[FFmpeg-cvslog] r15334 - trunk/libavcodec/golomb.h

[Aurelien Jacobs]

Uoti Urpala wrote:

> On Mon, 2008-09-15 at 22:30 +0200, michael wrote:
> > Log:
> > Fix issue245.
> 
> IMO mentioning only a bug tracker number is not a good idea. It takes
> extra time if you want to look it up (it's not even a link you could
> directly click), and the information may not be available in the future.
> Is there any guarantee the bug tracker tickets will survive with the
> same numbering if FFmpeg for example changes bug tracker implementation?

I agree that is not really a good commit message.

But more importantly, it breaks regression tests. More precisely,
the following command goes into an endless loop:
./ffmpeg_g -y -flags +bitexact -dct fastint -idct simple -sws_flags +accurate_rnd+bitexact -i ./tests/data/a-jpegls.avi -f rawvideo -pix_fmt yuv420p -sws_flags area+bitexact ./tests/data/jpegls.vsynth.out.yuv

Interrupting it with gdb shows the culprit:

Program received signal SIGINT, Interrupt.
0x000000000069f763 in ls_decode_line (state=0xffcb50, s=0xf78570, 
    last=0x1050a61, dst=0x1050ee1, last2=187, w=1056, stride=3, 
    comp=<value optimized out>, bits=8) at libavcodec/jpeglsdec.c:92
92	    for(k = 0; (state->N[Q] << k) < state->A[Q]; k++);

This is the endless loop. And I have no time right now to analyze it
further. Sorry.

Aurel