TST/FIX twinx and twiny w/ constrainedlayout

[jklymak]

PR Summary

As per #10404, the twinx/y code in _base.py for constrainedlayout was not being called in any of the tests, and hence an import was inadvertently erased. This adds the import and two tests. I also tested interactively, and everything works as it should to window resizing.

PR Checklist

• Has Pytest style unit tests
• Code is PEP 8 compliant

[QuLogic]

It's too bad this was not noticed before, but you should probably use style='default' for any new test images.

[jklymak]

Do you mean style='mpl20'? Confused why I'd set it to the default (presumably not specifying is the default).

[QuLogic]

Not specified defaults to classic, or we'd have to recreate all the test images. mpl20 is also good, and maybe better to be more explicit.

[jklymak]

I can do that for the two that aren't merged here. Happy to do it for the other ones as well, but there will still be dupes in the repo.

[jklymak]

Actually, removed the images. Not really needed for the test, which is just that the axes track each other....

Happy to change the other (already committed) image tests if thats warranted.

[QuLogic]

fig.canvas.draw() is what's used elsewhere.

[QuLogic]

The images are already there, so I guess there's no saving in changing the style now. Just best to remember it the next time they're updated.

[QuLogic]

codecov says the line with the undefined name is not getting hit with these new tests.

[QuLogic]

I can run the tests without the import and it still passes, so it seems they are not checking that code just yet.

[jklymak]

Great catch @QuLogic Thanks a lot. That code path was never getting called, because only the twinx logic in _subplots is used. Guess thats why I never noticed there was no import 🐑

What really has me flummoxed is that twinx works w/o that logic. I added to subplot anyway, because I think its important to keep a layoutbox with each axis. But it never got into the situation where that was causing a problem.

[jklymak]

Small fix - the layout box in the twinned axis was already being made, so this just constrains it to be the same size as its twin. This was already happening due to logic @efiring added, but this makes it explicit in the layout constraints, which seems a good idea if things get complicated.

[jklymak]

Thanks again for your help @QuLogic I'd missed that that code wasn't even being called.