[SDL2] Build a fake SDL2 library

[madebr]

This is useful on CI to quickly build a library from scratch where we don't care about the internals.

This is related to #9580, but is not a header-only library:
it creates a shared SDL2 library that looks identical to the original, but aborts immediately when you try to run it.

I first tried plumbing this in src/dynapi/SDL_dynapi.c so it does not abort and forwards the calls to a "real" SDL2 library, like sdl2-compat does.
Perhaps @icculus has more success ;)

I'd like to use this for creating binary releases of the SDL2 satellite libraries such that they do not need to rebuild SDL2 for every VC/mingw architecture.

Description

Existing Issue(s)

[slouken]

Does it make sense to have CI download the last SDL2 release artifacts instead?

[icculus]

Yeah, this was more or less what I was envisioning in that previous discussion. I also think it would be massively beneficial to setup-sdl's build times to have a mode like this, when the CI just needs to link against something but doesn't plan to distribute it.

[madebr]

Does it make sense to have CI download the last SDL2 release artifacts instead?

I did not know an easy way to get them in a reliable way, but I think I found one.
This command echos the tag of the latest SDL2 release:

RELEASE_TAG=$(gh -R libsdl-org/SDL release list  --json tagName,createdAt,name --jq '[.[]|select(.name|startswith("2."))]|max_by(.createdAt)|.tagName)
echo $RELEASE_TAG

Then download the latest release:

gh -R libsdl-org/SDL release download $RELEASE_TAG -p '*mingw*

I'll try to use that approach now.

[madebr]

Yeah, this was more or less what I was envisioning in that previous discussion. I also think it would be massively beneficial to setup-sdl's build times to have a mode like this, when the CI just needs to link against something but doesn't plan to distribute it.

I think dynapi forwarding can be greatly simplified if the SDL API would include va_list ap alternatives of the vararg functions.
Then,

• normal functions can simply forward (as they do now), and
• vararg functions can simply forward to their va_list equivalent (this code should move into the SDL source tree)

A header-only SDL would then not need to worry about vararg functions.

[icculus]

That's an interesting idea. We should try this before the ABI locks down. If nothing else, it would remove some ugliness in dynapi.

[smcv]

Language binding authors would perhaps also appreciate it if there was always a way to get equivalent functionality without varargs, for example:

/* equivalent to SDL_SetError("%s", message) */
SDL_bool SDL_SetErrorLiteral(const char *message);