Trying to figure out why Pytest is crashing on Travis on the other branches, and I'm wondering if it's because there are too many warnings. I'm not sure if we should backport this as it bumps the pytest requirement (won't be a problem in Fedora at least)?

Should this be backported to the 2.2.x branch?

Lets see if it backports cleanly

@meeseeksdev backport to 2.2.x

@ArchangeGabriel if that does not go cleanly, are you interested in doing the backport? Failing that, we should just pin/document that you must use old versions of pytest with the 2.2.x branch.

Something went wrong ... Please have a look at my logs.

It seem that the branch you are trying to backport to does not exists.

@meeseeksdev backport to v2.2.x

sigh, second times the charm?

Owee, I'm MrMeeseeks, Look at me.

There seem to be a conflict, please backport manually. Here are approximate instructions:

1. Checkout backport branch and update it.

$ git checkout v2.2.x$ git pull

2. Cherry pick the first parent branch of the this PR on top of the older branch:

$ git cherry-pick -m1 ca444e2a74759e559198403dbb1b96ff61489e79

3. You will likely have some merge/cherry-pick conflict here, fix them and commit:

$ git commit -am 'Backport PR #12154: Avoid triggering deprecation warnings with pytest 3.8.'

4. Push to a named branch :

git push YOURFORK v2.2.x:auto-backport-of-pr-12154-on-v2.2.x

5. Create a PR against branch v2.2.x, I would have named this PR:

"Backport PR#12154 on branch v2.2.x"

And apply the correct labels and milestones.

Congratulation you did some good work ! Hopefully your backport PR will be tested by the continuous integration and merged soon!

If these instruction are inaccurate, feel free tosuggest an improvement.

I’ve backported cleanly.

The only difference is thatwith _cleanup_cm(): is not present in 2.2.x, so there is one level indentation difference.

The relevant part:

diff --git a/lib/matplotlib/testing/conftest.py b/lib/matplotlib/testing/conftest.pyindex 40ce56c4895..d2f699c2be8 100644--- a/lib/matplotlib/testing/conftest.py+++ b/lib/matplotlib/testing/conftest.py@@ -24,19 +24,19 @@ def mpl_test_settings(request):     original_settings = matplotlib.rcParams.copy()     backend = None-    backend_marker = request.keywords.get('backend')+    backend_marker = request.node.get_closest_marker('backend')     if backend_marker is not None:         assert len(backend_marker.args) == 1, \             "Marker 'backend' must specify 1 backend."-        backend = backend_marker.args[0]+        backend, = backend_marker.args         prev_backend = matplotlib.get_backend()     style = '_classic_test'  # Default of cleanup and image_comparison too.-    style_marker = request.keywords.get('style')+    style_marker = request.node.get_closest_marker('style')     if style_marker is not None:         assert len(style_marker.args) == 1, \             "Marker 'style' must specify 1 style."-        style = style_marker.args[0]+        style, = style_marker.args     matplotlib.testing.setup()     if backend is not None:@@ -64,7 +64,7 @@ def mpl_image_comparison_parameters(request, extension):     # pytest won't get confused.     # We annotate the decorated function with any parameters captured by this     # fixture so that they can be used by the wrapper in image_comparison.-    baseline_images = request.keywords['baseline_images'].args[0]+    baseline_images, = request.node.get_closest_marker('baseline_images').args     if baseline_images is None:         # Allow baseline image list to be produced on the fly based on current         # parametrization.

This fixed ~2500 errors in our build (ArchLinux). I can make a PR.

Please do!

Should I revertd7e3789 in the same PR or in a follow-up one?

I will go with a follow-up one. Just a small note, seemshttps://github.com/matplotlib/matplotlib/blob/master/.appveyor.yml was not updated during this PR and still lists3.4 as the minimum version instead of3.6 everywhere else.

#13181

Just a small note that part of my backporting work in#13181 should have been applied to 3.0.x too, and especially this PR. When building 3.0.3 for Arch, I’ve encountered numerous errors in tests that have been fixed in 2.2.4 but not 3.0.3. But since 3.0.3 was the last bugfix release, I suppose it is not worth backporting now?

Drat, sorry we missed that 😞

3.0.3 is planned to be the last 3.0.x release and I am not sure doing a 3.0.4 to fix future compatibility with pytest is worth it, but if we find anything else we should fix this along with it.