[MPlayer-dev-eng] [PATCH] unified timing patch for H264

[Carl Eugen Hoyos]

Reimar Döffinger <Reimar.Doeffinger <at> gmx.de> writes:

> > Do we still care about -demuxer lavf -nocorrect-pts?
> > It is completely broken with your patch (and a PAFF stream).
> 
> On the other hand, I think it will make
> -demuxer lavf -fps ...
> work correctly I think?
> Of course the problem is that this breaks mencoder
> with any such stream which maybe isn't so great.

I don't see a difference in mencoders behaviour with or without the patch: In
both cases. PAFF in MPEG as input works ok with -demuxer lavf, but not with the
native demuxer.
(Is it known that mencoder -oac lavc -of avi is completely broken meaning it
produces avi's with seemingly funny audio timestamps?)

> I'm sure it was mentioned before, but could someone summarize
> what it fixes and what it breaks, possibly linking samples
> for each?

All PAFF samples in MPEG TS streams are fixed for the native demuxer (and
nocorrect-ots which is the default) with this patch, -demuxer lavf
-nocorrect-pts (which is not the default) is broken for all (?) MPEG TS streams
containing PAFF with the patch.

The advantage of the native TS demuxer is that seeking works much better (from a
users pov at least), sync after seek is fine and even with my ORF recordings
that don't want to seek with -demuxer lavf (roundup issue 1963), seeking works
fine with the native demuxer.

For the MPEG PS native demuxer, the patch makes no difference, since it does not
support H264 streams (or I haven't found a supported one yet)...

The sony 9-second sample is enough to reproduce this, but I also tested a long
Asian television recording (with visible mouth sync), the premiere sample and
some non-MPEG PAFF samples: MKV is completely out-of-sync with -nocorrect-pts
without the patch and playing slightly better (but not fine) with the patch
(tv.mkv), mov plays fine in all configurations (with and without patch, with
native and lavf demuxer, with correct-pts and nocorrect-pts).

Carl Eugen