Enable test for bbox_size limits in all Matplotlib 3 versions

[dhomeier]

Experimental PR to confirm or disprove the conclusions on how get_tightbbox calculation changes over Matplotlib version from #10900 (comment)
(tested locally against mpl 2.2.5, 3.0.3 and 3.3.2).

[dhomeier]

This does "unskip" the mpl2* tests, so probably needs a changelog? And there is no provision for a Matplotlib 4.0; could add either a comment or a mark.xfail to be prepared, though I do not expect a major version bump anytime soon.

[dhomeier]

Also the interface wants me to resolve the conflicts against master, while the mpl2 case is only relevant for the 4.0.x branch – not sure how to go about this.

[pllim]

@dhomeier , you might need to open 2 PRs for this:

• One against master and only for matplotlib >= 3.
• One against 4.0.x branch and for both matplotlib 2 and 3.

[dhomeier]

• One against 4.0.x branch and for both matplotlib 2 and 3.

Thanks. Since this is basically the version for 4.0.x, do you know if there is a way to change the "target" branch, or do I have to open a new PR?

[pllim]

Oh, weird. I used to be able to change the base branch right below the title, but I cannot seem to do that anymore. You might be able to create a new branch locally off the 4.0.x branch and then cherry pick this commit hash onto that branch, and then push that new branch out as a new PR?

[pllim]

Update: Okay, I found the way to change the base branch (I think they changed the UI a bit?). But unfortunately, it didn't work for your PR, as it introduced a bunch of unrelated commits. You'll have to manually do it. Thanks for your patience!

[dhomeier]

OK, creating a new PR from my branch also allowed me to pick 4.0.x, but then added the entire changeset from master, so I may have to fork off a new branch from 4.0.x.

[dhomeier]

I may have to fork off a new branch from 4.0.x.

But v4.0.x does not even have that test (last commit was 13 days ago...)

[pllim]

Err... right... it isn't backported yet... @bsipocz , any ideas?

[bsipocz]

Hmm, ok, I'll do a set of backports later today

[dhomeier]

v4.1.x does not include it either, but since rebasing on master here the MATPLOTLIB_LT_2x markers are also gone...

[dhomeier]

@bsipocz I may have become a bit confused over the milestones for this one (which is still a follow-up on #10900) – will there be v4.0.4, v4.1.1 and v4.2.1 branches in parallel now?

[bsipocz]

v4.1.x is not my business, but I don't think there will any more releases there

So, there will be a 4.0.x along with 4.2, maybe another one sooner during the RC period depending on how many fixes land in there. We already have 4.2.1 milestone as a storage for fixes that are not release blockers for 4.2. If anything there get merged before the 4.2 is out, I'll backport them and remilestone to 4.2 (this is basically the same process as it was in the past many years)

[dhomeier]

Marked no-changelog, since for this branch it just restores pre-#10900 behaviour of a test that does not have a changelog entry either.

[pllim]

Thanks!