TST: Enable pytest-xdist

[Kojoley]

No description provided.

[QuLogic]

Good to see test times down to about the same as nose.

[QuLogic]

Will this affect the determinism tests?

[tacaswell]

Aren't the tests that really care about this run in a shell-call with their own seeds set?

[QuLogic]

I don't see it set/unset PYTHONHASHSEED, so wouldn't it be inherited?

[tacaswell]

Yeah, I think you are right 🐑

[Kojoley]

Well, I have find out that xdist collection problem is triggered by test_mathtext.py, but I do not know what exactly, maybe globals() injection magic or something else. I think the best option for us is to rewrite test_mathtext to pytest fixtures and forget about this xdist issue.

[QuLogic]

Hmm, that definitely is weird; is there a way to parametrize the fileename passed to the image_comparison decorator?

[tacaswell]

It still looks longer, the nose version it matches is the one for coverage which does not use parallelism. Am I confused?

[QuLogic]

Oh, I wasn't looking at the parallelism, just that it wasn't exceptionally slow.

[Kojoley]

We do not run all the tests with nose. test_axes takes 2 mins on my machine with -n 2. IIRC nose builds previously had taken about 12 mins, what is close to the build time we see on pytest build here.

[Kojoley]

Also the slowdown may be related to the problem described here pytest-dev/pytest-xdist#18

[tacaswell]

sorry, I think I was too optimistic about merging this.