Fix loop factory parametrization for sync fixtures

[TLNing260310]

Fixes #1501.

Summary

• Parametrize synchronous tests over configured loop factories when their static fixture graph contains a pytest-asyncio-managed fixture.
• Run the plugin's generation hook after pytest and user parametrization hooks so direct parameters correctly shadow same-named fixtures while indirect parameters keep using their fixture definitions.
• Use the widest effective managed-fixture loop scope and resolve the loop-factory parameter before dependent fixtures inspect their caches.
• Cover fixture overrides, transitive dependencies, autouse fixtures, auto mode, teardown order, and direct/indirect parametrization; document the static-closure boundary.

Why

Previously only async test items requested the parametrized loop-factory fixture. A synchronous test consuming a managed async fixture therefore requested an unparametrized/default factory, changed the factory cache key, and could tear down shared fixtures at a sync/async test boundary. Parametrizing the affected synchronous test items keeps the factory cache key and fixture lifetime consistent.

Validation

• tests/test_loop_factory_parametrization.py: 55 passed with pytest 8.4.0
• tests/test_loop_factory_parametrization.py: 55 passed with pytest 9.1.1
• Broader pytest 8.4 run: 306 passed, 1 unrelated Windows nested-pytest test deselected; the local 8.4 environment did not include Hypothesis
• Broader pytest 9.1 run: 308 passed, 3 skipped, 1 unrelated Windows nested-pytest test deselected
• ruff check pytest_asyncio tests
• pyright pytest_asyncio tests

The documentation build was not run locally because Sphinx was not installed in the test environment.

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 94.83%. Comparing base (40342fa) to head (d86d02c).
⚠️ Report is 1 commits behind head on main.

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #1553      +/-   ##
==========================================
+ Coverage   94.50%   94.83%   +0.32%     
==========================================
  Files           2        2              
  Lines         510      542      +32     
  Branches       62       69       +7     
==========================================
+ Hits          482      514      +32     
  Misses         22       22              
  Partials        6        6              

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[tjkuson]

Thank you for the PR, but as-written it duplicates, coerces, and undermines pytest's internals in a way that's error-prone and difficult to maintain.

[tjkuson]

We should not be manipulating pytest internals like this.

[TLNing260310]

Addressed in d86d02c. I removed the object.setattr call and no longer mutate FuncFixtureInfo.initialnames. After pytest has created and pruned the Function item, the existing collection wrapper places the already-parametrized internal factory fixture first in the item's fixture resolution order, so its cache key changes before managed fixtures are read.

[tjkuson]

This changes the node IDs which is breaking for users that match on test names.

[TLNing260310]

Addressed in d86d02c. The original async pytest_generate_tests(tryfirst=True) ordering is restored. Sync fixture detection now uses a separate post-yield wrapper, so existing async IDs retain the factory-user_parameter order. I added collect-only regression coverage for both the preserved async order and the new sync user_parameter-factory order.

[tjkuson]

This reimplements pytest's fixture traversal system in a way that's error-prone is difficult to maintain.

[TLNing260310]

Addressed in d86d02c. The recursive fixture traversal, FixtureManager fallback, and override-chain bookkeeping have been removed. Sync detection now only scans the pytest-computed metafunc.fixturenames closure and reads the active fixturedefs[-1]. The docs explicitly limit the behavior to dependencies represented in the static closure rather than reimplementing traversal for dynamic or same-name override cases.

[TLNing260310]

Thanks for the review. I reworked the implementation in d86d02c around the three concerns: the custom recursive fixture traversal and FixtureManager fallback are gone; FuncFixtureInfo.initialnames is no longer mutated; and the original tryfirst ordering for async tests is restored with collect-only node-ID regression tests. Sync tests now use a separate post-yield generate wrapper, scan only the pytest-computed static fixture closure, and resolve the factory cache key before managed fixtures. The documented boundary is intentionally the closure pytest exposes: dynamic request.getfixturevalue dependencies and hidden same-name override chains are not rediscovered. The only remaining private access is a shallow, read-only lookup of the active FixtureDef through Metafunc._arg2fixturedefs because pytest 8.4-9.1 has no public equivalent. All CI checks pass, including pytest main, Python 3.10-3.14/3.14t on Windows and Linux, docs, lint, and packaging. Codecov reports all modified coverable lines covered. I would appreciate another review.