On 08.07.2015 18:07, Michael Niedermayer wrote: > On Wed, Jul 08, 2015 at 07:24:13AM -0400, Ronald S. Bultje wrote: >> - last of all, I want to remind you guys that we haven't done a very >> thorough code analysis to see if this can actually be triggered. As an >> example, look at commit 458446acfa1441d283dacf9e6e545beb083b8bb0. Before >> that, vp8 files smaller than 16x16 or vp9 files smaller than 64x64 could >> cause aborts. The vp9 ones I noticed due to some fate files being so small. >> I doubt you'll ever find such vp8 files, but they're valid. >> >> Let's not add silly checks to replace other silly ones.
Do you have any vp9 files triggering this assert? > if people prefer i ca revert b8c438e7623644edcfdb508c946a723a3b79b893 > or add a check for stride == 0 > > what do people prefer ? The assert can be triggered by the h264 decoder: mc_dir_part in libavcodec/h264_mb.c calls the 16 bit variant of emulated_edge_mc and thus the assert is triggered due to the sizeof(pixel) factor, even though block_w is smaller than buf_linesize. If that's valid behavior of the h264 decoder, then the assert should be removed, but if on the other hand this is a bug in the h264 decoder, then that should be fixed and the assert remain. Best regards, Andreas _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
