[Ffmpeg-devel-irc] ffmpeg-devel.log.20191101

Sat Nov 2 03:05:03 EET 2019

[00:24:38 CET] <philipl> So, I'm continuing to look at improving the handling of HLS discontinuities. Clearly, I have problems. Discontinuous segments are marked in the m3u8. That's great. One idea would be to mark the first packet out of the segment as discontinuous (a new packet flag) and then we'd consider how to treat that in other places. But there's a pretty thick layer between segments and packets. We use avio to 
[00:24:44 CET] <philipl> create the appearance of a continuous stream and packets are read out of that. This means that at the time a segment is opened, and we detect a discontinuity, we're actually ahead of packets being read out of the stream. The next packet isn't the one that needs to be marked; it might be a few packets later.
[00:25:51 CET] <philipl> I can't see a way to carry the discontinuity marker out-of-band. Do I have to insert an artificial 'barrier' into the avio stream? Seems that way.
[00:26:10 CET] <nicolas17> you mean when a discontinuity is found in the m3u8 you're still halfway processing packets in the previous .ts?
[00:33:28 CET] <philipl> Well, at least for ffmpeg.c, av_read_packet seems to be a couple of packets behind.
[00:34:01 CET] <philipl> And I can't attach a flag to the packet until it's read.
[00:43:27 CET] <nicolas17> I thought of adding extra readahead for the (misguided?) use-case of transcoding from a remote hls vod (for fast encodes, the latency of downloading new segments becomes a bottleneck) and it didn't seem easy to implement
[01:18:00 CET] <philipl> Looks like what we'd really have to do is modify the first packet out of the stream to set the mpeg ts discontinuity indicator - it's the only inline way to do it, and probably semantically reasonable.
[01:18:13 CET] <philipl> Then the mpegts demuxer can set the packet flag.
[01:22:49 CET] <philipl> The hls demuxer feeds everything into a new stream that goes into the mpegts demuxer and that's what returns the packets. so it's completely decoupled.
[10:30:56 CET] <cone-972> ffmpeg 03Steven Liu 07master:2dd71bf95e36: doc/filters: fix overlay_opencl document typo
[11:38:09 CET] <JEEB> > color callback is unsigned int > color is "int" in style
[13:40:04 CET] <cone-927> ffmpeg 03James Cowgill 07master:bb718d11ed99: avcodec/libtwolame: fix mono default bitrate
[15:58:17 CET] <JEEB> This text should be bold, italics and underline This text should be small and green
[15:58:23 CET] <JEEB> welp, now my code generates :things:
[20:00:37 CET] <cone-187> ffmpeg 03Mattias Wadman 07master:e447a4d112bc: libavformat/flacenc: reject too big picture blocks
[20:00:37 CET] <cone-187> ffmpeg 03Limin Wang 07master:a5e24be52ab0: swscale/swscale_unscaled: fix gbrap10be md5 different on big endian system
[20:00:37 CET] <cone-187> ffmpeg 03Limin Wang 07master:99fc5059fbee: avfilter/vf_lut3d: simplify code
[20:00:37 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:f011572e66c8: avcodec/utils: Check block_align
[20:32:25 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:9ee5096068f6: avcodec/truemotion2: Fix several integer overflows with *Yo, *Uo, *Vo
[20:32:26 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:2b655f55eaf0: avcodec/truemotion2: Fix several integer overflows in tm2_low_res_block()
[20:32:27 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:7730bacb413f: avcodec/aacdec_template: Check samplerate
[20:32:28 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:47a017aaaae3: avcodec/rv10: Check input for minimal frame size
[20:32:29 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:98a257c3235b: avcodec/aptx: Check the number of channels
[20:32:30 CET] <cone-187> ffmpeg 03Michael Niedermayer 07master:fd3ee7a92e92: tools/target_dec_fuzzer: Adjust threshold for G2M
[23:20:30 CET] <cone-187> ffmpeg 03Andriy Gelman 07master:1aec1183f3e8: avcodec/v4l2_buffers: Fix infinite loop
[23:20:31 CET] <cone-187> ffmpeg 03Andriy Gelman 07master:53c21c2d6beb: avcodec/v4l2_m2m: Fix indentation
[23:22:13 CET] <tmm1> i'm trying to teach ffprobe about FF_CODEC_PROPERTY_CLOSED_CAPTIONS but it appears not be set inside show_stream()
[00:00:00 CET] --- Sat Nov 2 2019