Missed several other places, will update PR soon.

Is this a continuation of #5325?

Also, please be aware of https://github.com/matplotlib/pytest-mpl

Pytest 3.0 released with this change:

yield-based tests are considered deprecated and will be removed in pytest-4.0.

😕

Coverage problem is because of if [[ $NOSE_ARGS="--with-coverage" ]]; then in .travis.yml. The expression should be either [ a=b ] or [[ a==b ]].

PR is now ready for review and merge.

Key notes:

• Collection phase is long due to image_comparison decorator
  (pytest runs setup_class during it)
• matplotlib.test is not aware of pytest yet
• pytest-xdist is disabled due to savefig has sideeffects #6899
• --collect-filter is blacklist by default (Include test files in coverage report #6902 (comment))
• --no-pep8 is implemented, but --pep8 is not

It looks like pytest is finding 5 more tests than nose so that is probably a victory.

It looks like pep8 is being run by pytest on travis, but not on appveyor, but that is ok for now.

There is a big CI hit time wise for this and we will have to (soon) pick some of the nose builds to drop.

It looks like pep8 is being run by pytest on travis, but not on appveyor, but that is ok for now.

I made this intentionally because we have separated pep8 only build on travis so there is no need to run them on all appveyor builds. Or I was wrong?

If anything pep8 should be skipped by travis pytest build. According to the handy output, it is only 45s which is not a big deal on top of our total runtime.

As far as I remember the main reason for splitting pep8 out in a separate build is not so much runtime but making it simpler to spot failures that are due to pep8 only