TST: Skip scipy tests if scipy isn't installed

[greglucas]

Scipy isn't a required dependency, so skip the tests that require scipy if scipy isn't found.

pytest.importorskip in the module-level imports caused flake8 warnings, so I decided to just try/except and skip instead which passes flake8 without requiring a bunch of extra ignores.

closes #2275

[rcomer]

I think contour itself does not need scipy, but specifically test_contour_linear_ring uses those two imported objects. So I suggest moving the imports within that test and putting the skip just there.

[greglucas]

Good point. Moved that down into the test.

[rcomer]

In here, everything except test_regridding_with_invalid_extent will pass providing you have at least one of scipy or pykdtree installed. Is it worth modifying this to try both imports? Or is it too unlikely that a developer would have pykdtree but not scipy?

I am also unclear how to add appropriate skips to the parametrization of test_regridding_with_invalid_extent if we did that...

[greglucas]

Sure, it is simple enough to do the nested try/except and put the skip only at the nested version. I also added a skip to the parameterization if scipy can't be imported there.

[rcomer]

If I run without scipy or pykdtree, I get failures in test_examples.py and test_imges.py. That could arguably be out of scope of this PR though.

[greglucas]

If I run without scipy or pykdtree, I get failures in test_examples.py and test_imges.py. That could arguably be out of scope of this PR though.

Good catch, I didn't realize I had pykdtree hanging around still. In fact, I get even more ERRORS due to image dependency imports in other modules like the ows tests. This is adding a lot of conditional logic for the skips in many places at this point considering no one else has complained about this before! I'm not sure how valuable this is...

[dopplershift]

I think it's definitely helpful given that neither of these are considered required, so IMO we should be able to run the test suite successfully without them.

[dopplershift]

I guess the alternative is to add one/both to the "test" optional dependencies.

[greglucas]

Alright, I refactored it a bit and tested this with the various combinations of pykdtree/scipy available/not available so that there are no errors.

There are a few problems that were slightly annoying and made this maybe a little more verbose than ideal:

1. We import our img_transform in other modules, for example to get _OWS_LIB_AVAILABLE we go through that entire module that has an import of img_transform, which requires us to skip above that import statement. This also means we can't set pytestmark = requires_pykdtree_scipy at the top of the module because that occurs after the module imports.
2. We have optional dependencies on two imports scipy or pykdtree, so pytest.importorskip I don't think will work in that situation. So I added a _HAS_PYKDTREE_SCIPY variable that we then use to just apply a pytest.skip() early on in the test module imports.

Significant changes and scope expansion since this approval

[QuLogic]

Seems to still be conflicts now?

[QuLogic]

Suggested change

	import pytest
	try:
	import scipy # noqa: F401
	except ImportError:
	pytest.skip("scipy is required for vector transforms", allow_module_level=True)
	import cartopy.crs as ccrs
	import cartopy.vector_transform as vec_trans
	from cartopy.tests.conftest import requires_scipy
	import cartopy.crs as ccrs
	import cartopy.vector_transform as vec_trans
	pytestmark = requires_scipy

[greglucas]

This one unfortunately doesn't work because import cartopy.vector_transform raises an import error before the pytestmark takes effect.

Forgot I had lingering outdated comment

[rcomer]

Sorry for previously messed up review - I started adding comments a couple of weeks ago but then PRed that change instead. Then forgot I still had the comment here when I approved it. Can’t seem to delete the comment now either.

Anyway, I think this is good to go.

[QuLogic]

Needs a rebase.