Improved wrappers and OpenMPI handling in Python MPI tests

[heplesser]

This PR fixes the problem we had with subprocess and mpirun. Thus, mpitests can now be run with conftest.py and the usual "missing" decorators can be applied to MPI tests.

See also #3466.

[nicolossus]

Nice, LGTM!

One suggestion you may consider (maybe in a future PR): As I understand the documentation, you can use the pytest_ignore_collect hook in conftest.py to skip tests in a specified directory based on a condition. This should work well for the MPI tests since they are in a dedicated directory, and would save you from having to decorate every MPI test function with skipif_incompatible_mpi.

[heplesser]

Nice, LGTM!

One suggestion you may consider (maybe in a future PR): As I understand the documentation, you can use the pytest_ignore_collect hook in conftest.py to skip tests in a specified directory based on a condition. This should work well for the MPI tests since they are in a dedicated directory, and would save you from having to decorate every MPI test function with skipif_incompatible_mpi.

Thanks for the hint, I had been looking for some way to skip entire directories. But I think it is not exactly what I would like, because if I understand this right, it would completely ignore the directory, so the test placed in it would be invisible in the test report. I would like them to be reported as skipped, so one is aware of the fact that tests were not run.

[nicolossus]

Nice, LGTM!
One suggestion you may consider (maybe in a future PR): As I understand the documentation, you can use the pytest_ignore_collect hook in conftest.py to skip tests in a specified directory based on a condition. This should work well for the MPI tests since they are in a dedicated directory, and would save you from having to decorate every MPI test function with skipif_incompatible_mpi.

Thanks for the hint, I had been looking for some way to skip entire directories. But I think it is not exactly what I would like, because if I understand this right, it would completely ignore the directory, so the test placed in it would be invisible in the test report. I would like them to be reported as skipped, so one is aware of the fact that tests were not run.

Good point. This would skip collecting the tests, so they will probably not show up in the summary. Another alternative could be to skip the tests at module level with pytest.skip(reason, allow_module_level=True), see https://docs.pytest.org/en/stable/how-to/skipping.html#skipping-test-functions