Strongly suggest using layout='constrained' for similar figure layouts.

Strongly suggest using layout='constrained' for similar figure layouts.

Yes, I know the constrained layout works fine with subfigures. But for my "real" use case (which requires different subfigures have some of their subplotpars synced so that their axes align), I found it was more straight forward to to call tight_layout and then to tweak subplotpars. If this is something we can already do with the constrained layout, I will be happy to learn how.

You can trigger a draw and then turn constrained layout off if that helps?
https://matplotlib.org/devdocs/users/explain/axes/constrainedlayout_guide.html#manually-turning-off-constrained-layout

And is there a way to align axes in different subfigures after that? With tight_layout, I run tight_layout for each of subfigures (they are set to have independent subplotpar attributes), take the maximum value ofsubplotpar.left of all subfigures, for example, and then apply that value to all subfigures.

By the way, independent of this can be done with the constrained layout, I think it is worthwhile to make TightLayoutEngine to support subfigures, given the fixes are rather straightforward.

It would be easier to see what you are asking with an example.

There is definitely no harm in making tight_layout work with subfigures if it doesn't require a lot of new knobs.

I think this also needs a test. I'm still not entirely clear what you are proposing the public API for this is? just attaching an engine to a single subfigure and running tight_layout on the subfigure?

For the public API, I am open to any suggestions. I personally think it is not necessary to introduce new public api. I don't expectTightLayoutEngine to be as robust asConstrained LayoutEnging for subfigures. On the other hand, it would be good if theTightLayoutEngine handles a figure with subfigures better, at least for simple cases.

I will push some more changes, including tests.

@leejjoon did you want to push forward with this?

For your use case, as I understand it, I would personally juts keep the axes in the same layout versus having them in subfigures. The idea of subfigures is that the layouts between the subfigures are independent. Of course if you are after a very specific layout - say with a larger column separation between columns of axes, then maybe subfigures+subplotpars is a reasonable way to go. Though even for that use case, I'd probably use width_ratios and blank subplots:

import numpy as npimport matplotlib.pyplot as pltfig, axs = plt.subplot_mosaic([['a', 'b', '.', 'c', 'd'],                         ['e', 'f', '.', 'g', 'h']                         ], width_ratios=[1, 1, 0.25, 1, 1],                         layout='constrained')for ax in axs:    axs[ax].set_xticks([])    axs[ax].set_yticks([])plt.show()

However, I appreciate that there are many other reasons you may want to use subfigures.