repro was easy and I learned something:

 draft @ cat test_something.py
import pytest

@pytest.fixture
def add_attr1(request):
    request.instance.attr1 = object()


@pytest.fixture
def add_attr2(request):
    request.instance.attr2 = request.instance.attr1


@pytest.mark.usefixtures('add_attr1')
class Parent:
    pass


@pytest.mark.usefixtures('add_attr2')
class TestThings(Parent):
    def test_attrs(self):
        self.attr1
        self.attr2

 draft @ pip-run 'pytest@git+https://github.com/pytest-dev/pytest@31df38f6a79b0734be3c1386eb7df6512b1a321b^1' -- -m pytest
  WARNING: Did not find branch or tag '31df38f6a79b0734be3c1386eb7df6512b1a321b^1', assuming revision or ref.
========================================================================= test session starts ==========================================================================
platform darwin -- Python 3.11.4, pytest-7.2.0.dev331+g15ac0349, pluggy-1.2.0
rootdir: /Users/jaraco/draft
plugins: anyio-3.6.2
collected 1 item                                                                                                                                                       

test_something.py .                                                                                                                                              [100%]

========================================================================== 1 passed in 0.00s ===========================================================================
 draft @ pip-run 'pytest@git+https://github.com/pytest-dev/pytest@31df38f6a79b0734be3c1386eb7df6512b1a321b' -- -m pytest
========================================================================= test session starts ==========================================================================
platform darwin -- Python 3.11.4, pytest-7.2.0.dev338+g31df38f6, pluggy-1.2.0
rootdir: /Users/jaraco/draft
plugins: anyio-3.6.2
collected 1 item                                                                                                                                                       

test_something.py E                                                                                                                                              [100%]

================================================================================ ERRORS ================================================================================
_______________________________________________________________ ERROR at setup of TestThings.test_attrs ________________________________________________________________

request = <SubRequest 'add_attr2' for <Function test_attrs>>

    @pytest.fixture
    def add_attr2(request):
>       request.instance.attr2 = request.instance.attr1
E       AttributeError: 'TestThings' object has no attribute 'attr1'

test_something.py:10: AttributeError
======================================================================= short test summary info ========================================================================
ERROR test_something.py::TestThings::test_attrs - AttributeError: 'TestThings' object has no attribute 'attr1'
=========================================================================== 1 error in 0.02s ===========================================================================

The issue isn't that the fixture on the parent class isn't being called, but that the fixtures are called in a different order with the fixture on the child being called before the fixture on the parent. Independent of this specific example, I'd expect fixtures on more basic classes to be called first, as the child (derived) class is liable to depend on behavior from the parent but not vice-versa (from a modeling perspective, the parent class is largely unaware of the child classes).

[RonnyPfannschmidt]

Found the bug

https://github.com/pytest-dev/pytest/blob/main/src/_pytest/mark/structures.py#L377 is missing a reversed when considering the mro

[RonnyPfannschmidt]

@jaraco seems like our timing was unfortuate