Honestly it's not clear to me that autoselecting papersize (and thus changing the figure size) is a desirable feature? (Not to block the improvement here...) We could perhaps support e.g. figsize="a4" or similar and deprecate/remove papersize autoselection?

I agree. More like you pointed out in #7551: the selection logic is most likely incorrect. So it may not be the best way to support automatic selection, but now it at least works as one might expect it to.

I have not looked very deep into this, so I do not really know when the paper size is and when it is simply ignored... But I agree that it may be better to see if one can find a better approach to deal with this. (And if none has shown up for 3.6, one can revisit this.)

The fact that it has never worked properly (likely for many many years) suggests that no one is really relying on this, though...

I'm inclined to merge something like this. Independent of if this is a good idea, we should at least do it in a non-broken way.

The fact that it has never worked properly (likely for many many years) suggests that no one is really relying on this, though...

Be careful here. It can well be that people have a specific figure size that selected a reasonable (maybe not optimal) page size. Or they might have set up some post processing to fix the size later. Changing, even if it is improving, would break those people. We should do this only once. Now changing the detection algorithm only to abandon it in the next release can cause extra migration pain. Let's decide once how this should be handled and stick with it.

@timhoffm Fair. I take back my statement.