[Libav-user] Frames corruption when seeking...

[Jean-Yves Avenard]

On 5 June 2014 11:41, Jean-Yves Avenard <jyavenard at gmail.com> wrote:
> On 5 June 2014 03:46, wm4 <nfxjfg at googlemail.com> wrote:
>
>> My hw doesn't do mpeg4, so I have no way to reproduce.
>
> Interestingly, I was googling for something else entirely (ffmpeg
> vdpau render2) and stumble upon this thread:
> http://lists.mplayerhq.hu/pipermail/mplayer-users/2013-August/086497.html
>
> What he describes is exactly what I experience...
>
> but the discussion stops short, and new news since.
>
> Otherwise, back to the original topic of my message and frames corruption.
> I can now reproduced it consistently with VC-1 videos only....
>
> I was checking what xbmc was doing, and saw this comment in their
> code, in their draw slice code:
>   // ffmpeg vc-1 decoder does not flush, make sure the data buffer is
> still valid

this was a red herring...

after debugging, that code is never reached even after seeking.

What I do notice however is while myth displays part of the previous
frames (from before the seek) after; xbmc will not display anything
for several seconds after a seek (you only get audio).

Not sure what is best.. Getting part of the frames after a seek, but
slightly corrupted.. Or nothing at all..

neither behaviour are acceptable IMHO..

something in that VC-1 decoder I can't pinpoint