[FFmpeg-devel] [PATCH 1/1] avformat/dashenc: Added #EXT-X-PROGRAM-DATE-TIME to HLS playlists

[Joep Admiraal]

Hi Karthick,

Thanks for looking into this.
I'll have a look at the suggested improvement.

Regards,
Joep

On Tue, Feb 26, 2019 at 6:25 AM Jeyapal, Karthick <kjeyapal at akamai.com>
wrote:

>
> On 2/22/19 12:25 PM, joepadmiraal wrote:
> > From: joepadmiraal <joep at groovytunes.nl>
> >
> > ---
> >  libavformat/dashenc.c | 29 ++++++++++++++++++++---------
> >  1 file changed, 20 insertions(+), 9 deletions(-)
> >
> > diff --git a/libavformat/dashenc.c b/libavformat/dashenc.c
> > index 37a7547b12..d0d0d6f030 100644
> > --- a/libavformat/dashenc.c
> > +++ b/libavformat/dashenc.c
> > @@ -122,6 +122,7 @@ typedef struct DASHContext {
> >      int64_t last_duration;
> >      int64_t total_duration;
> >      char availability_start_time[100];
> > +    int64_t start_time_ms;
> >      char dirname[1024];
> >      const char *single_file_name;  /* file names as specified in
> options */
> >      const char *init_seg_name;
> > @@ -432,6 +433,8 @@ static void write_hls_media_playlist(OutputStream
> *os, AVFormatContext *s,
> >      const char *proto = avio_find_protocol_name(c->dirname);
> >      int use_rename = proto && !strcmp(proto, "file");
> >      int i, start_index, start_number;
> > +    time_t start_time_s = c->start_time_ms / 1000;
> > +    double prog_date_time = start_time_s;
> >
> >      get_start_index_number(os, c, &start_index, &start_number);
> >
> > @@ -464,16 +467,21 @@ static void write_hls_media_playlist(OutputStream
> *os, AVFormatContext *s,
> >      ff_hls_write_init_file(c->m3u8_out, os->initfile, c->single_file,
> >                             os->init_range_length, os->init_start_pos);
> >
> > -    for (i = start_index; i < os->nb_segments; i++) {
> > +    for (i = 0; i < os->nb_segments; i++) {
> >          Segment *seg = os->segments[i];
> > -        ret = ff_hls_write_file_entry(c->m3u8_out, 0, c->single_file,
> > -                                (double) seg->duration / timescale, 0,
> > -                                seg->range_length, seg->start_pos, NULL,
> > -                                c->single_file ? os->initfile :
> seg->file,
> > -                                NULL);
> > -        if (ret < 0) {
> > -            av_log(os->ctx, AV_LOG_WARNING, "ff_hls_write_file_entry
> get error\n");
> > +        double duration = (double) seg->duration / timescale;
> > +        if (i >= start_index) {
> > +            double prog_date_time_copy = prog_date_time;
> > +            ret = ff_hls_write_file_entry(c->m3u8_out, 0,
> c->single_file,
> > +                                    duration, 0,
> > +                                    seg->range_length, seg->start_pos,
> NULL,
> > +                                    c->single_file ? os->initfile :
> seg->file,
> > +                                    &prog_date_time_copy);
> > +            if (ret < 0) {
> > +                av_log(os->ctx, AV_LOG_WARNING,
> "ff_hls_write_file_entry get error\n");
> > +            }
> >          }
> > +        prog_date_time += duration;
> Hi Joep,
>
> It is great to see that you are submitting your changes back to the
> community for improving dashenc features.
> Really appreciate your effort on this. Thank you very much.
>
> In terms for the actual changes, I have review comment. The algorithmic
> complexity for calculating prog_date_time will increase with the time
> software has been running continuously. For example, if the software has
> been running for seven days continuously (with window_size parameter
> enabled for a live input), it will take smaller cpu time to compute
> prog_date_time on the first day, and much longer time on the seventh day.
> Such kind of algorithms affect the stability of the software in
> long-duration runs for live input. Could you please modify the algorithm so
> that the prog_date_time calculation's complexity doesn't increase linearly
> with time.
>
> Regards,
> Karthick
> >      }
> >
> >      if (prefetch_url)
> > @@ -1583,9 +1591,12 @@ static int dash_write_packet(AVFormatContext *s,
> AVPacket *pkt)
> >          os->first_pts = pkt->pts;
> >      os->last_pts = pkt->pts;
> >
> > -    if (!c->availability_start_time[0])
> > +    if (!c->availability_start_time[0]) {
> > +        int64_t start_time_us = av_gettime();
> > +        c->start_time_ms = start_time_us / 1000;
> >          format_date_now(c->availability_start_time,
> >                          sizeof(c->availability_start_time));
> > +    }
> >
> >      if (!os->availability_time_offset && pkt->duration) {
> >          int64_t frame_duration = av_rescale_q(pkt->duration,
> st->time_base,
>
> _______________________________________________
> ffmpeg-devel mailing list
> ffmpeg-devel at ffmpeg.org
> https://ffmpeg.org/mailman/listinfo/ffmpeg-devel
>