Major mmdevapi and winmm audio bugs

[Joerg-Cyril.Hoehle at t-systems.com]

Andrew Eikum wrote:

>And, not too surprisingly, TimerQueue isn't sufficient. I've attached
>my tests here. The tests ask to execute the callback every 12 ms. On
>my Windows 7 VM, I found that it executed with intervals like:
>20, 10, 10, 10, 10, 20, 10, 10, 10, 10, ...
>With an 11 occasionally interspersed.
>So, forget that.

No. Did you notice that on *average*, a callback occurs every 12ms (5 within 60ms)?
This is the same essential property that winmm:setTimer exhibits too.
Wine's CreateTimerQueue is buggy, because it doesn't do that.
Wine's winmm:timer does.
A patch is needed for CTQ.

Now if Wine's CTQ would be fixed, Wine's audio would have fewer underruns.

The periodicity is another topic.

>my Windows 7 VM
I don't trust timings from a VM.  10/20ms may be an artefact. OTOH it may be
a design decision from MS related to scheduler quantums or whatever.

>Since we're already using poll() in winmm and it seems to work well,
>that would be my suggestion.
Sure, but then you need an own thread in mmdevapi, don't you?
So we're back at the point where I said earlier that I can well imagine a fix
to CTQ for wine-1.4.0, but less so a move to an own thread + poll.

Regards,
 Jörg Höhle