[Bug]: Crash when removing a subfigure with a subplot in a figure #31319

Activity

1. tacaswell

   added

   status: confirmed bug

   on Mar 17, 2026
2. tacaswell commented on Mar 17, 2026

   

   tacaswell

   on Mar 17, 2026

   Member

   More actions

   The problem is that we everything inherits from Artist (including the Figure family of clases), but the type of obj.axes is not stable across all classes: for the Figure family it is list[Axes], for Axes it is None, and for everything else it is None | Axes [Yes this is non-ideal, no we would not do it this way if designing from scratch now, no it is not worth the deprecation cost to change it at this point]. Additionally we customize the remove functionality by setting a callback at run time. Thus, the assumption in the base Artist.remove is that if the artists have a remove callback set and the self.axes attribute is not falsey, then we are in the third category so we can treat the self.axes as an Axes object and need to clean up Axes stuff that was set in add_artist.

   Looking at this I see a couple of paths:

   • update the _remove_method callback for sub-figures to recursively remove their axes/subfigures so that by the time we hit the code is Artist.remove self.axes == []
   • update Artist.remove to bail on self.axes being a list
   • update Artist.remove to bail on Figure subclasses of self
   • update Artist.romove to only not bail on Axes of self.axes
   • add a custom remove method someplace in the Figure family

   None of these are super great, but I think overriding the remove method is the worst choice (mixing methods of injecting logic always makes things worse). Making sure SubFigures recursively clean them selves will make removing them a bit more expensive and will complicate some future plans but might help with releasing memory (breaks reference cycles) and is more consistent with the current behavior else where (e.g. this code block where we are making sure that for a data artist (Line2D) that is being removed from an Axes that they get fully decoupled). Changing the test from the truthyness of self.axes to a type check + not None check is probably the minimal change that will fix the problem with the fewest side effects.
3. tacaswell

   added this to the v3.10.9 milestone on Mar 17, 2026
4. QuLogic commented on Mar 17, 2026

   

   QuLogic

   on Mar 17, 2026

   Member

   More actions

   Oh, that's unfortunate. We actually have a test for SubFigure removal, but it's an empty one:
   

   matplotlib/lib/matplotlib/tests/test_figure.py

   Lines 1575 to 1579 in dfa19f6

   	def test_subfigure_remove():
   	fig = plt.figure()
   	sfs = fig.subfigures(2, 2)
   	sfs[1, 1].remove()
   	assert len(fig.subfigs) == 3
5. Rahasyapandey

   added a commit that references this issue on Mar 18, 2026

   Fix AttributeError in artist.remove for SubFigures (matplotlib#31319)

   ...

   3ccb873
6. rcomer

   closed this as completedin #31323on Mar 27, 2026