[SDL] depth buffer

[Ryan C. Gordon]

> before calling SDL_SetVideoMode(). Note that a 32-bit depth-buffer is
> requested - it's most likely not possible to get, but it assures that
> you'll be supplied with the best possible bit-resolution (24 bits on
> most recent video hardware).

Not true: on X11, SDL_SetVideoMode() will fail if it can't get the 
requested depth buffer (or maybe just "at least the requested depth 
buffer"). SDL_GetError() will report "Couldn't find matching GLX visual".

Other platforms will likely do the same.

SDL doesn't protect you here like it does with shadow surface fallbacks 
on 2D visuals, or automatic audio conversion.

A better strategy is to ask for what you want through 
SDL_GL_SetAttribute(), and if SDL_SetVideoMode() fails, start lowering 
your requirements and try again until you either get a GL context or you 
can't work with the limitations.

Quake 3 actually does this pretty well:

   http://svn.icculus.org/*checkout*/quake3/trunk/code/unix/sdl_glimp.c

GLW_SetMode() is what I'm talking about. Ignore the quake-specific bits 
and look at the for-loop. It tries everything at high quality, and 
failing that, it starts reducing one attribute at a time, in the order 
of importance, until SDL_SetVideoMode() succeeds or the visual wouldn't 
work out for the game.

The code itself could be cleaner with some enums and better structuring, 
but the concept is basically perfect for this.

--ryan.