Fix session fixture teardown exceptions being reported as duplicate XFAILs

[FazeelUsmani]

Fixes #8375

When a session-scoped autouse fixture raises an exception during teardown, and the last test in the suite is marked @pytest.mark.xfail, pytest was incorrectly showing an extra XFAIL line (duplicated) instead of reporting the teardown failure as an ERROR.

Problem

Before this fix:

• Output: 1 passed, 2 xfailed (incorrect - duplicate xfail)
• Session fixture teardown exceptions were being treated as expected test failures

After this fix:

• Output: 1 passed, 1 xfailed, 1 error (correct)
• Session fixture teardown exceptions are properly reported as errors

Root Cause

The xfail handling in pytest_runtest_makereport was being applied to all test phases (setup, call, teardown). This caused any exception during teardown to be converted to an xfail result if the test was marked with @pytest.mark.xfail.

Solution

Restrict xfail handling to only apply during the "call" phase. Setup and teardown failures are now properly reported as errors, regardless of xfail markers on the test.

This aligns with the principle that xfail should only apply to test execution logic, not to fixture infrastructure failures (setup/teardown).

Changes

• Modified src/_pytest/skipping.py: Added if call.when == "call" condition to xfail handling
• Added regression test: test_session_fixture_teardown_exception_with_xfail
• Updated existing tests to expect correct behavior (errors instead of xfails for setup/teardown failures)

[The-Compiler]

You have an if here already, so you can just combine this to

Suggested change

	if call.excinfo:
	if call.excinfo and call.when == "call":

instead of having to indent everything under a second if.

[The-Compiler]

This is a major behavior change, which probably warrants some more discussions (and possibly a major release, might be too late for 9.0.0 though).