Skip unwrapping plugin attributes that cannot hide a fixture

[aryansk]

Fix #14877.

FixtureManager.parsefactories() calls _check_for_wrapped_fixture() — which runs inspect.unwrap() — on every attribute of every registered plugin, even though most plugin attributes (plain functions, classes, modules, and values) can never contain a fixture definition. In a default session most registered plugins define no fixtures, so the scan is pure overhead repeated for every Config that is built (thousands of times in pytest's own pytester-based test suite).

Change

Only attributes which could hide a fixture definition behind another decorator need the unwrap:

• staticmethod/classmethod descriptors (e.g. @classmethod above @pytest.fixture), and
• objects with a __wrapped__ chain (e.g. from functools.wraps; a bare FixtureFunctionDefinition also sets __wrapped__ via functools.update_wrapper, so genuine fixtures still take this path — where _check_for_wrapped_fixture is a cheap no-op).

Everything else is skipped. safe_getattr (not hasattr) is used for the __wrapped__ probe so that objects with a broken __getattr__ do not raise (the existing test_parsefactories_evil_objects_issue214 guard, pytest#214).

The existing warning behaviour is preserved exactly: all four wrapped-fixture warning scenarios (@staticmethod, @classmethod, and functools.wraps-style custom decorators, with or without an intervening @classmethod) still emit their warnings, and the tests for them pass unchanged.

Validation

• New regression test test_parsefactories_only_unwraps_attributes_that_can_hide_fixtures asserts that plain values, methods, and functions are never unwrapped while fixture-bearing attributes and descriptors still are.
• Full suite: 3731 passed (the only failure, test_error_diffs[Compare attrs classes], is a pre-existing attrs-version issue on main).
• Local micro-benchmark of parsefactories on a plugin-like object with 600 plain attributes: 0.90 ms → 0.49 ms per scan (~45% faster).
• pre-commit (incl. mypy) passes on the changed files.

Note

This issue and this contribution were developed with AI assistance; the change itself is small and scoped, and the above describes exactly what it does.

[RonnyPfannschmidt]

this will severely downplay the effect of a different speefup im still working on, but thats a good thing

[aryansk]

Thanks Ronny — appreciated. Happy to adjust if the in-flight speedup changes the tradeoff; the focused fixture-wrap check stays as is unless you'd prefer otherwise.