[FFmpeg-devel] [PATCH v7 1/8] Add generic metadata injection using AV_PKT_DATA_METADATA_UPDATE

[Romain Beauxis]

Le lun. 10 mars 2025 à 15:46, Michael Niedermayer
<michael at niedermayer.cc> a écrit :
>
> On Mon, Mar 10, 2025 at 08:50:06AM -0500, Romain Beauxis wrote:
> > Le dim. 9 mars 2025 à 13:31, Michael Niedermayer
> > <michael at niedermayer.cc> a écrit :
> > >
> > > On Tue, Feb 25, 2025 at 04:01:14PM -0600, Romain Beauxis wrote:
> > > > libavcodec/decode.c: intercept `AV_PKT_DATA_METADATA_UPDATE` packet
> > > > extra data, attach them to the next decoded frame.
> > > >
> > > > ---
> > > >  libavcodec/decode.c | 20 ++++++++++++++++++++
> > > >  1 file changed, 20 insertions(+)
> > > >
> > > > diff --git a/libavcodec/decode.c b/libavcodec/decode.c
> > > > index cac7e620d2..96e2f0ce95 100644
> > > > --- a/libavcodec/decode.c
> > > > +++ b/libavcodec/decode.c
> > > > @@ -97,6 +97,8 @@ typedef struct DecodeContext {
> > > >      int lcevc_frame;
> > > >      int width;
> > > >      int height;
> > > > +
> > > > +    AVDictionary *pending_metadata;
> > > >  } DecodeContext;
> > > >
> > > >  static DecodeContext *decode_ctx(AVCodecInternal *avci)
> > > > @@ -729,6 +731,8 @@ int attribute_align_arg avcodec_send_packet(AVCodecContext *avctx, const AVPacke
> > > >  {
> > > >      AVCodecInternal *avci = avctx->internal;
> > > >      DecodeContext     *dc = decode_ctx(avci);
> > > > +    const uint8_t *side_metadata;
> > > > +    size_t size;
> > > >      int ret;
> > > >
> > > >      if (!avcodec_is_open(avctx) || !av_codec_is_decoder(avctx->codec))
> > > > @@ -746,6 +750,14 @@ int attribute_align_arg avcodec_send_packet(AVCodecContext *avctx, const AVPacke
> > > >          ret = av_packet_ref(avci->buffer_pkt, avpkt);
> > > >          if (ret < 0)
> > > >              return ret;
> > > > +
> > > > +        side_metadata = av_packet_get_side_data(avpkt, AV_PKT_DATA_METADATA_UPDATE, &size);
> > > > +        if (side_metadata) {
> > > > +            av_dict_free(&dc->pending_metadata);
> > > > +            ret = av_packet_unpack_dictionary(side_metadata, size, &dc->pending_metadata);
> > > > +            if (ret < 0)
> > > > +                return ret;
> > > > +        }
> > > >      } else
> > > >          dc->draining_started = 1;
> > > >
> > > > @@ -815,6 +827,7 @@ fail:
> > > >  int ff_decode_receive_frame(AVCodecContext *avctx, AVFrame *frame)
> > > >  {
> > > >      AVCodecInternal *avci = avctx->internal;
> > > > +    DecodeContext     *dc = decode_ctx(avci);
> > > >      int ret;
> > > >
> > > >      if (!avcodec_is_open(avctx) || !av_codec_is_decoder(avctx->codec))
> > > > @@ -887,6 +900,12 @@ int ff_decode_receive_frame(AVCodecContext *avctx, AVFrame *frame)
> > > >          }
> > > >      }
> > > >  #endif
> > > > +
> > > > +    if (dc->pending_metadata) {
> > > > +        av_dict_copy(&frame->metadata, dc->pending_metadata, AV_DICT_APPEND);
> > > > +        av_dict_free(&dc->pending_metadata);
> > > > +    }
> > > > +
> > > >      return 0;
> > > >  fail:
> > > >      av_frame_unref(frame);
> > >
> > > Does this work correctly, if theres a delay/latency in the decoder ?
> >
> > Can you describe a scenario that you're thinking about?
>
> The users feeds several packets into a multi threaded decoder
> and then depending on the threads and luck sooner or later
> one frame comes out.
>
> Passing some data in a way that disregards this, feels wrong
> Hypothetically there also could be a 2nd AV_PKT_DATA_METADATA_UPDATE
> going in before the frame corresponding to the first comes out
> but i may be missing something

As far as I can tell, the mechanism is only used by ogg streams and
for ogg, these scenarios would not really apply I believe.

Currently, DTS are broken for chained ogg streams so passing packets
in any other order than the stream's sequential order would not work.

I think this concern could be addressed in the planned follow-up work,
which is to remove the ogg headers from the demuxer and attach
AV_PKT_DATA_METADATA_UPDATE to a packet with actual media content.

In this case, the metadata would be immediately attached to the frame
when the packet is decoded.

-- Romain