[MPlayer-dev-eng] Making lavf ogg demuxer the default

[Reimar Döffinger]

On Sun, May 23, 2010 at 11:31:13PM +0200, Reimar Döffinger wrote:
> On Sun, May 23, 2010 at 11:21:22PM +0200, Diego Biurrun wrote:
> > On Sun, May 23, 2010 at 09:09:09PM +0200, Reimar Döffinger wrote:
> > > On Sun, May 23, 2010 at 09:01:58PM +0200, Reimar Döffinger wrote:
> > > > since our own ogg demuxer cannot demux theora if compiled without libtheora,
> > > > I think the lavf ogg demuxer should be made default.
> > > > Does anyone know of particular issues with it?
> > > > I remember keyframe issues (IIRC fixed), subtitle issues (fixed as well I think),
> > > > problem with language <-> stream mapping (not sure), and a few others that should be
> > > > fixed.
> > > > If it works, that may also be one reason less for the tremor code...
> > > 
> > > Well, already found one issue, at least with
> > > http://samples.mplayerhq.hu/ogg/bgc.sub.dub.ogm
> > > seeking is extremely (I guess > 10x) slower...
> > 
> > Is your last commit and the patch you submitted to ffmpeg-devel enough
> > to fix the issue?  I'm all for using lavf by default for Ogg demuxing.
> 
> Well... For Windows and OS/2 I think it will only get it to the
> "barely acceptable" level.
> For POSIX-systems I am also not sure for systems that set siginterrupt to
> 0 by default, if that also applies to usleep.

I think performace should now be fairly ok in all cases.
Probably non-interleaved mov performance will be improved as well, at the
cost of high CPU wakeup load (there should be a bugzilla bug about that,
if someone wants to test - not it will only really help for locally stored
mov, not so much over http).