BUG: fix default math font in yt plots against matplotlib >= 3.4

[neutrinoceros]

PR Summary

Endgoal: resolve #3514
Basic testing seems to indicate this works, however I need to further check a few things. Is it still possible to change the math font using the matplotlib API ?
Submitting as a draft for now so I can at least check if image comparison tests pass as is.

Note: this reverts a temporary patch in tests/matplotlibrc from #3153.

PR Checklist

• New features are documented, with docstrings and narrative docs
• Adds a test for any bugs fixed. Adds tests for new features.

[neutrinoceros]

test cases to add:

import matplotlib as mpl
from yt.testing import fake_amr_ds
from yt.visualization.api import SlicePlot

ds = fake_amr_ds(fields=[("gas", "density")], units=["g/cm**-3"])
sp = SlicePlot(ds, "z", ("gas", "density"))

Expected result

# this case requires matplotlib 3.4 or greater
sp = ySlicePlot(ds, "z", ("gas", "density"))
sp.set_font({"math_fontfamily": "dejavusans"})

# same result is expected here 
with mpl.rc_context({"mathtext.fontset": "dejavusans"}):
    sp = yt.SlicePlot(ds, "z", ("gas", "density"))

Expected result

Note that this last case currently only works with matplotlib 3.4 + yt's main branch. It doesn't work with matplotlib 3.3 or earlier or with the current state of this PR, so I'm not 100% sure we need to support it, but it would be nice if we could avoid closing a door that matplotlib just opened for us.

[neutrinoceros]

So the image tests revealed two kinds of failures:

• line plots seem to not be properly affected by this patch (acutal bug)
• exponents in minor tick labels in colorbars for phase plots with logscale (weirdly specific...) are affected, but IMO they actually got better, so I think I'll propose to adopt this change

For illustration:

main branch

this patch

[neutrinoceros]

I'm now down to 9 errors (previously 41) for GH answer tests. All remaining errors are the kind I described in the previous comment, so I believe it' is mature enough to warrant bumping the answer store.
On the other hand, some of the failures on Jenkins at least are real: tests that check wether font parameters can be changed from the outside are showing that this feature is broken, so I still need to work on that. I'll open a PR to the answer store for starters, and focus on solving real issues on Jenkins before I can ask to bump this baseline too if needed.

[neutrinoceros]

Here's the PR to the answer store yt-project/answer-store#23
requesting early reviews from @matthewturk, @chummels, @Xarthisius regarding this point. Would it be okay to merge that PR before the present one is 100% finished or should I focus on getting things working on Jenkins first ?

[neutrinoceros]

Alright I believe that all failures, GH and Jenkins alike, are the kind I propose to adopt. Opening this proposing to approvals before I can bump references.

[neutrinoceros]

Note to reviewers: please consider reviewing this soon, the reports are not going to stay available for more than a couple days, after which we’d need to manually retrigger tests.

[matthewturk]

lgtm

[neutrinoceros]

@yt-fido test this please

[neutrinoceros]

I got "no old answer" twice on Jenkins after bumping the answer number. Either I don't understand how it's supposed to be done anymore or something isn't working. @Xarthisius, can you have a look ?

[Xarthisius]

@yt-fido test this please

[chummels]

Wait, so can we no longer manually set the font in our plots? I don't think i ever have, but it might be something people have used in the past.

[neutrinoceros]

No, it will still be possible, thanks to inheritance. Because I'm making ProfilePlot a subclass of PlotContainer, we don't need to have this method here, which was duplicated code from PlotContainer anyway :-)

[chummels]

OK, I'm a bit confused based on the description of this PR and the comments what this PR actually does (e.g., two "expected results" plots) Presumably this continues to use the old math font instead of the new MPL fonts for our plots. can you confirm?

[neutrinoceros]

Presumably this continues to use the old math font instead of the new MPL fonts for our plots. can you confirm?

Yes, this PR forces matplotlib to use the "cm" math font family, which was its default before matplotlib 3.4.0

A side effect of doing so, using matplotlib 3.4's api, is that negative exponents in tick labels are also forced to use this font, something we apparently were not doing in yt already, but arguably should have. This is the reason that I have a "bump answer store" PR attached :)

[chummels]

OK, looks good to me. Thanks!

[neutrinoceros]

Great, it's now down to a single error, same kind as the ones already bumped. Can I bump that one too ?

[Xarthisius]

Sure, but be careful about numbering. local_cylindrical_background_008 already exists. Possibly created by some different PR

[neutrinoceros]

alright, going straight to 009 then, thanks for the heads !

[neutrinoceros]

Something went wrong but at least it doesn't look like I corrupted the database. Let's try bumping one last time.

[neutrinoceros]

🎉 this is finally ready for merging

[matthewturk]

@meeseeksdev backport to yt-4.0.x

[lumberbot-app]

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 yt-4.0.x
$ git pull

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

$ git cherry-pick -m1 c6caf13b80c9495a6a8e8bb1ea6d5983ae8566e3

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

$ git commit -am 'Backport PR #3520: BUG: fix default math font in yt plots against matplotlib >= 3.4'

4. Push to a named branch :

git push YOURFORK yt-4.0.x:auto-backport-of-pr-3520-on-yt-4.0.x

5. Create a PR against branch yt-4.0.x, I would have named this PR:

"Backport PR #3520 on branch yt-4.0.x (BUG: fix default math font in yt plots against matplotlib >= 3.4)"

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!

Remember to remove Still Needs Manual Backport label once the PR gets merged.

If these instruction are inaccurate, feel free to suggest an improvement.

[neutrinoceros]

@meeseeksdev backport to yt-4.0.x

[lumberbot-app]

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 yt-4.0.x
$ git pull

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

$ git cherry-pick -m1 c6caf13b80c9495a6a8e8bb1ea6d5983ae8566e3

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

$ git commit -am 'Backport PR #3520: BUG: fix default math font in yt plots against matplotlib >= 3.4'

4. Push to a named branch :

git push YOURFORK yt-4.0.x:auto-backport-of-pr-3520-on-yt-4.0.x

5. Create a PR against branch yt-4.0.x, I would have named this PR:

"Backport PR #3520 on branch yt-4.0.x (BUG: fix default math font in yt plots against matplotlib >= 3.4)"

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!

Remember to remove Still Needs Manual Backport label once the PR gets merged.

If these instruction are inaccurate, feel free to suggest an improvement.

[neutrinoceros]

@meeseeksdev backport to yt-4.0.x