Sorry, I retract my comment, it does not seem to work reliably even with this change.

The critical change was to place the drawing logic inside the if self.got_artist part. This is because if someone clicks on a line in the legend it triggers two on_pick events: one for generally clicking on the legend and one for clicking on the line. In the previous code it would only draw once evt.artist == self.ref_artist was true, which was only true in the case of the legend, so the first event. Now, it handles the legend event, which sets self.got_artist = True like before, then as self.got_artist is True it now also handles the line event.

But in which scenarios did it not work for you? I have tried testing it again with the code example provided in the original bug description and can't spot any issue

Ah, I sorted out what happened.

I changed the fig.canvas.draw() in the example code to fig.canvas.draw_idle() and the flashing came back.

Ah okey I see, was that the only change? Because it seems to still work for me even when using fig.canvas.draw_idle()

Okey I did notice that with fig.canvas.draw_idle() the legend disappears when spamming on a line. Adding draw/blit to the on_release seems to fix it

... the legend disappears when spamming on a line.

I drink too much ☕ !

Hi! Did any maintainer maybe get the chance to test this?

Given that this is code we do not have unit tests for, and the docs failure is due to the config changing on main but not getting propogated to correctly (this is understood...we build the runner based on the config on the branch then manually create the merge commit to build the docs which leaves the requirements assuming a different base version of Python than the runner has) I'm going to go ahead and merge this before CI finishes.

Thank you for you work on this @vittoboa and thank you for sticking with it to find the a better solution than the initial one!

Congratulations on your first merged Matplotlib PR 🎉 I hope we hear from you again.