Creating a patch release to fix matplotlib import error

[jgostick]

The deprecation of the matplotlib.docstring module to matplotlib._docstring was fixed in our previous patch, BUT it was not compatible with both new and old versions of matplotlib, so this patch uses a try/except to work with both versions.

[codecov]

Codecov Report

Merging #2857 (79e81d6) into release (3dbe3e5) will decrease coverage by 0.2%.
Report is 2 commits behind head on release.
The diff coverage is 45.0%.

❗ Current head 79e81d6 differs from pull request most recent head 8e320be. Consider uploading reports for the commit 8e320be to get more accurate results

Additional details and impacted files

@@            Coverage Diff            @@
##           release   #2857     +/-   ##
=========================================
- Coverage     89.9%   89.8%   -0.2%     
=========================================
  Files          148     148             
  Lines         8608    8624     +16     
=========================================
+ Hits          7742    7747      +5     
- Misses         866     877     +11     

[ma-sadeghi]

@jgostick Given that Matplotlib is going to remove this module altogether, can't we just host the code in our side? I think the license allows it.

[jgostick]

The windows builds are failing because the pypardiso team is having trouble with mkl again. They have a lot of recent activity and I'm sure they'll get this fixed soon. I am tempted to merge this now so we can get the pip release working with all version of mpl, which is more important than pardiso, I think?

[ma-sadeghi]

Maybe give them just a few more days? Either the issue is on pypardiso side, which I think will get fixed soon, or on Intel side, which probably won't get fixed soonish

[jgostick]

But technically pypardiso is not in our dependencies. This is only failing because we use it in some of our tests...but you're right, lets give it a few more days.