pytest-ification

[max-sixty]

So far, only the changes that unittest2pytest could fix. Need to work out how to sort through the exisiting xarray ones

EDIT: ready to merge (and keen to get this in before there are divergences from other branches)

[fujiisoup]

Thanks @maxim-lian,
This is a great step to clean up our test suit :)

Are you going to make our tests more pytest-ized (e.g. by adopting more @pytest.mark.parametrize)?
I think the current change is a good size.

There are some test failures (maybe due to mis-import).
After they are fixed, I will review more carefully and merge this so that this change would not be outdated.

[fujiisoup]

Maybe we can remove from xarray.core.pycompat import PY3 in line 3.

[max-sixty]

I think this is now passing throughout except for Python3.4, which I think is because pytest is on version 3.0.5, which doesn't have the match keyword from pytest 3.3.

Does anyone know why an old version of pytest is being installed there? From what I can see on the pytest home page, 3.4 is fully supported, and the 3.4 ci requirements don't force a lower version of pytest.

I could unravel that but if anyone has insight (and is anyone using 3.4? Though this PR prob not the place to make that decision)

I think we should try and merge this asap, as there are going to be (manageable) merge conflicts with other branches

[max-sixty]

Are you going to make our tests more pytest-ized (e.g. by adopting more @pytest.mark.parametrize)?

Not yet, I think that's a bigger exercise, and more difficult to script

[Zac-HD]

This should also apply to unsigned integers, and the resulting dtype should be given as a numpy dtype. Plus, we can make this much more efficient for small integers:

elif np.issubdtype(dtype, np.integer):
    if dtype.itemsize <= 2:
        dtype = np.float32
    else:
        dtype = np.float64
    fill_value = np.nan

[max-sixty]

Tests now pass, including flake8, apart from the pytest.warns.

Given how much code this touches, I'm keen to merge this ASAP, before divergences creep in. What's the best course on 3.4? pandas is in the process of dropping it.

Alternatively I could revert the pytest.warns changes, though there isn't one clean commit to revert.

[jhamman]

@maxim-lian - I opened #1829 to discuss dropping Python 3.4.

[fujiisoup]

Using pip seems to temporary solve this issue.
ci/requirements-py34.yml

name: test_env
dependencies:
  - python=3.4
  - bottleneck
  - flake8
  - pandas
  - pip:
    - coveralls
    - pytest>=3.3
    - pytest-cov

maybe not the best solution, though.

[fujiisoup]

There are two comment-outed lines using self.assertEqual in test_dataarray and test_dataset.

We are still using some methods of TestCase class, such as assertVariableNotEqual.
Can we add equivalent function to testings.py?

[max-sixty]

There are two comment-outed lines using self.assertEqual in test_dataarray and test_dataset.

That are comment-out as part of this PR? Or prior? If prior, then I think we should leave until we attempt to add those back in

We are still using some methods of TestCase class, such as assertVariableNotEqual.
Can we add equivalent function to testings.py?

I think the best way of handling those is xfailing an assert_equal test, rather than having a extra -not test for each variant. I also think we should do that in a separate PR (generally am keen to get this in before there are divergences)

[max-sixty]

Using pip seems to temporary solve this issue.

Nice, thanks. I've put this in. If we're on the way to deprecate 3.4, from which source we install pytest is flexible IMHO

[jhamman]

This looks good to me. Thanks for the cleanup.