List<T> is now a really ugly macro.
Added a workaround for jack.h not putting `void` in function
prototypes for functions that take no arguments. I made upstream
pull requests to jack1 and jack2 but I don't have high hopes
about them getting merged.
I removed the lock-free atomic asserts. clang reports
non-lock-free atomics when in fact it does have lock-free
atomics. I inspected the generated code for gcc and clang
for fetch_add, load, and store, on x86_64 and armhf, and
it's all lock free.
Closes#45.
* doc clarifications
* examples compile with c99 not c11
* fix pulseaudio on_backend_disconnected not firing only during
flush events
* make wait events more efficient
* fix alsa devices race condition
* fix backend disconnected code handling
* add overflow test
* fix on_events_signal not called at correct times
* refactor pulseaudio device scanning
* fix SoundIoErrorNoSuchDevice string value
This changes the semantics of the callbacks so that instead of
a single `requested_frame_count` or `available_frame_count`,
the callbacks get a minimum frame count and maximum frame count.
The callback must write at least the minimum or get an underflow.
The minimum will be 0 on ALSA, PulseAudio, and Dummy, and will
equal the maximum on CoreAudio and JACK.
This ensures optimal behavior regardless of buffer size.
