[FFmpeg-devel] [PATCH 4/4] avformat/mpegtsenc: use the correct stream_types and write HDMV descriptors for m2ts

[John Stebbins]

On Fri, 2020-04-10 at 21:44 +0200, Marton Balint wrote:
> Fixes ticket #2622.
> 
> Signed-off-by: Marton Balint <cus at passwd.hu>
> ---
>  libavformat/mpegtsenc.c | 58
> ++++++++++++++++++++++++++++++++++++++++++++++++-
>  1 file changed, 57 insertions(+), 1 deletion(-)
> 
> diff --git a/libavformat/mpegtsenc.c b/libavformat/mpegtsenc.c
> index add35aca89..b154675d60 100644
> --- a/libavformat/mpegtsenc.c
> +++ b/libavformat/mpegtsenc.c
> @@ -359,6 +359,54 @@ static int get_dvb_stream_type(AVFormatContext
> *s, AVStream *st)
>      return stream_type;
>  }
>  
> +static int get_m2ts_stream_type(AVFormatContext *s, AVStream *st)
> +{
> +    int stream_type;
> +
> +    switch (st->codecpar->codec_id) {
> +    case AV_CODEC_ID_MPEG2VIDEO:
> +        stream_type = STREAM_TYPE_VIDEO_MPEG2;
> +        break;
> +    case AV_CODEC_ID_H264:
> +        stream_type = STREAM_TYPE_VIDEO_H264;
> +        break;
> +    case AV_CODEC_ID_VC1:
> +        stream_type = STREAM_TYPE_VIDEO_VC1;
> +        break;
> +    case AV_CODEC_ID_HEVC:
> +        stream_type = STREAM_TYPE_VIDEO_HEVC;
> +        break;
> +    case AV_CODEC_ID_PCM_BLURAY:
> +        stream_type = 0x80;
> +        break;
> +    case AV_CODEC_ID_AC3:
> +        stream_type = 0x81;
> +        break;
> +    case AV_CODEC_ID_DTS:
> +        stream_type = (st->codecpar->channels > 6) ? 0x85 : 0x82;
> +        break;
> +    case AV_CODEC_ID_TRUEHD:
> +        stream_type = 0x83;
> +        break;
> +    case AV_CODEC_ID_EAC3:
> +        stream_type = 0x84;
> +        break;
> +    case AV_CODEC_ID_HDMV_PGS_SUBTITLE:
> +        stream_type = 0x90;
> +        break;
> +    case AV_CODEC_ID_HDMV_TEXT_SUBTITLE:
> +        stream_type = 0x92;
> +        break;
> +    default:
> +        av_log(s, AV_LOG_WARNING, "Stream %d, codec %s, is muxed as
> a private data stream "
> +               "and may not be recognized upon reading.\n", st-
> >index, avcodec_get_name(st->codecpar->codec_id));
> +        stream_type = STREAM_TYPE_PRIVATE_DATA;
> +        break;
> +    }
> +
> +    return stream_type;
> +}
> +
>  static int mpegts_write_pmt(AVFormatContext *s, MpegTSService
> *service)
>  {
>      MpegTSWrite *ts = s->priv_data;
> @@ -372,6 +420,14 @@ static int mpegts_write_pmt(AVFormatContext *s,
> MpegTSService *service)
>      q += 2; /* patched after */
>  
>      /* put program info here */
> +    if (ts->m2ts_mode) {
> +        put_registration_descriptor(&q, MKTAG('H', 'D', 'M', 'V'));
> +        *q++ = 0x88;        // descriptor_tag -
> hdmv_copy_control_descriptor
> +        *q++ = 0x04;        // descriptor_length
> +        put16(&q, 0x0fff);  // CA_System_ID
> +        *q++ = 0xfc;        // private_data_byte
> +        *q++ = 0xfc;        // private_data_byte
> +    }
>  
>      val = 0xf000 | (q - program_info_length_ptr - 2);
>      program_info_length_ptr[0] = val >> 8;
> @@ -401,7 +457,7 @@ static int mpegts_write_pmt(AVFormatContext *s,
> MpegTSService *service)
>              break;
>          }
>  
> -        stream_type = get_dvb_stream_type(s, st);
> +        stream_type = ts->m2ts_mode ? get_m2ts_stream_type(s, st) :
> get_dvb_stream_type(s, st);
>  
>          *q++ = stream_type;
>          put16(&q, 0xe000 | ts_st->pid);

I just wanted to comment on PGS subtitles.  PGS in Matroska have all
segments combined in one packet.  m2ts requires each segment in it's
own packet.  So remuxing PGS from Matroska to m2ts will not work
correctly without the addition of a bsf to split segments.  

I just happen to be starting testing of just such a bsf (I also wrote a
merge bsf to go the other way).  I hacked in some of you're patch to do
a smoke test converting back and forth between m2ts and Matroska. Split
bsf emits multiple packets for each input packet and merge consumes
multiple packets for each input packet.  So I also had to do
significant surgery on the auto bsf code in mux.c

After cleaning up the patches and doing some additional testing, I'll
be submitting these bsf.

I also have a PGS encoder working.  But, like dvbsub, it may need to
emit 2 packets for 1 input subtitle.  What's being done for dvbsub in
ffmpeg.c to work around this is ugly and I'd like to propose something
like avcodec_send_frame/avcodec_receive_packet interface for subtitles
to clean this up.  Would anyone object to an avcodec_send_subtitle and
just extend avcodec_receive_packet to handle subtitles?