[FFmpeg-devel] [PATCH 1/2] avcodec/decode: inject missing global side data to output frames

[James Almer]

On 2/26/2025 1:57 PM, Andreas Rheinhardt wrote:
> James Almer:
>> ff_decode_frame_props() injects global side data passed by the caller (Usually
>> coming from the container) but ignores the global side data the decoder
>> gathered from the bitstream itself.
>> This commit amends this.
>>
>> Signed-off-by: James Almer <jamrial at gmail.com>
>> ---
>>   libavcodec/decode.c | 9 +++++++++
>>   1 file changed, 9 insertions(+)
>>
>> diff --git a/libavcodec/decode.c b/libavcodec/decode.c
>> index cac7e620d2..c5a577f4f1 100644
>> --- a/libavcodec/decode.c
>> +++ b/libavcodec/decode.c
>> @@ -1570,6 +1570,15 @@ int ff_decode_frame_props(AVCodecContext *avctx, AVFrame *frame)
>>       if (ret < 0)
>>           return ret;
>>   
>> +    for (int i = 0; i < avctx->nb_decoded_side_data; i++) {
>> +        const AVFrameSideData *src = avctx->decoded_side_data[i];
>> +        if (av_frame_get_side_data(frame, src->type))
>> +            continue;
>> +        ret = av_frame_side_data_clone(&frame->side_data, &frame->nb_side_data, src, 0);
>> +        if (ret < 0)
>> +            return ret;
>> +    }
>> +
>>       if (!(ffcodec(avctx->codec)->caps_internal & FF_CODEC_CAP_SETS_FRAME_PROPS)) {
>>           const AVPacket *pkt = avctx->internal->last_pkt_props;
>>   
> 
> Why is this attached to every frame instead of just the first one?

These entries are stream wide and apply to every frame (unless 
overridden by frame specific side data), same as the global entries that 
came from the container that are attached immediately above this chunk.

> If I read this correctly, it will be possible for there to be
> decoded_side_data propagated to the output, then in-stream side-data of
> the same type (not overridden) and in the next frame, the decoder adds
> the decoded_side_data again, presuming that the side data from the
> earlier frame was meant to be non-persistent.

If a side data that's coded in a frame is meant persist beyond the frame 
where it appeared, then the decoder is who should take care of that by 
attaching it to every frame that needs it before returning it, and not 
the generic code.
This is the case with for example mastering display metadata in h264/5.

-------------- next part --------------
A non-text attachment was scrubbed...
Name: OpenPGP_signature.asc
Type: application/pgp-signature
Size: 495 bytes
Desc: OpenPGP digital signature
URL: <https://ffmpeg.org/pipermail/ffmpeg-devel/attachments/20250226/72671411/attachment.sig>