Johan & All,

Following recent mailing list discussions, i decided to implement some 
of the ideas that have been suggested recently. :-)

So i heavily modified the A2DP streaming part of the pcm plugin

The user-visible features of this patch:
   * much reduced latency (You should love this one Frederic)
   * supports headsets that are unable to throttle the data flow and
     that play too fast or drop packets or whatever (Marcel, could you
     please confirm that fixes the issues you had with some headsets ?).

This patch is fairly intrusive, and removes some code that i judjed 
useless now, like the bandwidth measurement stuff, and has greatly 
simplifyed transfer loop for avdtp (see comments in the code for the 
justifications)

It brings in a thread that has for only reponsibility to increment the 
so called hardware pointer (which doesn't point to anything in our 
case), and notify the application trough writing in a pipe that some 
room is available (virtual room again), so that it could send more data.

The attached file details the tests i've run to check i didn't break 
anything. All is conclusive except playing with XMMS that goes underrun 
for a reason i haven spotted yet. Apart from that aplay, mplayer, 
gstreamer are fine.

I'm looking forward for any questions you have on the patch.

Nest step would be to fix the SCO part of the plugin, but this one is 
looks harder as i think it will require tweaking the kernel SCO part.

Regards,

Fabien