Flag enum types and flags typedefs

[kainino0x]

Originally posted by @mrshannon in #158 (comment):

Any idea on how to improve WGPUBufferUsage vs. WGPUBufferUsageFlags?

You technically don't need WGPUBufferUsageFlags as WGPUBufferUsage will > always be wide enough for the bitwise combination because we have WGPUBufferUsage_Force32 = 0x7FFFFFFF which is effectively doing what is done for the C++ headers with:

enum class BufferUsage : uint32_t {

This is strictly true, but in Dawn's C++ bindings we also have operator overloads, so when you do wgpu::BufferUsage::CopySrc | wgpu::BufferUsage::CopyDst you get a wgpu::BufferUsage value. In C WGPUBufferUsage_CopySrc | WGPUBufferUsage_CopyDst will give you an int of some kind. So this would force callers of the API to cast back to the appropriate type.

desc.usage = static_cast<WGPUBufferUsage>(WGPUBufferUsage_CopySrc | WGPUBufferUsage_CopyDst);

Apparently, the cast is only needed in C++. C seems happy without a cast. So I guess we could define operator overloads for these types behind #ifdef __cplusplus...?

[kainino0x]

i.e.

#ifdef __cplusplus
WGPUBufferUsage operator|(WGPUBufferUsage a, WGPUBufferUsage b) {
    return static_cast<WGPUBufferUsage>(a | b);
}

and so on

[kainino0x]

FWIW, we copied this pattern from Vulkan:
#1 (comment)

in any case, whatever we do, we need to make sure it works fully in both C and C++.

[Kangz]

Adding operators only in C++ if that's where it causes problems would make sense. We could have something similar to Dawn's EnumClassBitmask.h.

in any case, whatever we do, we need to make sure it works fully in both C and C++.

Yeah right now webgpu.h doesn't work in C because of #84 even.

[kainino0x]

A Dawn contributor just so happened to find a pattern just like this in mingw's winnt.h: https://github.com/mirror/mingw-w64/blob/1de9cc347da5215889f6e8ff3bd4bcc820aa50bf/mingw-w64-headers/include/winnt.h#L682

#ifdef __cplusplus
#define DEFINE_ENUM_FLAG_OPERATORS(ENUMTYPE) \
extern "C++" { \
inline ENUMTYPE operator | (ENUMTYPE a, ENUMTYPE b) { return ENUMTYPE(((int)a) | ((int)b)); } \
...

[kainino0x]

meeting: agreed to do that, to patch over the difference between C and C++.

However I forgot that the original point of this issue was whether we wanted to get rid of the *Flags types, as we don't need this change otherwise (in C++, enum | enum -> int). I think we should get rid of the Flags types.

[kainino0x]

Obsoleted by #273 due to replacing all bitflag enums with uint64_t.