1. When one uses UBSan (Undefined behaviour sanitizer) compiled with the -fsanitize-trap=undefined option (that makes the resulting binary trap on undefined behaviour instead of simply emitting a warning), lots of FATE tests (about 960 on an x64) fail. Over 200 of these are false positives because of unaligned accesses inside #ifdef HAVE_FAST_UNALIGNED, but even accounting for that, the number is impressive.
2. ubitux (who runs a FATE box that uses UBSan [1]) has therefore proposed that the UBSan toolchain configure options be changed to trap by default so that undefined behaviour doesn't go unnoticed. What does the community think about this? 3. I have also fixed some of the occurences of undefined behaviour. With the exception of the mpeg12dec patch all these patches are little no-brainer. - Andreas [1]: http://fate.ffmpeg.org/report.cgi?slot=x86_64-archlinux-gcc-ubsan&time=20190919170122 Andreas Rheinhardt (11): avformat/sdsdec: Fix undefined shifts avcodec/mpeg12dec: Sanitize start codes earlier avcodec/wavpackenc: Fix undefined shifts avcodec/tdsc: Fix undefined shifts avformat/nutenc: Don't pass NULL to memcmp avcodec/pcm: Fix undefined shifts avcodec/pcm: Cosmetics avformat/movenc: Fix undefined shift avcodec/g723_1dec: Fix invalid shift avcodec/truespeech: Fix invalid shifts avcodec/ac3enc: Fix invalid shift libavcodec/ac3enc.c | 2 +- libavcodec/g723_1dec.c | 2 +- libavcodec/mpeg12dec.c | 4 +++- libavcodec/pcm.c | 30 +++++++++++++++--------------- libavcodec/tdsc.c | 6 +++--- libavcodec/truespeech.c | 8 ++++---- libavcodec/wavpackenc.c | 6 +++--- libavformat/movenc.c | 4 +++- libavformat/nutenc.c | 5 +++-- libavformat/sdsdec.c | 6 +++--- 10 files changed, 39 insertions(+), 34 deletions(-) -- 2.20.1 _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email ffmpeg-devel-requ...@ffmpeg.org with subject "unsubscribe".