Discussion: Consider `typedef enum` a few types that are currently just `typedef`

[Lancelotbronner]

I'm working on Swift bindings for SDL3, the language's convention-based automatic C interop helps a lot but does the wrong thing in a few places, forcing me to redefine large sets of constants.

Since some of those are inconsistent depending on API, seems like mostly flags as well. I thought I'd still open the discussion.

I'm guessing this has to do with C's default int type for enums, but I'm hoping maybe its just that a few were missed (like SDL_keymod which definitely fits in an int).

The APIs I've encountered (or spotted in the headers) so far, no doubt I'm missing a few, are:

• SDL_PenInputFlags
• SDL_InitFlags
• SDL_MessageBoxFlags
• SDL_MessageBoxButtonFlags
• SDL_BlendMode
• SDL_SurfaceFlags
• SDL_WindowFlags
• SDL_Keycode (this one hurts a lot)
• SDL_keymod

Some additional Swift issues, which don't pertain to SDL but just for completeness sake, or in case someone magically has a fix:

• Any "complex" macros (ie. more than a literal constant) fail to import so I had to eg. duplicate logic from SDL's version macros
• All opaque pointers lose their type safety. Less type-safe than C is wild.

[slouken]

@icculus, do you remember why we switched these from enum to Uint32? I remember it was an important fix, but I don't remember why. SDL_WindowFlags needed to be 64-bit and 64-bit enum support was flaky, and SDL_Keycode is an unbounded set of codepoints, but everything else could be an enum except... ?

[madebr]

C++ emits warnings when you OR together enum values without casting.

[slouken]

Ah, thank you, that's why we did it.

[icculus]

I think once the some of the enums became defines, we switched almost everything for consistency.

I don't remember a stronger reason than that at the moment. I can try to dig up the conversation.

[slouken]

I think once the some of the enums became defines, we switched almost everything for consistency.

I don't remember a stronger reason than that at the moment. I can try to dig up the conversation.

No, it was the bitflags operator issue.