[FFmpeg-devel] Longevity Test Report

[Reimar Döffinger]

On Sat, Feb 07, 2009 at 09:25:27PM -0800, Mike Melanson wrote:
> So now I'm more confused than before. Maybe a leak in the demux 
> subsystem when it has to demux more than one stream? One more test for 
> good measure:
> 
> ffmpeg -i file.mp4 -vcodec rawvideo -acodec pcm_s16le -y out.avi
> 
> leaks as bad as the original problem.

Better don't talk about a "leak", most people expect that to mean that
memory will not be freed.
What (at least with MPlayer, maybe this is not possible with FFmpeg)
is a very common issue is that there is A-V desync (either in the
source or the output file), causing it to buffer e.g. a lot of video
data, and is not strictly "wrong".
Another popular reason is the index generation, since that can usually
be written at the end all the data for it must be kept in memory.
This is probably less of an issue for FFmpeg, since MPlayer e.g. often
generates 1000 fps files with lots of empty video frames (consume little
space in the generated file, but make the index grow huge).
Thus the suggestion of looking at all allocations and see which codes
allocates more data than you think it should.

Greetings,
Reimar D?ffinger