Environment SDL_AUDIODRIVER may have invalid values

[ericoporto]

If someone has used an SDL1 software or driver, on Windows, they may have set an environment variable of SDL_AUDIODRIVER to dsound, but it now has been renamed as directsound. Since the environment variables don't have a version number SDL2_..., you may hit an audio driver system that doesn't initialize.

https://www.libsdl.org/release/SDL-1.2.15/docs/html/sdlenvvars.html

It would be useful if SDL_InitSubSystem(SDL_INIT_AUDIO) would not fail in an invalid environment variable, and instead attempt known good audio drivers before failing - like wasapi, directsound and winmm in Windows, but not dummy or disk.

[sulix]

We do handle the difference between "pulse" and "pulseaudio" as of #4693 — perhaps something similar could be done (though avoiding directsound may be worth doing on more modern systems…)

[ericoporto]

It would be good if SDL3 has these versioned to avoid clashing (e.g.: SDL3_AUDIODRIVER )

I was sketching this for AGS: ericoporto/ags@30b7185

But it felt like it was supposed to be in SDL2 instead. MojoAL and other libraries don't have a way to reuse a device because they also pass additional information when opening for things like frequency, which gets checked by SDL2. Essentially I would like SDL2 to only fail if neither of the sensible audio drivers works - this may be a different issue, but just thought I should inform the context.

though avoiding directsound may be worth doing on more modern systems

I think people are avoiding WASAPI because of #5538 , since the bug reports about this environment variable came from people with discreet audio cards.

[sezero]

(though avoiding directsound may be worth doing on more modern systems…)

sdl12-compat probably needs that

[slouken]

We do handle the difference between "pulse" and "pulseaudio" as of #4693 — perhaps something similar could be done (though avoiding directsound may be worth doing on more modern systems…)

Yep, I'll go ahead and implement this.