Re: [FFmpeg-devel] [PATCH v2 2/2] avformat/flvenc: add option to read metadata from file

[Gyan Doshi <ffmpeg@gyani.pro>]

On 2023-02-04 03:46 pm, Paul B Mahol wrote:
> On 2/4/23, Gyan Doshi <ffmpeg@gyani.pro> wrote:
>>
>> On 2023-02-03 09:04 pm, Andreas Rheinhardt wrote:
>>> Gyan Doshi:
>>>> Useful, in conjuntion with option meta_period, to vary metadata during
>>>> stream.
>>>>
>>>> File format is ffmetadata.
>>>> ---
>>>>    configure            |   2 +-
>>>>    doc/muxers.texi      |   3 +
>>>>    libavformat/flvenc.c | 133 +++++++++++++++++++++++++++++++++----------
>>>>    3 files changed, 107 insertions(+), 31 deletions(-)
>>>>
>>>> diff --git a/configure b/configure
>>>> index 9d78a244a3..de371632c4 100755
>>>> --- a/configure
>>>> +++ b/configure
>>>> @@ -3433,7 +3433,7 @@ eac3_demuxer_select="ac3_parser"
>>>>    f4v_muxer_select="mov_muxer"
>>>>    fifo_muxer_deps="threads"
>>>>    flac_demuxer_select="flac_parser"
>>>> -flv_muxer_select="aac_adtstoasc_bsf"
>>>> +flv_muxer_select="aac_adtstoasc_bsf ffmetadata_demuxer"
>>>>    gxf_muxer_select="pcm_rechunk_bsf"
>>>>    hds_muxer_select="flv_muxer"
>>>>    hls_demuxer_select="adts_header ac3_parser"
>>>> diff --git a/doc/muxers.texi b/doc/muxers.texi
>>>> index 02ecddf186..000c92b2a7 100644
>>>> --- a/doc/muxers.texi
>>>> +++ b/doc/muxers.texi
>>>> @@ -555,6 +555,9 @@ With every video packet.
>>>>    @end table
>>>>    Note that metadata will always be re-emitted if a metadata update event
>>>> is signalled.
>>>>
>>>> +@item meta_filename
>>>> +Specify a ffmetadata file from which to load metadata.
>>>> +
>>>>    @end table
>>>>
>>>>    @anchor{framecrc}
>>>> diff --git a/libavformat/flvenc.c b/libavformat/flvenc.c
>>>> index d1c7a493d1..1332b18b41 100644
>>>> --- a/libavformat/flvenc.c
>>>> +++ b/libavformat/flvenc.c
>>>> @@ -122,6 +122,10 @@ typedef struct FLVContext {
>>>>        double framerate;
>>>>        AVCodecParameters *data_par;
>>>>
>>>> +    char *meta_filename;
>>>> +    AVFormatContext *meta_ctx;
>>>> +    AVDictionary *meta_dict;
>>>> +
>>>>        int flags;
>>>>        int meta_period;
>>>>    } FLVContext;
>>>> @@ -277,6 +281,92 @@ static void put_amf_bool(AVIOContext *pb, int b)
>>>>        avio_w8(pb, !!b);
>>>>    }
>>>>
>>>> +static int read_metadata_from_file(AVFormatContext *s, unsigned int ts,
>>>> int header)
>>>> +{
>>>> +    FLVContext *flv = s->priv_data;
>>>> +    const AVInputFormat *meta_format = NULL;
>>>> +    AVDictionary *current_meta_dict = NULL;
>>>> +    float timestamp = ts/1000.f;
>>>> +    int ret;
>>>> +
>>>> +    if (!flv->meta_filename)
>>>> +        return 0;
>>>> +
>>>> +    meta_format = av_find_input_format("ffmetadata");
>>>> +    if (!meta_format) {
>>>> +        av_log(s, AV_LOG_ERROR, "ffmetadata demuxer not found.\n");
>>>> +        return AVERROR(ENOSYS);
>>>> +    }
>>>> +
>>>> +    avformat_close_input(&flv->meta_ctx);
>>>> +
>>>> +    ret = avformat_open_input(&flv->meta_ctx, flv->meta_filename,
>>>> meta_format, NULL);
>>>> +    if (ret < 0) {
>>>> +        av_log(s, AV_LOG_ERROR, "Failed to read metadata from file %s t:
>>>> %f.", flv->meta_filename, timestamp);
>>>> +        if (flv->meta_dict)
>>>> +            av_log(s, AV_LOG_ERROR, " Continuing with old metadata.");
>>>> +        av_log(s, AV_LOG_ERROR, "\n");
>>>> +        return ret;
>>>> +    }
>>>> +
>>>> +    if (flv->meta_dict) {
>>>> +        av_dict_copy(&current_meta_dict, flv->meta_dict, 0);
>>>> +        av_dict_free(&flv->meta_dict);
>>>> +    }
>>>> +
>>>> +    ret = av_dict_copy(&flv->meta_dict, flv->meta_ctx->metadata, 0);
>>>> +    if (ret < 0) {
