[FFmpeg-devel] [PATCH] libavcodec/avcodec.h: fix typos in AVCodecContext.pkt_timebase description

[Anton Khirnov]

Quoting Stefano Sabatini (2023-04-02 21:03:21)
> On date Sunday 2023-04-02 20:14:16 +0200, Stefano Sabatini wrote:
> > On date Sunday 2023-04-02 20:05:35 +0200, Stefano Sabatini wrote:
> > > ---
> > >  libavcodec/avcodec.h | 6 +++---
> > >  1 file changed, 3 insertions(+), 3 deletions(-)
> > > 
> > > diff --git a/libavcodec/avcodec.h b/libavcodec/avcodec.h
> > > index 1e91b9cb53..dec5e40eac 100644
> > > --- a/libavcodec/avcodec.h
> > > +++ b/libavcodec/avcodec.h
> > > @@ -1781,9 +1781,9 @@ typedef struct AVCodecContext {
> > >      enum AVPixelFormat sw_pix_fmt;
> > >  
> > >      /**
> > > -     * Timebase in which pkt_dts/pts and AVPacket.dts/pts are.
> > > -     * - encoding unused.
> > > -     * - decoding set by user.
> > > +     * Timebase in which pkt_dts/pts and AVPacket.dts/pts are expressed.
> > > +     * - encoding: unused.
> > > +     * - decoding: set by user.
> > >       */
> > >      AVRational pkt_timebase;
> > 
> > BTW, also I'm still confused by this field.
> > 
> > What happens in case it is not set?

The majority of decoding code treats timestamps as opaque numbers, they
are simply copied from packets to their corresponding frames. You don't
need to know the timebase to do that.

> > In this case I assume the decoder/libavcodec will pick a suitable
> > timebase which is directly set in the decoded frames? Is the field
> > internally set in this case, or the information is only stored in
> > the output frames?

Some code does more involved timestamp manipulation, e.g. adjusting
timestamps after skipping audio samples. When the timebase is not known,
this manipulation is just not done.

> > How libavcodec detects if the field is set or not?

By num&den being >0.

> 
> Also, is there any guarantee that the decoded frame time base should
> be set based on the pkt_timebase value?
> In practice this does not happen (so that I had to set the time_base
> explicitly in the decoded frame in transcode.c). Is this expected or
> the library should explicitly set the time_base in the decoded frames?
> The latter is what I would expect.

AVFrame/AVPacket.time_base were added quite recently and (to avoid
surprises for the API users) the libraries do not use them yet. If and
when that changes, it will probably require explicit opt-in from the
caller.
I've also recently started using them in ffmpeg CLI.

-- 
Anton Khirnov