pyplot.subplots() accepts subplot arguments through **kwargs - #16738

Wow, that is an old issue!

I share @QuLogic 's concerns about hard-coding a sub-set of kwargs to be hoisted up to subplot_kw. Right now, there is a clear seperation: subplot kwargs go in a dict, fig_kw go as "normal" kwarg so there is no ambiguity over where the input is going to get routed. By special casing just some of the subplot keyword arguments we are setting our selves up bugs in the future of the form "This kwarg X works, but kwarg Y doesn't".

This also has the logical extension that we should route some keys to gridspec_kw if we know they should go there which would make everything more complex.

In short, I am worried that half measures will make things more confusing and inconsistent and the full solution may grow too complex.

I checked add_subplot()'s documentation and it mentions "projection", "polar", "sharex", "sharey", and "label". In pyplot.subplot()'s signature the "sharex" and "sharey" arguments are explicitly passed, so I just extracted the three remaining ones from **kwargs.
@tacaswell Just to make sure I understand the issue correctly: The issue with this is that it's not extendable and everytime the arguments need to be changed, the list needs to be updated as well. Also extending this idea to gridspec_kw could make things more complicated. Correct? So your suggestion is to look for a more fundamental solution?

@sabulikia That is a reasonable summary.

On one hand add_subplot, in addition to the keywords it consumes, also pass**kwargs through to the subplots class factory. This means we can never fully get rid of subplots_kw (as it is technically possible for the user to provide their on Axes sub-class (which something like windrose [1] does). On the flip side plt.figure also takes **kwargs and passes them onto new_figure_manager which also take **kwargs and pasess them on. Thus we can't do the inverse and identify all of the kwargs that should go to figure by name (and hence bundle the rest of them into the subplots_kw).

The fundamental issue is we have two things that consume all of the "extra" keyword arguments and are user-plug able. I do not think there is a more fundamental solution possible here, but I would be happy to be proved wrong!

It seems simpler to say "input to subplots goes here and input to Figure goes there" than to add "...except for XYZ". Any one exception is easy to remember, but en-masse they are huge sources of friction.

[1] https://github.com/python-windrose/windrose