[FFmpeg-devel] [PATCH] md5: change flags to match framemd5.

[Michael Niedermayer]

On Sat, Sep 05, 2015 at 02:22:48PM +0200, Hendrik Leppkes wrote:
> On Sat, Sep 5, 2015 at 2:16 PM, Michael Niedermayer <michaelni at gmx.at> wrote:
> > On Sat, Sep 05, 2015 at 07:11:39AM -0400, Ronald S. Bultje wrote:
> >> This prevents the following warning: "Application provided invalid, non
> >> monotonically increasing dts to muxer in stream 0: 9 >= 9".
> >> ---
> >>  libavformat/md5enc.c | 3 ++-
> >>  1 file changed, 2 insertions(+), 1 deletion(-)
> >>
> >> diff --git a/libavformat/md5enc.c b/libavformat/md5enc.c
> >> index 8e87f09..e1a5368 100644
> >> --- a/libavformat/md5enc.c
> >> +++ b/libavformat/md5enc.c
> >> @@ -107,7 +107,8 @@ AVOutputFormat ff_md5_muxer = {
> >>      .write_header      = write_header,
> >>      .write_packet      = write_packet,
> >>      .write_trailer     = write_trailer,
> >> -    .flags             = AVFMT_NOTIMESTAMPS,
> >> +    .flags             = AVFMT_VARIABLE_FPS | AVFMT_TS_NONSTRICT |
> >> +                         AVFMT_TS_NEGATIVE,
> >>      .priv_class        = &md5enc_class,
> >>  };
> >>  #endif
> >
> > The warning indicates a problem, I dont think we should "remove" the
> > warning without fixing the problem
> >
> > also some kind of md5 checksum muxer which validates timestamps for
> > regression tests should be kept. Validating timestamps is important
> >
> > Tests which pass timestamp validation currently should raise a big
> > red flag when they fail.
> > Ive not checked but i suspect md5 would not give any indication that
> > good valid timestamps changed into bad invalid ones, it would give
> > the same checksum with no warnings either way
> > That would be rather bad for regression tests as worsening timestamp
> > validity could be easily missed
> >
> 
> IMHO if you wanted to test timestamps of individual frames, you would
> use framemd5, as that will immediately show the difference in the
> timestamps in the output, and not abort on a somewhat confusing
> generic avformat error.

I would use framemd5, but some of our fate tests use plain md5
these should probably be changed to framemd5 then
also i would suggest to clarify the md5 docs to mention explicitly
that md5 ignores timestamps

[...]
-- 
Michael     GnuPG fingerprint: 9FF2128B147EF6730BADF133611EC787040B0FAB

In a rich man's house there is no place to spit but his face.
-- Diogenes of Sinope
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 181 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20150905/663b1c58/attachment.sig>