On 18.06.2015 23:43, Ronald S. Bultje wrote: > OK, looked further, it's actually quite complicated and I don't think > it's easy to do "the correct thing". The reason is that on frame header > error, we don't move the reference frame state forward (that's the code > in vp9_decode_frame right after reading the frame header), so all kind > of things get in a weird state. Your patch should probably go in, it > doesn't "fix" the weird state (i.e. I'm not convinced it's technically > "correct") but at least it's not "bad" (crash-causing), just "weird" > (technically possibly wrong, but it already was so that's ok). > > To do the correct thing, I should probably implement error resilience, > it's not exactly easy so I'll leave that for sometime later.
Thanks for this analysis. I've pushed the patch now. By the way, afl [1] is really a great tool for finding such problems. I can only recommend everyone to try it out yourself. Best regards, Andreas 1: http://lcamtuf.coredump.cx/afl/ _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
