Milestoning 3.1 because it would be nice if this would make it into the docs soon. But feel free to re-milestone.

I actually think we should just deprecate Figure.show() in favor of e.g. plt.show(figures=[fig1, fig2]) (if you really want to raise just a subset of figures).

Figure.show only works for pyplot-managed figures, so it makes sense to have that functionality in pyplot, plus having it it plt.show() will properly spin up the event loop if needed.

If you are managing your own GUI then hopefully you are not using pyplot (...), in which case fig.show() is not useful.

Given #13101 (comment), can we really deprecate Figure.show()?

Not an expert on this, but there seem to be two differences betweeen the two show() functions:

1. Rasing all vs. only one figure
2. Managing an event loop.

plt.show(figures=[fig1, fig2]) only addresses 1.

But plt.show(figures=[fig1, fig2]) would also reuse the code that already exists in plt.show() to manage the event loop.

I think the improvements to plt.show can be done separately.

Currently plt.show calls show from the backend bases template which calls manager.canvas.figure.show() which then calls manager.show() so there is clearly some room for simplifying that call chain.... We will never be able to deprecated manager.show() as that is typically inherited from the underlying GUI widget.

I'm not talking about deprecating manager.show() (which is indeed needed as a backend customization point), just figure.show().

Thanks, I think this will be helpful 👍