Thanks @dswelch. Have you checked what would happen if you had more than two rows or columns? From a quick look at the code, I think the outer rows/columns might end up wider than the middle ones.

I’m also wondering if there’s a reason we can’t just use the subplotspec’s get_position method.

Hi @rcomer, thank you for the feedback! You're correct that it does not work in the case of more than two rows or columns. Could you elaborate where I would use the get_position method? When experimenting with it by just returning the results of get_position, I got these results, with the constrained layout on the left and non constrained layout on the right:

This is the same example that I used from the original issue, with wspace and hspace set to 0.2. I don't believe I fully understand what this function does so any guidance would be appreciated.

Thanks for checking.

As far as I can tell, the get_position method is designed to do what we need except for subplots rather than subfigures. However, I had missed that the underlying GridSpec.get_grid_positions relies on using Figure.subplotpars to get the left, right, wspace, etc. properties. So that does tie it to subplots, and is the “reason we can’t just use the subplotspec’s get_position method.”

I just looked again at this and I realised that GridSpec.get_grid_positions doesn't have to rely on Figure.subplotpars provided the relevant attributes are set on the gridspec itself.

So perhaps using get_grid_positions would work, but only if we originally set up the gridspec to not have any white space around the edges. I.e. pass left=0, right=1, bottom=0, top=1 here:

@dswelch are you still interested in working on this and if so do you want to try that? I can't guarantee it will work!

@rcomer I'm sorry for the late response, I would like to open up this issue to others if possible. If there is anything I should do to make that easier please let me know.

Thanks for letting us know @dswelch. I’ll add the “orphan” label, then someone might adopt it.

After some more discussion at #25511 I opened #25960 with a slightly different approach. So I will close this one in favour of that. Thank you for your work on this @dswelch. This one turned out to be somewhat thorny problem. I hope that that has not put you off and that we hear from you again some time.