On date Saturday 2023-02-11 03:30:00 +0100, Stefano Sabatini wrote:
> On date Thursday 2023-02-09 00:41:18 +0100, Michael Niedermayer wrote:
> > On Mon, Feb 06, 2023 at 02:25:23AM +0100, Stefano Sabatini wrote:
> [...]
> > > Subject: [PATCH 2/2] ffmpeg: review -dts_delta_threshold and -dts_delta_error
> > >  options
> > > 
> > > Review handling of -dts_delta_threshold and -dts_delta_error options,
> > > specify them as floating point expressed in seconds.
> > > 
> > > Also, review and simplify logic. Adjust values for tests, since in
> > > some cases the new values does not trigger the correction logic.
> > > 
> > > PR: https://patchwork.ffmpeg.org/project/ffmpeg/list/?series=8252
> > > ---
> > >  doc/ffmpeg.texi             | 36 ++++++++++++++++---
> > >  fftools/ffmpeg.c            | 72 ++++++++++++++++++++-----------------
> > >  fftools/ffmpeg.h            |  2 ++
> > >  fftools/ffmpeg_demux.c      |  3 ++
> > >  tests/fate/filter-audio.mak |  2 +-
> > >  tests/fate/mpeg4.mak        |  2 +-
> > >  6 files changed, 77 insertions(+), 40 deletions(-)
> > 
> > This seems to break a case with concat and vsync
> > ./ffmpeg -y -i 'concat:///home/michael/videos/angels.mpg|/home/michael/videos/angels.mpg'  -vsync 0 -an file.avi
> > 
> > ...
> > cpb: bitrate max/min/avg: 0/0/200000 buffer size: 0 vbv_delay: N/A
> > [mpeg4 @ 0x55e051b8d4c0] Invalid pts (0) <= last (11)00.00 bitrate=N/A speed=   0x    
> > [vost#0:0/mpeg4 @ 0x55e051b9d700] Error submitting video frame to the encoder
> > Conversion failed!
> >
> > 
> > Ill mail you the angels.mpg, i think its online somewhere but i cant
> > find it
> 
> Fixed, now the code should be equivalent to the previous
> implementation.
> 
> What happened in this case (and apparently in the other fate tests
> failing), is that some sort of limit correction is applied:
> 
> detected dts:-0.041711 < dts_limit:0.358789
> ts delta 0.5005 applied => ts_offset_discont:0.5005 dts:0.458789
> 
> preventing the invalid pts error.
> 
> The limit correction, hardcoded in the ffmpeg.c code, is completely
> unrelated to the dts_delta_threshold value, no idea if it would make
> sense to make this parametric (but at least now it should be a bit
> more clear from the code/logs).

Moving the refactoring changes to a dedicated thread.

Updating the doc extensions.