>>>> +        av_log(s, AV_LOG_ERROR, "Could not transfer metadata from %s at
>>>> %f seconds. Continuing with old metadata.\n", flv->meta_filename,
>>>> timestamp);
>>>> +        av_dict_free(&flv->meta_dict);
>>>> +        av_dict_copy(&flv->meta_dict, current_meta_dict, 0);
>>>> +        av_dict_free(&current_meta_dict);
>>>> +        return ret;
>>>> +    }
>>>> +
>>>> +    av_log(s, AV_LOG_VERBOSE, "Metadata from file %s updated %s at %f
>>>> seconds.\n", flv->meta_filename, header ? "in header" : "in video
>>>> packet", timestamp);
>>>> +    av_dict_free(&current_meta_dict);
>>>> +    avformat_close_input(&flv->meta_ctx);
>>>> +
>>>> +    return 0;
>>>> +}
>>>> +
>>>> +static int write_user_metadata_tag(AVFormatContext *s, AVDictionaryEntry
>>>> *tag, AVIOContext *pb, int *metadata_count)
>>>> +{
>>>> +
>>>> +    av_log(s, AV_LOG_DEBUG, "Writing tag %s with value %s count: %d\n",
>>>> tag->key, tag->value, *metadata_count);
>>>> +
>>>> +    if(   !strcmp(tag->key, "width")
>>>> +        ||!strcmp(tag->key, "height")
>>>> +        ||!strcmp(tag->key, "videodatarate")
>>>> +        ||!strcmp(tag->key, "framerate")
>>>> +        ||!strcmp(tag->key, "videocodecid")
>>>> +        ||!strcmp(tag->key, "audiodatarate")
>>>> +        ||!strcmp(tag->key, "audiosamplerate")
>>>> +        ||!strcmp(tag->key, "audiosamplesize")
>>>> +        ||!strcmp(tag->key, "stereo")
>>>> +        ||!strcmp(tag->key, "audiocodecid")
>>>> +        ||!strcmp(tag->key, "duration")
>>>> +        ||!strcmp(tag->key, "onMetaData")
>>>> +        ||!strcmp(tag->key, "datasize")
>>>> +        ||!strcmp(tag->key, "lasttimestamp")
>>>> +        ||!strcmp(tag->key, "totalframes")
>>>> +        ||!strcmp(tag->key, "hasAudio")
>>>> +        ||!strcmp(tag->key, "hasVideo")
>>>> +        ||!strcmp(tag->key, "hasCuePoints")
>>>> +        ||!strcmp(tag->key, "hasMetadata")
>>>> +        ||!strcmp(tag->key, "hasKeyframes")
>>>> +    ){
>>>> +        av_log(s, AV_LOG_DEBUG, "Ignoring metadata for %s\n",
>>>> tag->key);
>>>> +        return AVERROR(EINVAL);
>>>> +    }
>>>> +    put_amf_string(pb, tag->key);
>>>> +    avio_w8(pb, AMF_DATA_TYPE_STRING);
>>>> +    put_amf_string(pb, tag->value);
>>>> +    (*metadata_count)++;
>>>> +
>>>> +    return 0;
>>>> +}
>>>> +
>>>>    static void write_metadata(AVFormatContext *s, unsigned int ts)
>>>>    {
>>>>        AVIOContext *pb = s->pb;
>>>> @@ -360,36 +450,11 @@ static void write_metadata(AVFormatContext *s,
>>>> unsigned int ts)
>>>>        }
>>>>
>>>>        ff_standardize_creation_time(s);
>>>> -    while ((tag = av_dict_iterate(s->metadata, tag))) {
>>>> -        if(   !strcmp(tag->key, "width")
>>>> -            ||!strcmp(tag->key, "height")
>>>> -            ||!strcmp(tag->key, "videodatarate")
>>>> -            ||!strcmp(tag->key, "framerate")
>>>> -            ||!strcmp(tag->key, "videocodecid")
>>>> -            ||!strcmp(tag->key, "audiodatarate")
>>>> -            ||!strcmp(tag->key, "audiosamplerate")
>>>> -            ||!strcmp(tag->key, "audiosamplesize")
>>>> -            ||!strcmp(tag->key, "stereo")
>>>> -            ||!strcmp(tag->key, "audiocodecid")
>>>> -            ||!strcmp(tag->key, "duration")
>>>> -            ||!strcmp(tag->key, "onMetaData")
>>>> -            ||!strcmp(tag->key, "datasize")
>>>> -            ||!strcmp(tag->key, "lasttimestamp")
>>>> -            ||!strcmp(tag->key, "totalframes")
>>>> -            ||!strcmp(tag->key, "hasAudio")
>>>> -            ||!strcmp(tag->key, "hasVideo")
>>>> -            ||!strcmp(tag->key, "hasCuePoints")
>>>> -            ||!strcmp(tag->key, "hasMetadata")
>>>> -            ||!strcmp(tag->key, "hasKeyframes")
