I committed my sndio diff together with Jag's README diff.

On Tue, 23 Jul 2024 11:24:46 +0200
Alexandre Ratchov <alex@caoua.org> wrote:

> For non-blocking writes to work, the caller must call poll(2) or an
> equivalent to wake up when the device can accept data. I couldn't find
> any poll(2) calls in the ALSA backend so I don't understand how writes
> are triggered. Does it retry periodically? If so, you need to call
> sio_pollfd(), poll(), sio_revents() before the non-blocking
> sio_write().

The write trigger is MySound_EndWrite0 in OSGLUXWN.c, which returns
true when the emulator has filled 1 buffer (with kOneBuffLen samples).
The ALSA backend might do a partial write, leave the leftover samples
in the ring of 16 buffers, wait for the next trigger, then try to
write the leftover sound plus the new buffer.

When I tried the non-blocking sio_write, the leftover samples seemed
to grow longer than the ring ("sound buffer over flow" in OSGLUXWN.c).
I heard 3 seconds of audio, because it stopped calling sio_write after
3 seconds.  I did use sio_pollfd, poll, sio_revents; but might have
made some other mistake.

I switched to the blocking sio_write because I didn't want the
leftover samples to grow.

On Tue, 23 Jul 2024 07:53:14 -0400
Jag Talon <jag@aangat.lahat.computer> wrote:

> Ok to add the following diff to the README to remove the note about
> missing audio?

I forgot to update the README.  Thank you for doing it.
--gkoehler