[FFmpeg-devel] Bug: Regression in audio decoding of some FLV videos.

[Corey O'Connor]

Sorry for posting via devel list but roundup crashed every time I try
to submit this bug. :-P

I've been using SVN-r5487 plus a few patches for FLV decoding with
great success for some time now. I upgraded to SVN-r14372 (tip of
trunk) and re-ran all my regression tests. The tests indicated that
the audio decoding of some FLVs was producing noise instead of the
expected output. EG:

http://www.dailymotion.com/get/7/320x240/flv/522594.flv?key=fa6ffcf8e3752718ad41aecaeb9d80ff1357ce4

http://www.dailymotion.com/get/9/320x240/flv/557931.flv?key=e3fda5a075b91da4bbb1c7450f3ca1f01357ce3

SVN-r5487 did not have this issue.

I haven't been able to track down what change caused the regression. I
was able to reproduce the change in r5487 by changing where the
packets pts was set in libavformat/flvdec.c to instead set the dts.
The inverse applied to tip-of-trunk did not resolve the issue, but
maybe that provides a clue as to what is going on.

-Corey O'Connor