> On 16 Jul 2020, at 10:36, Dongwon Kim <[email protected]> wrote: > > > HLS spec RFC 8216(https://tools.ietf.org/html/rfc8216) version 7 added > supporting fMP4(CMAF) format for segment. However, when requesting seek > operation, > the hls demuxer doesn't work properly due to previous implementation of the > HLS demuxer was only for supporting MPEG-TS format for segment. > > So, I added procedure to reopen segment format on HLS demuxer when seeking > requested. It's quite similar to MPEG-DASH demuxer(libavformat/dashdec.c) > seeking procedure, as you know, MPEG-DASH also support fMP4(CMAF) format for > fragment as default. Refer to > https://github.com/FFmpeg/FFmpeg/blob/master/libavformat/dashdec.c#L1904. > > > Please test with below command: > > Server: > $ ffmpeg -y -i {input} -vcodec libx264 \ > -force_key_frames 30 \ > -sc_threshold 0 \ > -acodec aac \ > -hls_segment_type fmp4 \ > -hls_time 6 \ > -hls_playlist_type vod \ > -map 0:a -map 0:v -f hls \ > -var_stream_map "a:0,agroup:aud,default:yes v:0,agroup:aud" \ > -hls_segment_filename "segment_%v_%03d.m4s" \ > -master_pl_name playlist.m3u8 \ > hlscmaf_%v.m3u8 > > > Player: > $ ffplay playlist.m3u8 > > > > On 20. 7. 15. 오후 4:20, Dongwon Kim wrote: >> Signed-off-by: Dongwon Kim <[email protected]> >> --- >> libavformat/hls.c | 67 +++++++++++++++++++++++++++++++++++++++++++++++ >> 1 file changed, 67 insertions(+) >> >> diff --git a/libavformat/hls.c b/libavformat/hls.c >> index ba17c4ed96..561b42ea6b 100644 >> --- a/libavformat/hls.c >> +++ b/libavformat/hls.c >> @@ -2139,6 +2139,68 @@ static int compare_ts_with_wrapdetect(int64_t ts_a, >> struct playlist *pls_a, >> return av_compare_mod(scaled_ts_a, scaled_ts_b, 1LL << 33); >> } >> +static int reopen_demux_for_component(AVFormatContext *s, struct playlist >> *pls) >> +{ >> + ff_const59 AVInputFormat *in_fmt = NULL; >> + AVDictionary *in_fmt_opts = NULL; >> + uint8_t *avio_ctx_buffer = NULL; >> + int ret = 0; >> + >> + if (pls->ctx) { >> + av_freep(&pls->pb.buffer); >> + memset(&pls->pb, 0x00, sizeof(AVIOContext)); >> + pls->ctx->pb = NULL; >> + avformat_close_input(&pls->ctx); >> + pls->ctx = NULL; >> + } >> + >> + if (ff_check_interrupt(&s->interrupt_callback)) { >> + ret = AVERROR_EXIT; >> + goto fail; >> + } >> + >> + if (!(pls->ctx = avformat_alloc_context())) { >> + ret = AVERROR(ENOMEM); >> + goto fail; >> + } >> + >> + avio_ctx_buffer = av_malloc(INITIAL_BUFFER_SIZE); >> + if (!avio_ctx_buffer ) { >> + ret = AVERROR(ENOMEM); >> + avformat_free_context(pls->ctx); >> + pls->ctx = NULL; >> + goto fail; >> + } >> + ffio_init_context(&pls->pb, avio_ctx_buffer , INITIAL_BUFFER_SIZE, 0, >> pls, read_data, NULL, NULL); >> + pls->pb.seekable = 0; >> + >> + if ((ret = ff_copy_whiteblacklists(pls->ctx, s)) < 0) >> + goto fail; >> + >> + pls->ctx->flags = AVFMT_FLAG_CUSTOM_IO; >> + ret = av_probe_input_buffer(&pls->pb, &in_fmt, "", NULL, 0, 0); >> + if (ret < 0) { >> + av_log(s, AV_LOG_ERROR, "Error when loading segment, playlist >> %d\n", pls->cur_seq_no); >> + avformat_free_context(pls->ctx); >> + pls->ctx = NULL; >> + goto fail; >> + } >> + >> + pls->ctx->pb = &pls->pb; >> + pls->ctx->io_open = nested_io_open; >> + >> + ret = avformat_open_input(&pls->ctx, "", in_fmt, &in_fmt_opts); >> + av_dict_free(&in_fmt_opts); >> + if (ret < 0) >> + goto fail; >> + ret = avformat_find_stream_info(pls->ctx, NULL); >> + if (ret < 0) >> + goto fail; >> + >> +fail: >> + return ret; >> +} >> + >> static int hls_read_packet(AVFormatContext *s, AVPacket *pkt) >> { >> HLSContext *c = s->priv_data; >> @@ -2351,6 +2413,11 @@ static int hls_read_seek(AVFormatContext *s, int >> stream_index, >> pls->seek_stream_index = -1; >> pls->seek_flags |= AVSEEK_FLAG_ANY; >> } >> + >> + if (pls->ctx && pls->ctx->iformat && >> strcmp(pls->ctx->iformat->name, "mov,mp4,m4a,3gp,3g2,mj2") == 0) { >> + pls->init_sec_buf_read_offset = 0; >> + reopen_demux_for_component(s, pls); >> + } >> } >> c->cur_timestamp = seek_timestamp; > > _______________________________________________ > ffmpeg-devel mailing list > [email protected] > https://ffmpeg.org/mailman/listinfo/ffmpeg-devel > > To unsubscribe, visit link above, or email > [email protected] with subject "unsubscribe”.
I have previously submitted another means of achieving this here: https://patchwork.ffmpeg.org/project/ffmpeg/list/?series=1018 <https://patchwork.ffmpeg.org/project/ffmpeg/list/?series=1018> My approach was to mirror in mp4 demux the same support that mpeg2ts has for detecting a position reset on the stream. And to improve the seek semantics in HLS demux. Some older history here: https://trac.ffmpeg.org/ticket/7485 <https://trac.ffmpeg.org/ticket/7485> _______________________________________________ ffmpeg-devel mailing list [email protected] https://ffmpeg.org/mailman/listinfo/ffmpeg-devel To unsubscribe, visit link above, or email [email protected] with subject "unsubscribe".
