Additionally, we are now python 3.10+ on main anyway, so not extensions is not actually needed for Self.

Note however that to be backported to the next micro release, we would need to take care there, since it still runs on py 3.9.

Additionally, we are now python 3.10+ on main anyway, so not extensions is not actually needed for Self

typing.Self only became part of the standard library in 3.11

Ah, I remembered Self coming in (at least added in some places) with recent changes (#28503) that were broadly related to py 3.10 being the new minimum.

Regardless, relying on typing_extensions inpyi files (or guarded byif TYPE_CHECKING) is not new here, though perhaps we could be more explicit about it somewhere.

I do not think we want to make it aruntime dependency necessarily.

All that said, in this instance, I don't actually thinkSelf describes what actually happens...

For twinx/twiny, the Axes that is returned is created byFigure.add_subplot orFigure.add_axes, which both returnAxes, specifically.

This means that if you calltwinx on a Polar plot (or geographic projection or 3D plot, etc) you get a "normal" axes back... (See#12506).

Thus the actual correct type hint isAxes here, notSelf:

Created here:

Type hints for those methods:

This generally falls into the category of_AxesBase is nottotally wrong...Axes is a subclass thereof, butSelf is also not correct when you go further down the inheritance tree. That leaves us with a weird circular reference problem, but I don't think that should actually be a problem in pyi files.

However... I thinkSelf may be the "ideal" type hint, just not actually reflective of the code as it stands...

This means that if you calltwinx on a Polar plot (or geographic projection or 3D plot, etc)

It's generally not clear what twinning should mean on arbitrary axes or that it's useful. For example, where should the second radial axis on polar be placed? Polar does not even have a second radial axis. While it would be technically feasible for 3d and geo axis, the resulting plot is likely very difficult to interpret.

For now, I'd pragmatically say that we just don't provide proper twinning for these niche cases and set the return type toAxes.

While it would be technically feasible for 3d and geo axis, the resulting plot is likely very difficult to interpret

Um this has me realizing that a twinx geoAxis that shows the 0-360 convention and the 180W-180E convention could be useful for teaching both 😅

But also that's functionality that probably could live in cartopy if anyone wanted it.

Thanks@randolf-scholz! Congratulations on your first PR to Matplotlib 🎉 We hope to hear from you again.