Just a braindump from an IM conversation with Sam, for changing the audio conversion API for 1.3. 

Ryan C. Gordon
Jabber
12:27
as for conversion code:
12:27
Ideally, it looks something like:
12:28
int SDL_ConvertBufferSize(SDL_AudioCVT *cvt, int input_buffer_size);
12:28
 returns number of bytes you need to convert input_buffer_size bytes to cvt format.
12:28
and then a static buffer converter.
Sam Lantinga
GTalk
12:28
Hmm, I like that.
Ryan C. Gordon
Jabber
12:28
But also something like what zlib does:
12:29
you feed it data, and say "i want X bytes now" and it gives you more, or tells you to feed it more.
12:29
It's a little more high level, but it's pleasant for streaming.
Sam Lantinga
GTalk
12:29
We'd probably want size functions both ways
12:29
SDL_ConvertBufferInputSize(), and SDL_ConvertBufferOutputSize()
Ryan C. Gordon
Jabber
12:29
The second approach requires SDL to buffer some data between calls.
Sam Lantinga
GTalk
12:30
Well, the AudioCVT structure could be changed to stream
12:30
maybe
Ryan C. Gordon
Jabber
12:30
But honestly, when you have to do this with zlib, it's actually really nice for the caller.
Sam Lantinga
GTalk
12:30
Yeah
12:30
That sounds awesome!