[FFmpeg-devel] [PATCH v5 1/6] libavcodec: Add generic metadata injection using AV_PKT_DATA_METADATA_UPDATE

[Romain Beauxis]

Le mar. 18 févr. 2025 à 22:11, Romain Beauxis
<romain.beauxis at gmail.com> a écrit :
>
> Le mar. 18 févr. 2025 à 17:28, Lynne <dev at lynne.ee> a écrit :
> >
> > On 17/02/2025 17:19, Romain Beauxis wrote:
> > > libavcodec/decode.c: intercept `AV_PKT_DATA_METADATA_UPDATE` packet extra data,
> > > attach them to the next decoded frame.
> > >
> > > The metadata needs to be cached because there is no guarantee that each packet
> > > will generate a decoded frame immediately.
> > >
> > > `AV_PKT_DATA_METADATA_UPDATE` does not seem used in `libavcodec` at the moment
> > > so regression risks seem low.
> > >
> > > This generic mechanism could be reused to support more in-band metadata update
> > > in the future.
> > >
> > > ---
> > >   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);
> > > @@ -2314,4 +2333,5 @@ void ff_decode_internal_uninit(AVCodecContext *avctx)
> > >       DecodeContext *dc = decode_ctx(avci);
> > >
> > >       av_refstruct_unref(&dc->lcevc);
> > > +    av_dict_free(&dc->pending_metadata);
> > >   }
> >
> > btw, jamrial mentioned that AVSTREAM_EVENT_FLAG_METADATA_UPDATED exists.
> > I only see it used to inform API users that metadata has been updated,
> > but you should set the flag.
> >
> > Other than that, I'm fine with the patchset. You should resend it with
> > the change, and if there are no comments, I'll merge it in a few days.
>
> Thank you for the review.
>
> Thanks for pointing that out. Since ogg vorbis already supports
> AVSTREAM_EVENT_FLAG_METADATA_UPDATED I think it would make sense to
> favor our the whole process from there.
>
> However, I have some questions about how this is implemented in vorbis.
>
> The vorbis implementation:
> * Parses new comment metadata and adds them to the AVStream's metadata
> * Entries are appended in this case there is already an existing one:
>
>          if (av_dict_get(*m, t, NULL, 0))
>             av_dict_set(m, t, ";", AV_DICT_APPEND);
>         av_dict_set(m, t, v, AV_DICT_APPEND);
>
> * Then AVSTREAM_EVENT_FLAG_METADATA_UPDATED is set on the AVStream.
>
> I can see a couple of issues:
> * This is pretty unexpected behavior. Any consumer of the stream's
> metadata would expect a single title, not an accumulation of all
> previously parsed titles
> * This is also a memory leak.
>
> Should we make it so the stream's duplicated metadata are taking the
> last value instead?

Actually, I'd miss a sneaky `av_dict_free` call!

Updated patchset coming soon!

-- Romain