If you want real Pythonic enums (that actually inherit from enum.Enum) you can reuse the P11X_DECLARE_ENUM macro I wrote in mplcairo.

I can't say I understand all of that macro, but I can certainly copy it over if that's the way we want to go.

(I don't really mind either way, but you mentioned that real enums may be better, so I suggested it. I can also try to explain better how it works...)

The failure in the docs actually shows the confusion with the "just a bunch of constants" approach:

Also, given that the enum values are bitfields already, not bit positions, the math there doesn't make sense.

(I don't really mind either way, but you mentioned that real enums may be better, so I suggested it. I can also try to explain better how it works...)

On second reading, I have some understanding of it all. It would be nice though if I could add a docstring to each class (as done here), and avoid the C++ enum (but I think I need this one for the variant to produce the deprecation anyway.)

I guess to add a docstring you need to do something like p11x::enums[py_name].attr("__doc__") = ...; as there's no way to set it in the function-style API.
The macro is explicitly designed to interface with a C(++) enum but you can probably get away without it if you also remove the entire typecaster machinery?

I guess to add a docstring you need to do something like p11x::enums[py_name].attr("__doc__") = ...; as there's no way to set it in the function-style API.

Yep, looks like that worked nicely.

The macro is explicitly designed to interface with a C(++) enum but you can probably get away without it if you also remove the entire typecaster machinery?

I started making a copy without the is_enum_v and hardcoding instead of underlying_type_t, but realized I still needed the enum for the deprecation, I think, so gave up on that. Would be nice to de-dupe, but at least the two copies are next to each other now.

Ah, apparently using #ifdef in macro calls is undefined, and MSVC doesn't accept it. I'll just comment out the new ones until we update FreeType.

If I've understood the remaining failure, MSVC doesn't expand __VA_ARGS__ in the P11X_ENUM_TYPE macro call to be multiple arguments, so everything falls apart after. I worked around that by just manually passing the type, unless you have any better ideas?

If I've understood the remaining failure, MSVC doesn't expand __VA_ARGS__ in the P11X_ENUM_TYPE macro call to be multiple arguments, so everything falls apart after. I worked around that by just manually passing the type, unless you have any better ideas?

You likely need to pass /Zc:preprocessor (or the older equivalent /experimental:preprocessor, which is what I use in mplcairo) (https://learn.microsoft.com/en-us/cpp/preprocessor/preprocessor-experimental-overview?view=msvc-170) in extra_compile_args (or whatever that's called for meson).

OK, I checked and /experimental:preprocessor should be old enough for us to depend on.

So this is failing on AppVeyor because we're still on the 2017 image; I opened #28869 separately to see if we can bump to the 2019 one.

We're passing now that AppVeyor has been updated.

sorry, took two tries to get the rebase right.