Error not raised in session fixture when mark.xfail is set on last test

Version
6.2.2

Context
Running a test suite with a session fixture that does some cleaning up

Description
If the last test of the suite has a mark.xfail, the error will be marked as an additional xfailed test

Minimal example

import pytest

@pytest.fixture(autouse=True, scope='session')
def failme():
    yield

    raise RuntimeError('cleanup fails for some reason')

def test_01_OK():
    assert 1 == 1

@pytest.mark.xfail()
def test_02_fail():
    assert 0 == 1

% py.test -v test.py
test.py::test_01_OK PASSED
test.py::test_02_fail XFAIL
test.py::test_02_fail XFAIL

To me, it would seem more appropriate to raise the error since the fixture is executed in the context of the session, not the test

Start by running the minimal example in test.py with pytest -v and observe the session-fixture cleanup failure after the final xfailed test. Trace how session fixture teardown errors are reported in this execution path. Done means the cleanup RuntimeError is raised or reported as an error rather than recorded as an additional xfailed test.