Re: [AD] get_audio_stream_buffer

[Elias Pschernig <elias@xxxxxxxxxx>]

On Sun, 2004-07-25 at 15:32 +0300, Serge Semashko wrote:
> Chris wrote:
> 
> >> As nobody else can reproduce the problem, that could probably be 
> >> compiler optimization bugs.
> > 
> > What happens if you just use plain audio streams? If there isn't a 
> > problem with using just them, then the problem is with DUMB/DUMBOGG 
> > (BTW, Ben Davis said there might be some problems between the latest 
> > DUMB and DUMBOGG.. that could be the problem too)..
> 
> A good suggestion. I have tested 'exstream.c' example included in 
> Allegro distributive. After running it for a very long time, this 
> problem shows up here too, so it is not related with DUMB or DUMBOGG.
> Sympthoms are the same, the sound changes to noise and this demo example 
> stops reacting at keyboard input (it can't be stopped using ESC key as 
> usually).
> 
> Once again, I would like anybody who is running Linux try to check if 
> this bug affects him too, or I'm just fighting with the windmills. Just 
> start exstream.c example in the evening, turn off speakers and go to 
> sleep. When you wake up in the morning, turn the speakers on and check 
> if you have the symptoms described above.
> 
> I made a quick look at Allegro sources. There is a function oss_update 
> in uoss.c. If I get it right, it is periodically called to feed data to 
> oss to produce sound. Maybe there are some race conditions and if this 
> function is called when my code is inside of get_audio_stream_buffer, 
> something bad can happen? I did not see any thread synchronization code 
> in get_audio_stream_buffer function.
> 

Could be, it sounds a lot like a race condition. Which means, it is hard
to track down. Since it seems you are the only one who can reproduce it,
coud you try a debug build, then run it in gdb, and when the problem
occurs, gdb might give some useful information.

Once the problem occurs, get into gdb with ctrl-c, and I don't know, try
"info threads" and "bt" - maybe something gives a hint what is wrong. 

(I let your last example run for 5 hours btw, and no change.)

-- 
Elias Pschernig