ENH: Align titles

[AnsonTran]

PR summary

Adds a function Figure.align_titles, that aligns titles of axes on the same row.

Closes #22376. Continuation of PRs #25591, #22793

Updated gallery example for align labels

PR checklist

• "closes #0000" is in the body of the PR description to link the related issue
• new and changed code is tested
• Plotting related features are demonstrated in an example
• New Features and API Changes are noted with a directive and release note
• Documentation complies with general and docstring guidelines

[AnsonTran]

Hello, would appreciate having someone look over this PR whenever you have the chance. Thanks!

[rcomer]

Thanks for your work on this @AnsonTran, I think it's looking great! I just have some minor comments.

[jklymak]

I think it's a valid question if align_labels should also align the titles

[rcomer]

I think it's a valid question if align_labels should also align the titles

I don't have an opinion on whether it should but, if it did, I guess it would need to be optional at least for a deprecation period.

I don't think we necessarily need to decide within this PR as it could be done as a follow-up.

[AnsonTran]

Not sure why this failed

[rcomer]

That test is currently failing on all the PRs. #28007 is trying to fix it.

[jklymak]

Looks good except for one nit

[rcomer]

The image test is unfortunately failing for a platform that was added since the previous commit #27723. I downloaded the test images and the diff looks like this:

I'm very confused how this is a failure.

[AnsonTran]

I see... Is it just on the new test? or other image comparison tests as well?

[ksunden]

If I shift the levels to increase the contrast, we see that there are very slight differences in the right hand plot surrounding the drawn line:

These differences are not meaningful, and are likely due to things like rounding modes at the hardware level (it is failing on macos ARM machines).

We had to add tolerances to ~50 image tests in #27723 for similar reasons. This is not that unexpected, just some (usually floating point) calculations are not quite exact and so lead to such things. 0.02 is a pretty low RMS.

Here is an example of adding such tolerance:

https://github.com/QuLogic/matplotlib/blob/48bc62d6088c262b1b5507981f2948d117e63601/lib/matplotlib/tests/test_contour.py#L167-168

The number needs to be at least as big as the number stated in the failure, so with a small buffer I might use something like 0.022 for this test (which says 0.020).

[rcomer]

Hi @AnsonTran we prefer rebase rather than merge to get the branch up-to-date with main. Are you happy to do that?

Edit: in case I'm not clear

git fetch upstream
git rebase upstream/main

[AnsonTran]

Yes sorry about that, just wanted to figure out what caused the codecov to fail