Refactor the testsuite

[tkoskela]

Closes #200

Each Fortran test now has its own parametrized python test that specifies the values it checks from the output. I would have liked to also make results a fixture, but I couldn't figure out how to pass the path into a fixture.

[davidbowler]

Can you update the README.md to reflect how we now create new tests, and in particular how we can specify a different precision to the default?

[tkoskela]

Can you update the README.md to reflect how we now create new tests, and in particular how we can specify a different precision to the default?

Thanks, good point! I've done it, I hope this has made it a bit simpler to add tests as well.

In the future, you can select "request changes" in the review if you would like to request changes :smiley:

[davidbowler]

Can you add an example for how to set a precision that is not the default (or add it to the README.md)?

[tkoskela]

To set a custom precision I would maybe just modify the precision matching the key in the test function. Something like:

    @pytest.mark.parametrize("key", ['Harris-Foulkes energy',
                                     'Max force',
                                     'Force residual',
                                     'Total polarisation'])
    def test_003(self, key, default_precision, testsuite_directory):

        path = os.path.join(testsuite_directory, "test_003_bulk_BTO_polarisation")
        res = results(path, key)
        if key is 'Total polarisation':
          precision = custom_precision
        else
          precision = default precision
        np.testing.assert_allclose(res[0], res[1], rtol = precision, verbose = True)

Doing it through the fixtures seems too complicated

[davidbowler]

Could we also parametrize the precision in the same way as the key?

[tkoskela]

Not quite the same way, because then pytest would run through all [precision, key] combinations, and I don't think that's what we want? I've been struggling with how to make the parametrization do exactly what I want, given we want slightly different treatment for the keys, the test directories, and the precisions. I can talk you through this tomorrow morning

What I propose we can do, is make precision a function of key, which we can then call for each key value. See #207

[tkoskela]

You already merged this PR, do you want me to open another one?