On 01/12/2013 04:46 PM, Tanu Kaskinen wrote: > On Sat, 2013-01-12 at 10:10 +0100, David Henningsson wrote: >> On 01/11/2013 08:18 PM, Tanu Kaskinen wrote: >>> On Fri, 2013-01-11 at 14:04 +0100, David Henningsson wrote: >>>> This fixes a bug where pulseaudio would give up the device (due to >>>> a request from JACK), but then immediately grab it again because >>>> the monitor callback fired, telling that the device is now available. >>>> >>>> (Note: the protocol does not specify a timeout, i e if pulseaudio >>>> is requested to give its device up but JACK does not grab the dbus name, >>>> at what point is PulseAudio allowed to re-grab it?) >>>> >>>> Signed-off-by: David Henningsson <david.henningsson at canonical.com> >>>> --- >>>> src/modules/reserve-monitor.c | 30 +++++++++++++++++++----------- >>>> 1 file changed, 19 insertions(+), 11 deletions(-) >>>> >>>> Will commit this to stable-3.x and master in a few days if there are >>>> no objections. >>>> >>>> @Lennart, would you mind committing this to the upstream reserve.git repo as well? >>> >>> This seems pretty equivalent to a patch[1] that I sent earlier, with the >>> difference that with your patch change_cb() is called also in "busy -> >>> busy" transitions (i.e. when the bus name changes owner, and neither old >>> or new owner is pulseaudio). >>> >>> [1] http://thread.gmane.org/gmane.comp.audio.pulseaudio.general/15053 >> >> There is the problem of busy not being correctly initialized, i e, the >> initial value of busy does not take ourselves into account. Explicitly >> checking is_really_busy(old) works around this issue. (Properly >> initializing busy is a more elegant solution though.) > > Good point. So, there are two bugs: m->busy isn't initialized properly, > and change_cb() is sometimes called even when m->busy doesn't change. Well, I figured that it would be - maybe mostly theoretical here, but still - the case that given three apps A, B, and C with descending priorities, that B want to know of a change from A to C so that it could grab the sound card now that a lower priority app has it. > How to proceed? If I was to decide, we'd take my patch for fixing the > redundant change_cb() calls, and adapt your patch to fix the > initialization (is_really_busy() should be called in rm_watch(), instead > of using it in filter_handler() to work around the initialization bug). > If that sounds good, do you wish to send an updated patch, or do you > prefer me to do it? I don't have time/priority to fix the initialization (I believe it requires a call to org.freedesktop's GetNameOwner function?), so if you do the initialization within a few days time I'm okay with that. Otherwise I suggest we push my patch as it is. -- David Henningsson, Canonical Ltd. https://launchpad.net/~diwic
