[theora] Fwd: ffmpeg2theora 0.24 regression: accelerated video output (converted from h264)

[Jason Self]

On Thu, Jul 16, 2009 at 6:22 AM, <xiphmont at xiph.org> wrote:
> let the PTS on the video/audio packets dictate the actual sync.

It appears that the folks at handbrake.fr are alread doing just that.

I've never experienced these sync issues using HandBrake (they also
make use of libavcodec and libavformat from the FFmpeg project.)

Per their wiki, "HandBrake writes its own time codes to the audio and
video streams after they have been decoded in the sync thread
(sync.c). It uses the Presentation Time Stamp (PTS) that is in the
source packets in order to sync up the streams and then overwrite the
existing PTS values with internal ones (that remain consistent)."

http://trac.handbrake.fr/wiki/LibHandBrakeSync.

I wonder if there's a reason that FFmpeg doesn't also do this?

It might be interesting to see what they did to accomplish that (I
never seem to get around to it.) Depending on what they did, perhaps
it could also be applied to ffmpeg2theora?