Many people agree that ALSA is probably the biggest weakness of the GNU/Linux operating system. ALSA is massive, poorly understood, and badly documented. Some design decisions that it took early on also proved to be fatal. The reason is simple: historically, soundcards didn't behave predictably (in fact, some of them included hardware mixers while others didn't), and a lot of drivers had to be reverse-engineered.
Today, we're stuck with a historical relic coupled with PulseAudio, a half-baked software mixer. My question is simple: if we could rewrite ALSA today (from scratch is probably too hard, but atleast reusing parts of other software), how would it be different from the current ALSA, design-wise?