Good catch!

We should have a bit of a think about what to do about pyplot.annoate though...

there's no problem with pyplot.annotate? it'll be subject to the same deprecation period. it's _make_keyword_only that interacts poorly with pyplot.

>>> plt.annotate(s="hi", xy=(0,0), xytext=(1,1))
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
TypeError: annotate() missing 1 required positional argument: 'text'
>>> plt.gca().annotate(s="hi", xy=(0,0), xytext=(1,1))
__main__:1: MatplotlibDeprecationWarning: The 's' parameter of annotate() has been renamed 'text' since Matplotlib 3.1; support for the old name will be dropped in 3.3.
Text(1, 1, 'hi')

With this patch applied, passing an explicit s kwarg to pyplot.annotate does cause a TypeError.
The function is not decorated with the _rename_parameter decorator.

good catch :/
I guess this is another argument in favor of #14130 (comment) (@tacaswell do you want to comment on that?)

if the general problem doesn't get fixed quickly you can always go with manually writing the pyplot wrapper as a stopgap (#14130 (comment)).

I'm willing to implement at least the stopgap mentioned by @anntzer, if that is what we want for now

I was just reminded of this, I've seen some PRs in recent days/weeks regarding annotate and some of the renaming of params and such, but did not track them closely enough to see if they were fixing this problem in another way, enabling this PR to work as intended, or just actually unrelated to this change entirely.

So, just checking in, seeing if there is anything that needs to be done to this PR to catch up (it appears to me that the docstring and the method signature are still mismatched as of now)

There are no merge conflicts, so thats usually a good sign that no one has touched the section of the code you have touched. OTOH, you have some documentation failures, and should rebase with master to make sure all the CI is using recent CI.

Technically this breaks plt.annotate(s=...) without a deprecation, and the fix would be to merge #15254. If we decide to ignore the APi break, I agree (some version of) this should go in ASAP.

CI Failure looks unrelated to me... (Looks like two image files just failed to write properly, nothing to do with the actual code difference here)

I did test out #15254, the change proposed here is still needed (to Axes.annotate), but that PR does prevent the API deprecation problem. (There is a small merge conflict, as both edit pyplot.py, but it is only in the generated portion of that file, anyway)

Closing and opening to rerun CI.

Rebased on top of master after #15254 was merged

Rebased on top of master to avoid merge conflict