_type_ class attributes in ctypes, and fix ctypes.wintypes.BYTE

[jorenham]

This adds the _type_ class attributes to the relevant types of ctypes. It makes the individual c_* types structurally distinguishable. For the integer and floating C types, the C99 standard is assumed: https://en.cppreference.com/w/c/language/arithmetic_types#Integer_types as reference

But why?

A Protocol can now be used to e.g. only accept C types could have _type_ == "l", i.e. signed integers of at least 32 bits on 32- and 64-bits systems. This will match c_int, c_long, c_longlong, c_int32, c_int64, c_time_t, c_ssize_t, and on windows also HRESULT. The alternative would be to use a platform-specific union of these 7 or 8 types, which would be rather messy and verbose.

Also, because stubs are often used as form of documentation (or maybe that's just me), this will also help in that sense.

[jorenham]

Another advantage: stubtest is now able to recognize that BYTE = c_byte in ctypes.wintypes is incorrect :)

[srittau]

There's a merge conflict now, but apart from that LGTM – for my limited ctypes experience.

[jorenham]

but apart from that LGTM – for my limited ctypes experience.

@seberg you're pretty C-savvy; does this look right to you?

[brianschubert]

Would it make sense for these to be Final?

[jorenham]

Would it make sense for these to be Final?

Good question. I guess that partially depends on whether mixed ClassVar and Final generally supported by the major type-checkers yet.

[github-actions]

According to mypy_primer, this change has no effect on the checked open source code. 🤖🎉