[Ffmpeg-devel] ffserver segfault during regression tests

[Alex Beregszaszi]

Hi,

> I finally tracked down the exact commit that made ffserver SEGV during
> regression tests, it seems to be r6822.
> 
> use lambda/lmax instead of qscale/qmax as end check for vbv retry
> 
> commit to mpegvideo.c
> 
> Will try to find out more.

I have found other bugs last week, as I am now using ffserver for a
project, and it works nicely.

The bug I found among others, was in the avi muxer code, where it stuffs
zero length packets in order to achieve sync. This looped forever,
probably because ffserver sends fake PTSes. PTS handling must be fixed.

However, this fix didnt fixed the regressions, but when you started to
view an AVI stream, ffserver went up to 100%cpu/100%mem and eventually
killed by the OOM killer.

--
Alex