SubFigure.get_figure does return a Figure, even if the subfigures are nested.

In [6]: sfig2 = sfig.subfigures()

In [7]: sfig2.get_figure()
Out[7]: <Figure size 640x480 with 1 Axes>

This seems inconsistent and likely we haven't thought about it when subfigures were introduced.

Intuitively, I'd recommend the following behavior: get_figure() returns the closest (sub)figure. If we were to go all the way up, it's not possible to get to the next level only.

That would imply:

• Change documentation/typing of Artist.get_figure
• Change behavior of Subfigure.get_figure to return the containing (sub)figure. - Note that get_figure only returns self.figure, so self.figure should be changed for nested subfigures. I consider it a bug that this is not the containing subfigure. @jklymak do you agree?

Hmm that seems inconsistent at least so one way or another is a bug

It seems that a subfigure wants to know its parent rather than its root?

Ooops sorry crosspost w above. I'd have to look at the code a bit.

This seems inconsistent and likely we haven't thought about it when subfigures were introduced.

Intuitively, I'd recommend the following behavior: get_figure() returns the closest (sub)figure. If we were to go all the way up, it's not possible to get to the next level only.

That would imply:

• Change documentation/typing of Artist.get_figure
• Change behavior of Subfigure.get_figure to return the containing (sub)figure. - Note that get_figure only returns self.figure, so self.figure should be changed for nested subfigures. I consider it a bug that this is not the containing subfigure. @jklymak do you agree?

Hmmm, I fear it's not so simple. I explicitly set self.figure = parent.figure which therefore resolves to the root figure, and I think that is because artists have things like self.figure.stale and self.figure.suppressComposite which would somehow have to magically thread themselves through.

So I wonder if the correct course of action is to override get_figure to return the parent figure, but keep self.figure as-is. In hindsight, it should have been private, I think?

So I wonder if the correct course of action is to override get_figure to return the parent figure, but keep self.figure as-is. In hindsight, it should have been private, I think?

It's bad enough that we have a mixture of attributes and getters, having both with slightly different semantics is even worse. I propse to

• expand get_figure() with a parmeter to select between the direct parent and the top-level figure, choosing the appropriate default. - The default should become the same for artists and subfigures, which means an API change for one of them.
• create whatever internal state is necessary to support the behavior (likely self._parent_figure)
• keep self.figure with the current semantics, but discourate (deprecate?) it's use in favor of get_figure() - if needed turn it into a property.

The above seems good to me.
Something like:

get_figure(root=False)

root : bool
    If False return the (Sub)Figure this artist is on.  If True, return the root Figure if there is a nested tree of SubFigures.

Ah, this also make sense as to why the subfigure_mosaic has problems with getting the nesting right.

Strongly agree we need a version that gets the parent rather than the root.

keep self.figure with the current semantics, but discourate (deprecate?) it's use in favor of get_figure() - if needed turn it into a property.

FWIW deprecating .figure would be extremely disruptive for me -- I use this API all the time, typically because I only keep track of axes, and go back to the figure as needed via ax.figure...