ASIO: fix for FL Studio ASIO quirk

[Sin-tel]

It seems like some drivers (notably "FL Studio ASIO") do not use the ASIO double buffering mechanism, and always pass a buffer_index of 0, instead of the usual 0 - 1 swapping.

This means we can not rely on buffer_index to check when to silence the buffer.

src/host/asio/stream.rs:326

let silence = current_buffer_index.load(Ordering::Acquire) != callback_info.buffer_index;

This check always fails, so the buffer never gets cleared. Output will only get added, stacking up forever.

I replaced it with a similar check, using a global counter instead that gets incremented once in the buffer_switch.

[roderickvd]

Nice catch on this driver quirk!

I was thinking though - since ASIO only supports one driver at a time, could we simplify this to just toggle between 0 and 1 instead of counting up indefinitely?

Something like:

static CALLBACK_FLAG: AtomicU32 = AtomicU32::new(1);

// In buffer_switch_time_info:
let callback_flag = CALLBACK_FLAG.fetch_xor(1, Ordering::Relaxed);

let callback_info = CallbackInfo {
    buffer_index: double_buffer_index,
    callback_flag,  // Alternates: 1, 0, 1, 0, ...
};

Then initialize the device storage to the opposite value:

current_callback_flag: Arc::new(AtomicU32::new(0)),

This would give the same behavior but:

• No wraparound concerns after 4 billion callbacks
• Only two states to reason about
• More explicit that we're just detecting "is this a new buffer switch"

What do you think?

[Sin-tel]

Yeah, a flag is simpler overall, but there's still some subtle issue with the initialization since it could be set to 0 or 1 by a previous stream. So I still set it to some other value like u32::MAX.

[roderickvd]

Thank you! I just added changelog entries and will merge when the CI gives the green light.
Then I think we're good for an asio-sys 0.2.4 and cpal 0.17 release 🎊