There was already an attempt of making the block argument explicit in #11170 It didn't work out for some reason. Maybe it'll help to get it straight here though.

I didn't know about #11170. It doesn't seems clear what happens when reading that though. So the problem might be in one or several backends?

It looks like the exact same issue in the failing tests, if I recall that correctly, which is about the WebAgg backend. See https://github.com/matplotlib/matplotlib/pull/11170/files#r186467036.

Yes, the WebAgg doesn't seem to work but isn't it solved by the the line that tacaswell comment? I try that now.

I don't know much about the different backends and how the backend class hierarchy works but to me it seems that the problem is that the show in backend bases take a keyword block but that this function is overridden in the WebAgg class with a show that takes no arguments.

WebAgg is taken care of in the backend_bases show.

I added a simple test of the input but I am not sure about the docstrings suggestions (or my own versions for that matter). I think that the explaination in the docstring that was suggested where to short.

I think that my docstring proposals contain the same amount of information, they are just less verbose. But a third opinion would be good.

A third opinion was being asked for: I can confirm that the two proposed changes by @timhoffm do indeed contain the same information.
But I can see how the initial proposals are easier to understand by inexperienced users. (On the other hand, more experienced users may be irritated by the verbosity compared to the ususal doc style.) In the end it may be a matter of taste and consistency.

This has been superseeded by

• Warn when trying to start a GUI event loop out of the main thread. #15504 addresses the signature of show by dynamic rewriting.
• DOC: Start to document interactive figures #4779 and Update plt.show() doc #14224 update the docstring.

@fredrik-1 sorry this stalled. Anyway, thanks for the contribution.