On 30/10/18 23:45, James Almer wrote: > On 10/30/2018 8:19 PM, Mark Thompson wrote: >> On 30/10/18 19:21, James Almer wrote: >>> Signed-off-by: James Almer <jamr...@gmail.com> >>> --- >>> libavcodec/cbs_vp9.c | 3 +++ >>> 1 file changed, 3 insertions(+) >>> >>> diff --git a/libavcodec/cbs_vp9.c b/libavcodec/cbs_vp9.c >>> index 7498be4b73..812be4ddd5 100644 >>> --- a/libavcodec/cbs_vp9.c >>> +++ b/libavcodec/cbs_vp9.c >>> @@ -403,6 +403,9 @@ static int cbs_vp9_split_fragment(CodedBitstreamContext >>> *ctx, >>> uint8_t superframe_header; >>> int err; >>> >>> + if (frag->data_size <= 0) >>> + return 0; >>> + >>> // Last byte in the packet. >>> superframe_header = frag->data[frag->data_size - 1]; >>> >>> >> >> Seems fine, but why would an empty fragment appear here? > > I noticed this when i tried to reimplement the vp9 parser using cbs_vp9. > Libavformat passed dummy zero sized packets that resulted in cbs errors > trying to read the frame header.
Seems like we'd prefer they didn't get this far (zero-sized packets are flush packets to decoders), but oh well. >> (Given that H.26[45] has pretty much the same check, maybe it should be in >> the generic code.) > > And AV1 has a while (size > 0), so yeah, it may be a good idea. I don't mind - this patch LGTM, make it common if you like. Thanks, - Mark _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
