[FFmpeg-devel] [PATCH v2 2/2] avformat/flvenc: add option to read metadata from file
Gyan Doshi
ffmpeg at gyani.pro
Sun Feb 5 12:01:41 EET 2023
On 2023-02-04 04:12 pm, Gyan Doshi wrote:
>
>
> On 2023-02-04 04:02 pm, Paul B Mahol wrote:
>> On 2/4/23, Gyan Doshi <ffmpeg at gyani.pro> wrote:
>>>
>>> On 2023-02-04 03:46 pm, Paul B Mahol wrote:
>>>> On 2/4/23, Gyan Doshi <ffmpeg at 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.
>>>>>>>
>>>>>>> + at 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(¤t_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(¤t_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(¤t_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?
>> There should be way to handle this generically see Anton's patch that
>> adds new AVOption type.
> Link?
So, I only found his patch to load filter options from files, but that's
only inside filtergraphs.
Anyway, this option is for dynamic metadata. Which means that even
meta_period would have to be a generic lavf option.
But most muxers don't emit metadata post-header. In fact, flvenc is the
only muxer that makes use of
AV***_EVENT_FLAG_METADATA_UPDATED. So, we will need a new AVFMT flag to
avoid reloading of file metadata in all other cases.
Seems overkill for something useful in one muxer.
Regards,
Gyan
More information about the ffmpeg-devel
mailing list