>>>> -        ){
>>>> -            av_log(s, AV_LOG_DEBUG, "Ignoring metadata for %s\n",
>>>> tag->key);
>>>> -            continue;
>>>> -        }
>>>> -        put_amf_string(pb, tag->key);
>>>> -        avio_w8(pb, AMF_DATA_TYPE_STRING);
>>>> -        put_amf_string(pb, tag->value);
>>>> -        metadata_count++;
>>>> -    }
>>>> +
>>>> +    while (tag = av_dict_get(s->metadata, "", tag,
>>>> AV_DICT_IGNORE_SUFFIX))
>>>> +        write_user_metadata_tag(s, tag, pb, &metadata_count);
>>>> +    while (tag = av_dict_get(flv->meta_dict, "", tag,
>>>> AV_DICT_IGNORE_SUFFIX))
>>>> +        write_user_metadata_tag(s, tag, pb, &metadata_count);
>>>>
>>>>        if (write_duration_filesize) {
>>>>            put_amf_string(pb, "filesize");
>>>> @@ -718,6 +783,7 @@ static int flv_write_header(AVFormatContext *s)
>>>>        if (flv->flags & FLV_NO_METADATA) {
>>>>            pb->seekable = 0;
>>>>        } else {
>>>> +        read_metadata_from_file(s, 0, 1);
>>>>            write_metadata(s, 0);
>>>>        }
>>>>
>>>> @@ -880,6 +946,7 @@ static int flv_write_packet(AVFormatContext *s,
>>>> AVPacket *pkt)
>>>>        if (meta_upd_flag ||
>>>>            (flv->meta_period == FLV_META_AT_KF        && par->codec_type
>>>> == AVMEDIA_TYPE_VIDEO && (pkt->flags & AV_PKT_FLAG_KEY)) ||
>>>>            (flv->meta_period == FLV_META_EVERY_PACKET && par->codec_type
>>>> == AVMEDIA_TYPE_VIDEO )) {
>>>> +        read_metadata_from_file(s, ts, 0);
>>>>            write_metadata(s, ts);
>>>>            if (meta_upd_flag)
>>>>                s->event_flags &= ~AVSTREAM_EVENT_FLAG_METADATA_UPDATED;
>>>> @@ -1053,6 +1120,11 @@ static void flv_deinit(AVFormatContext *s)
>>>>        }
>>>>        flv->filepositions = flv->head_filepositions = NULL;
>>>>        flv->filepositions_count = 0;
>>>> +
>>>> +    if (flv->meta_dict)
>>>> +        av_dict_free(&flv->meta_dict);
>>>> +
>>>> +    avformat_close_input(&flv->meta_ctx);
>>>>    }
>>>>
>>>>    static const AVOption options[] = {
>>>> @@ -1066,6 +1138,7 @@ static const AVOption options[] = {
>>>>        { "at_start",      "only once at start",        0,
>>>> AV_OPT_TYPE_CONST, {.i64 = FLV_META_ONCE_AT_START}, INT_MIN, INT_MAX,
>>>> AV_OPT_FLAG_ENCODING_PARAM, "meta_period" },
>>>>        { "at_keyframes",  "with every video keyframe", 0,
>>>> AV_OPT_TYPE_CONST, {.i64 = FLV_META_AT_KF},         INT_MIN, INT_MAX,
>>>> AV_OPT_FLAG_ENCODING_PARAM, "meta_period" },
>>>>        { "every_packet",  "with every video packet",   0,
>>>> AV_OPT_TYPE_CONST, {.i64 = FLV_META_EVERY_PACKET},  INT_MIN, INT_MAX,
>>>> AV_OPT_FLAG_ENCODING_PARAM, "meta_period" },
>>>> +    { "meta_filename", "ffmetadata file to import metadata",
>>>> offsetof(FLVContext, meta_filename), AV_OPT_TYPE_STRING, {.str = NULL},
>>>> 0, 0, AV_OPT_FLAG_ENCODING_PARAM },
>>>>        { NULL },
>>>>    };
>>>>
>>> What does this achieve that can't be achieved by other means (namely by
>>> an API-user updating the relevant metadata dictionaries)?
>> This option together with meta_period is for CLI users. Similar to
>> drawtext's textfile + reload.
>> It's been used in production for over a year at a client. Think it will
>> be broadly useful.
> Not everything that is useful now is correct on long term.
>
> Muxer specific metadata handling by reading from files is big hack.

What would you suggest?

Regards,
Gyan

_______________________________________________
ffmpeg-devel mailing list
ffmpeg-devel@ffmpeg.org
https://ffmpeg.org/mailman/listinfo/ffmpeg-devel

To unsubscribe, visit link above, or email
ffmpeg-devel-request@ffmpeg.org with subject "unsubscribe".