I'm not sure whats going on, but ignoring the spines in the tight_layout is probably not a good solution. I'll look into it later.

I wounder if it is a similar issue as #10989?

The plot actually looks ok again if tight_layout() is called one extra time.

Ignoring the spines are probably not a good idea but the tightbboxes for the spines are not correct when calculated in axes.get_tightbbox but are correct when calculated from ax from the python prompt.

Sorry on a phone right now but I am pretty certain I addressed the spine issue in a recent PR. But maybe just for constrained layout and not tight layout? Look at my recent closed PRs.

OK, see #11627. This is the same problem, caused by putting all child artists (including the spines) in the tight bbox calculation (#10682).

The problem with spines under zoom is that they don't get their transform updated in time for the tight_bbox calculation. That is the fundamental problem.

The issue with your fix is that if the spine is displaced from the axes position it will not be included in the tight_bbox and the subplot spacing won't be adequately large enough.

So, this bug all makes sense. I'm not sure what the fix for the spines not having the correct transform is, but that'd be the right thing to do.

The other thing we did for constrained_layout is not have it adjust the axes sizes if zoom or pan were on. Thats actually something that would make sense to me for fig.set_tight_layout(True) We don't want to keep changing the axes size while zooming is happening. But, others may disagree.

I updated my fix to just be used when tight_layout=True.

This is mostly a hack but the behavior is probably similar to the behavior in 2.2.2 so it shouldn't break peoples working process. I rather want this behavior than the behavior in #11742

I changed the order in figure.draw such that tight layout is calculated after the draw of the other artists.

The behavior is quite good and what I would expect in most cases but the tight_layout calculation is sometimes still slightly incorrect.

This won't position the axes properly until the second draw...

I see now that something doesn't work. It seems to work ok with one subplot but not with several.

But the thing I am after is that I think that tight_layout should be applied after the first draw and redraw the figure. But I can't find a way that does that...

OK, #11754 fixes the root of the problem, I hope.