[FFmpeg-devel] seeking in ts (was: mpegvideo_parser outputs incomplete frames)

[Wolfram Gloger]

Hi,

> Date: Mon, 3 Sep 2007 00:28:40 +0200
> From: Michael Niedermayer <michaelni at gmx.at>

> On Sun, Sep 02, 2007 at 09:33:02PM -0000, Wolfram Gloger wrote:
> >=20
> > Well I want/need to call av_find_stream_info() after the seek (stream
> > parameters might have changed a lot after a seek in ts), and at the
> > end of that we have
> 
> could we maybe rather work on the truncated frame issue without dragging
> this into your highly controversal view of how to use av_find_stream_info()

Ok, first things first.  av_find_stream_info() is really a different
issue (though I think a majority would expect that to be callable
anywhere within the stream, if not it should be clearly documented).

> and to return to the original question, do or do not the patches change
> seeking behavior in ts

They do change the seeking behaviour with regard to what happens
_after_ a seek -- I have already posted the changes for the seek
regression test -- but IMHO they only change the fact that the first
frame isn't always labeled as a key frame any more.

> question is just if it gets worse due to the patches

I have not seen any new issues due to these except that in the frame
copy case the "drop initial non-keyframe" code in ffmpeg.c now really
comes into action and it can lead to the following output:

(video frame dropped)
audio frame
audio frame
audio frame...
(video frame dropped)
audio frame
audio frame
audio frame...
video frame (first real I frame)
audio frame...

so we have up to 0.4 seconds of audio before the first video frame
arrives and e.g. dvdauthor doesn't like that.  I have a workaround for
that but for now I'd just like to point that out as the only new
problem I've seen.

Regards,
Wolfram.