[FFmpeg-cvslog] r210, r10k and avrp encoder

[Michael Niedermayer]

On Fri, Jan 27, 2012 at 07:56:05AM +0100, Reimar Döffinger wrote:
> On 26 Jan 2012, at 23:47, Michael Niedermayer <michaelni at gmx.at> wrote:
> > On Thu, Jan 26, 2012 at 11:08:16PM +0100, Reimar Döffinger wrote:
> >> On Thu, Jan 26, 2012 at 10:42:43PM +0100, Michael Niedermayer wrote:
> >>> On Thu, Jan 26, 2012 at 10:10:02PM +0100, Reimar Döffinger wrote:
> >>>> On 26 Jan 2012, at 20:01, Michael Niedermayer <michaelni at gmx.at> wrote:
> >>>>> On Thu, Jan 26, 2012 at 08:45:09AM +0100, Reimar Döffinger wrote:
> >>>>>> 
> >>>>>>> +    avctx->coded_frame->reference = 0;
> >>>>>>> +    avctx->coded_frame->key_frame = 1;
> >>>>>>> +    avctx->coded_frame->pict_type = AV_PICTURE_TYPE_I;
> >>>>>> 
> >>>>>> That makes yet two more encoders for which pts is not handled correctly.
> >>>>> 
> >>>>> why ?
> >>>>> 
> >>>>> avctx->coded_frame = avcodec_alloc_frame();
> >>>>> should set all the pts fields to AV_NOPTS_VALUE
> >>>> 
> >>>> And why should the outgoing pts be AV_NOPTS_VALUE when the incoming is not?
> >>>> And how would you distinguish the cases where the encoder just never sets pts vs. an encoder that just (possibly with reordering) passed through a AV_NOPTS_VALUE it got as input?
> >>> 
> >>> CODEC_CAP_DELAY could be used
> >> 
> >> Could be. Excuse my inability to contain the snark, but
> >> after a few years of using it, it might be a good idea to
> >> decide on who the API should work...
> >> Because with the current code in FFmpeg an encoder returning
> >> AV_NOPTS_VALUE as far as I can tell is just as broken in the final
> >> result as one returning 0.
> > 
> > So to summarize, the API is inadequately documented and ffmpeg is
> > containing a bug.
> > 
> > The question is are you volunteering to fix these or should i do it?
> 
> I have a patch for FFmpeg and rawenc, the problem is
> a) whether it is the best thing to do

> b) the many H.264 test failures (which are also visible with my previous hacks in exactly the same way)

you could add a '-vsync ignore' that ignores timestamps and then add
this vsync option to the tests that need it. That way we also would
keep track of which tests have timestamp issues.



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

The real ebay dictionary, page 3
"Rare item" - "Common item with rare defect or maybe just a lie"
"Professional" - "'Toy' made in china, not functional except as doorstop"
"Experts will know" - "The seller hopes you are not an expert"
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: Digital signature
URL: <http://ffmpeg.org/pipermail/ffmpeg-cvslog/attachments/20120127/dabf9e68/attachment.asc>