`--run-parallel` doesn't work well with old style `@pytest.mark.xfail("expr", ...)`

[colesbury]

Like #13, but with @pytest.mark.xfail with a string condition.

I don't have a minimal reliable reproducer, but I encountered this when running the following pybind11 tests with --run-parallel:

https://github.com/pybind/pybind11/blob/master/tests/test_unnamed_namespace_a.py

module 'pybind11_tests.buffers' has no attribute 'PYBIND11_INTERNALS_VERSION'

During handling of the above exception, another exception occurred:
Error evaluating 'xfail' condition
    (m.PYBIND11_INTERNALS_VERSION <= 4 and (m.defined___clang__ or not m.defined___GLIBCXX__)) or (m.PYBIND11_INTERNALS_VERSION >= 5 and not m.defined_WIN32_or__WIN32 and (m.defined___clang__ or m.defined__LIBCPP_VERSION))
AttributeError: module 'pybind11_tests.buffers' has no attribute 'PYBIND11_INTERNALS_VERSION'

...

ERROR tests/test_unnamed_namespace_a.py::test_have_both_class_any_struct - Failed: Error evaluating 'xfail' condition

[andfoy]

I cannot reproduce this failure case any longer:

============================================================= test session starts ==============================================================
platform linux -- Python 3.13.2, pytest-8.3.3, pluggy-1.5.0 -- /home/andfoy/anaconda3/envs/nogil/bin/python3
cachedir: .pytest_cache
hypothesis profile 'default' -> database=DirectoryBasedExampleDatabase(PosixPath('/home/andfoy/Documentos/Quansight/nogil/pybind11/tests/tests/.hypothesis/examples'))
C++ Info: 11.4.0 C++17 __pybind11_internals_v8_system_libstdcpp_gxx_abi_1xxx_use_cxx11_abi_1__ PYBIND11_SIMPLE_GIL_MANAGEMENT=False PYBIND11_NUMPY_1_ONLY=False
rootdir: /home/andfoy/Documentos/Quansight/nogil/pybind11/tests
configfile: pytest.ini
plugins: localserver-0.9.0.post0, hypothesis-6.118.8, run-parallel-0.2.1.dev0, order-1.3.0
collected 1222 items                                                                                                                           

../test_unnamed_namespace_a.py::test_have_class_any_struct[unnamed_namespace_a_any_struct] PASSED                                        [  0%]
../test_unnamed_namespace_a.py::test_have_class_any_struct[unnamed_namespace_b_any_struct] PASSED                                        [  0%]
../test_unnamed_namespace_a.py::test_have_at_least_one_class_any_struct PASSED                                                           [  0%]
../test_unnamed_namespace_a.py::test_have_both_class_any_struct PASSED                                                                   [  0%]

[rgommers]

I cannot reproduce this failure case any longer:

Would it make sense to add a test case with that syntax to the test suite in this repo, and then close this issue?