fix matplotlib errors for single level discrete colormaps

[keewis]

I'm not quite sure what happened, but our upstream-dev CI doesn't fail anymore. Looking at the build log, the install of the dev version of matplotlib seems to fail, so this tries to additionally remove the matplotlib-base package to see if the error comes up again if we make sure we get the dev version of matplotlib.

matplotlib released a preview version to PyPI so we can run our tests again. This follows the suggestion from #4226 (comment): convert single levels to degenerate levels.

• Closes failing upstream-dev CI: matplotlib #4226, towards ⚠️ Nightly upstream-dev CI failed ⚠️ #4947
• Passes pre-commit run --all-files

[keewis]

hmm... it seems that's because the repository is down, so now we're falling back to the PyPI version. @tacaswell, can you confirm?

[tacaswell]

It does look like the rackspace server we were pushing to seems to be down. This is more of a @matthew-brett question.

[matthew-brett]

Thanks @tacaswell :) Er - sorry - I'm lost - is the problem that xarray can't see the Matplotlib --pre binaries?

[keewis]

yes, we can't seem to use pip to install the pre binaries using

python -m pip install \
        -f https://7933911d6844c6c53a7d-47bd50c35cd79bd838daf386af554a83.ssl.cf2.rackcdn.com \
        --no-deps \
        --pre \
        --upgrade \
        matplotlib

also, when pointing my browser to that url, I get a resource not found page. I'm not sure if I did something different back then, but I think when I last did that about a month ago I got a page listing the available files instead.

[matthew-brett]

Yup - sorry - those URLs went down a while ago, as Tom said. Tom - are you uploading --pre binaries to Anaconda or similar?

[keewis]

if not, it might be worth copying numpy, pandas and sklearn which upload to https://pypi.anaconda.org/scipy-wheels-nightly/

[tacaswell]

Can you point me at the configs that upload to the correct place?

[keewis]

I don't know how those projects upload to that repository. @TomAugspurger, can you help with that?

Edit: https://github.com/MacPython/pandas-wheels/blob/master/azure/posix.yml, maybe?

[TomAugspurger]

Yep. I believe that @ogrisel can add you to the organization on anaconda.org so that you can create a key to upload to packages.

[max-sixty]

Is this & #4226 fixed now?

[keewis]

no, unfortunately not: right now, we're not installing matplotlib in that CI at all. I guess we're waiting on MacPython/matplotlib-wheels#13 to get fixed, so we can install the dev version of matplotlib. After that works, we might have to fix some test failures (and we might see #4226 again).

[max-sixty]

Ah, thanks @keewis !

[andersy005]

@keewis, is it safe to close this PR? As I understand it, this has been addressed in other PRs.

[keewis]

Not sure. Removing the matplotlib-base package as well is just a placeholder for the changes I opened this PR for. Back then we noticed some failures due to changes in matplotlib upstream, but before I could debug them the rackcdn index went down.

They made some progress with uploading to scipy-wheels-nightly since then, but that was using Travis so now they're moving the upload CI to github actions.

Once that works and the wheels are available on scipy-wheels-nightly I'm planning to change this PR to have the upstream-dev CI install from there and attempt to fix any bug that may be discovered.

[andersy005]

Removing the matplotlib-base package as well is just a placeholder for the changes I opened this PR for. Back then we noticed some failures due to changes in matplotlib upstream, but before I could debug them the rackcdn index went down.

Ah, I hadn't noticed that the rackcdn index was involved. My assumption was based on seeing the other upstream dev CI working :). Thank you for the clarification

[keewis]

if CI agrees this should finally be ready for merging. We still don't run our tests against nightly versions of matplotlib, but for that we will have to wait on upstream.