Use consistent tests

[akaszynski]

Ping @germa89:

Looks like #758 didn't quite resolve our issues. We're still seeing intermittent failures in test_load_array. You can validate this locally with:

pytest tests/test_mapdl.py -k test_load_array --count 100 -x

This fails because when our input array is shaped (10, 1), MAPDL will output a flat array (10,). Since numpy outputs flat arrays when the second dimension is 1, (e.g. numpy.arange), this PR changes the test to account for this edge case.

Lessons learned

If we want to test for a variety of edge cases, use parameterized inputs:

@pytest.mark.parametrize("dimx", [1, 3, 10])
@pytest.mark.parametrize("dimy", [1, 3, 10])
def test_load_array(mapdl, dimx, dimy):
    my_conv = np.random.rand(dimx, dimy)
    mapdl.load_array("my_conv", my_conv)

    # flatten as MAPDL returns flat arrays when second dimension is 1.
    if dimy == 1:
        my_conv = my_conv.ravel()
    assert np.allclose(mapdl.parameters["my_conv"], my_conv, rtol=0.0001)

This way, we test a variety of edge cases without resorting to random integers. Tracking down flaky tests is a nightmare. Let's avoid it by making tests consistent. If we want complete testing, let's focus on parameterized tests, not randomized tests. If we really want to do randomized tests correctly, checkout the hypothesis library. Disclaimer: it's going to make our tests longer and harder to maintain.

[germa89]

Hi @akaszynski
Thank you very much for working on this. This was a difficult one to debug (I believe). Happy to learn which steps you followed to discover the reason.