Template for unit tests

[RondeauG]

Pull Request Checklist:

• This PR addresses an already opened issue (for bug fixes / features)
  • Not really (at least not enough to resolve the issue)
• (If applicable) Documentation has been added / updated (for bug fixes / features)
• This PR does not seem to break the templates.
• HISTORY.rst has been updated (with summary of main changes)
  • Link to issue (:issue:number) and pull request (:pull:number) has been added

What kind of change does this PR introduce?

• Implements a working template on which to start creating unit tests for xscen.
• Added NotImplementedError in climatological_mean and compute_deltas when using daily data.

Does this PR introduce a breaking change?

• No.

Other information:

This is a start for #9

[RondeauG]

@aulemahal @Zeitsperre Feel free to work directly on this branch and to change/remove what I've done. Once we're happy with the formatting, we'll merge it and I'll start making real unit tests.

[RondeauG]

What I'm uploading here is a version that uses test_timeseries and hits pretty much all of climatological_mean. Since this is the best time to establish a format, I'm open to all suggestions!

[RondeauG]

Deux questionnements spécifiques:

• Est-ce que test_all_default/test_options est correct ? Est-ce que je devrais les combiner en une seule grosse fonction qui teste à la fois les défauts et options ? Est-ce que je devrais au contraire séparer chaque option dans son propre test ?
• Je dois redéclarer les mêmes timeseries plusieurs fois. Serait-ce mieux de les déclarer au tout début, en variable globale (ou à l'intérieur de TestClimatologicalMean) ?

[aulemahal]

 @RondeauG
Please see the changes introduced in my latest commit.

• Instead of looping inside the test function, I used pytest.mark.parametrize. The benefit here is that if a single element fails:
  • The others are also run
  • The pytest error output specifies the exact element causing the bug
  • And we can run a single element by hand if needed, while debugging.
• I switched to conventionnal assert when we are comparing scalars.
• I rewrote a check with a set logic instead of an all of a in comprehension list.
• Rewrote the calendar check to use only xarray properties and pytest parametrize

[RondeauG]

This should be good for review.

[Zeitsperre]

Will wait on @aulemahal's approval as well

[aulemahal]

One comment to the changes in climatological_mean.

For the test, I think this is a good example of what we want. Even, it is on the exhaustive side, which is good.

Warning : all tests here are done on artificial data, and thus it may fail to tests some real-life caveats. When adding more tests, we could make use of xclim's test data too.

[aulemahal]

This doesn't fit the docstring, which says None is possible!
I'm guessing you want to change the docstring?

[RondeauG]

Ah, true! I got confused because the default is a string, not None. I'll reverse that.