On Sun, Aug 23, 2015 at 6:58 PM, Michael Niedermayer <mich...@niedermayer.cc> wrote: > On Sat, Aug 22, 2015 at 04:26:53PM +0200, Nicolas George wrote: >> Le quintidi 5 fructidor, an CCXXIII, Ganesh Ajjanagadde a écrit : >> > + /* cast for -Wabsolute-value in clang 3.5+ */ >> > + return abs((int) (t1 - t2)) + abs((int) ((c1 & 0x000000ff) - (c2 & >> > 0x000000ff))) + >> > + abs((int) (((c1 & 0x0000ff00) >> 8) - ((c2 & 0x0000ff00) >> 8))) + >> > + abs((int) (((c1 & 0x00ff0000) >> 16) - ((c2 & 0x00ff0000) >> >> > 16))); >> >> I believe this is not correct, with or without the cast (the cast is >> implicit due to the prototype of abs(), of course): if the result is morally >> negative, converting the morally-negative unsigned to signed is an undefined >> behaviour. > > i dont know what "morally negative" means exactly but convertion > of out of range values to signed integers is not undefined behavior > in C. It is implementation defined or raises a implementation defined > signal > Please see 6.3.1.3 Signed and unsigned integers in ISO/IEC 9899:TC3 > > I doubt that FFmpeg will work on platforms that > define this significantly differently than how it is defined for the > currently supported platforms,
You are right that all standard platforms do this, which is why we never had a bug. However, we can achieve compliance with no cost with the patchv2. Maybe the macro/casts are ugly, but I currently treat standards compliance (minimal reliance on implementation defined behavior) as most important. > > [...] > > -- > Michael GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB > > The real ebay dictionary, page 3 > "Rare item" - "Common item with rare defect or maybe just a lie" > "Professional" - "'Toy' made in china, not functional except as doorstop" > "Experts will know" - "The seller hopes you are not an expert" > > _______________________________________________ > ffmpeg-devel mailing list > ffmpeg-devel@ffmpeg.org > http://ffmpeg.org/mailman/listinfo/ffmpeg-devel > _______________________________________________ ffmpeg-devel mailing list ffmpeg-devel@ffmpeg.org http://ffmpeg.org/mailman/listinfo/ffmpeg-devel
