[FFmpeg-cvslog] r26053 - in trunk: libavformat/asfdec.c tests/ref/fate/wmv8-drm

Reimar Döffinger Reimar.Doeffinger
Sun Dec 19 00:13:30 CET 2010


On Sat, Dec 18, 2010 at 08:01:05PM +0200, Kostya wrote:
> On Sat, Dec 18, 2010 at 06:19:00PM +0100, Reimar D?ffinger wrote:
> [...]
> > 
> > This time unfortunately the issue is indeed that the VC-1 decoder reads outside
> > the packet size.
> > Below hack would fix it, but that is of course not ok.
> > My proposal would be to remove that fate test an only keep the one I added
> > (does not decode), unless someone comes up with a way to fix the VC-1 decoder...
> 
> Well, how serious it can be? Bits overreading is checked every macroblock,
> and while this patch looks ok it may slow decoding down a bit (and it needs
> more checks in other places as well then).

One thing that makes me really suspicious now though is: why do we get different
FATE test results?
Should the error concealment code not at least discard all data we read from
_after_ the overread?
It seems to me like error concealment (if caused by overread) discards at least
one MB too little, or is that intentional?



More information about the ffmpeg-cvslog mailing list