http://bugs.winehq.org/show_bug.cgi?id=28723
--- Comment #109 from Jörg Höhle hoehle@users.sourceforge.net 2011-12-21 09:29:43 CST --- Michael, thank you. Your report is very dense so I've many questions.
restart PA
Although I've mentioned PA several times in this issue, I want it to not be about PA. Thus I'd be grateful if you could take PA out of the equation. I nevertheless want (even old) PA to work.
Andrew, what are the current mechanisms to select a particular device in winealsa? Back then there was the registry, nowadays I'm editing mmdevdrv.c:defname[] by hand to e.g. "plug:dmix" or "plughw:0,0". If there's none, that's a bug. There should be one, e.g. to select surround devices or whatever ALSA has to offer.
Do we have a specific Pulse Underrun bug entry? Not bug #10495
but most of the time the fuzz is constant once it starts
That means that my patch "2x snd_pcm_drop cause PA" is not effective enough.
#4 vs #5 fuzz differences What happens if you take the 2x snd_pcm_drop patch out of #5? What if you include it in #4? Is that what causes the differences?
restart PA for Wine (restarting Wine is not enough).
I've not found out a) when PA needs a complete restart; b) when snd_pcm_close+open("default") is enough (= other apps still sound fine) c) when 2x snd_pcm_drop suffices... b) and c) may be very close internally. In bug #25750, comment #6, Raymond mentions a PA disconnect/reconnect cycle.
I assume you use the newest PA + alsa_plugs >= 1.0.24. It's interesting that you still mention trouble. I'm using the outdated PA from Ubuntu Intrepid that is known to sometimes loose sound, sometimes after as much as 1h of play.
since sound changes in .30
What do you mean? Removal of DSound hw acceleration?
Would recording the sound
I don't care. Describing is enough to get a "picture". What helps is if you can state whether each and every trouble correlates with an underrun log. Then we'd need to know if the underrun was caused because audio was already disfunctional shortly before (normal padding&delay/position values) or whether high system load caused it.
Nevertheless, in this bug entry I'd prefer talking about how to support the Rage scenario - 10ms prefill (or even 0 prefill) - at most 10ms at each period event with a variety of ALSA devices, from hw-near ones with tiny buffers, to PA-alikes with huge ones and accessorily about underrun handling, or more generally, fault resilience, incl. detecting stalling of audio output.