[Ffmpeg-devel-irc] ffmpeg-devel.log.20191122

[burek]

[00:04:28 CET] <cone-417> ffmpeg 03Carl Eugen Hoyos 07release/4.0:5e98ede2da98: lavc/tableprint_vlc: Remove avpriv_request_sample() from included files.
[00:04:29 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:a6eb5e1f7086: avformat/mp3dec: Check that the frame fits within the probe buffer
[00:04:30 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:17d5c21b3d12: avcodec/sonic: Fix integer overflow in predictor_calc_error()
[00:04:31 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:6b430f066b9d: avcodec/apedec: Fixes integer overflow of res+*data in do_apply_filter()
[00:04:32 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:61620896a1ac: avcodec/qdm2: The smallest header seems to have 2 bytes so treat 1 as invalid
[00:04:33 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:2850004c6d3a: avcodec/vc1_block: Fix undefined behavior in ac prediction rescaling
[00:04:34 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:aca0578002d9: avcodec/wmadec: Require previous exponents for reuse
[00:04:35 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:38cfa9d5f974: avcodec/ralf: use multiply instead of shift to avoid undefined behavior in decode_block()
[00:04:36 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:ed04ecd2d3c3: avcodec/atrac3: Check for huge block aligns
[00:04:37 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:70bf2bdf3112: avcodec/iff: Move index use after check in decodeplane8()
[00:04:38 CET] <cone-417> ffmpeg 03Michael Niedermayer 07release/4.0:26e1d0d015bb: Changelog: update
[04:09:25 CET] <cone-844> ffmpeg 03Andreas Rheinhardt 07master:27c6c925344e: avformat/av1: Fix leak of dynamic buffer in case of parsing failure
[11:53:07 CET] <rcombs> okay I have an extremely dumb broadcast case
[11:53:33 CET] <rcombs> so SCC files don't have timestamps, they have SMPTE timecodes
[11:53:53 CET] <rcombs> which can be either drop-frame or non-drop-frame
[11:56:03 CET] <rcombs> as far as I can tell, what that ultimately means is "if the caption timecodes are non-drop-frame, and the video has a fractional frame rate, then they need to be adjusted by adding 2/30 of a second per minute except on minutes divisible by 10"
[11:57:03 CET] <rcombs> but if the video has an integer frame rate, then they don't need adjustment, or possibly would need the opposite adjustment if the captions are drop-frame
[11:59:01 CET] <rcombs> note that this isn't quite the same as being stretched by 1.001/1
[12:03:22 CET] <Dmitri_Ovch> Hi! I sent the small patch (https://patchwork.ffmpeg.org/patch/16382/) fixing constant QP settings for I, P pictures. I also sent a patch, (https://patchwork.ffmpeg.org/patch/16270/) proposed by Sitan Liu. This is a very small but useful patch that corrects the profile level option on AMF HEVC. Earlier, allowing the user to select the engine for AMF initialisation (https://patchwork.ffmpeg.org/patch/16015/). 
[12:03:22 CET] <Dmitri_Ovch> Could you review these patches?
[12:05:44 CET] <kierank> rcombs: yes
[12:06:00 CET] <kierank> what you do is run the video
[12:06:02 CET] <kierank> and match the timecodes
[12:06:14 CET] <kierank> iirc there is already code to do this in lavu otherwise it's trivial to write
[12:06:45 CET] <rcombs> kierank: how would that all fit together in ffmpeg, though
[12:06:52 CET] <kierank> doesn't
[12:26:47 CET] <rcombs> kierank: so here's one thing I'm not clear on: say if the frame rate is (exactly) 24; are the frame numbers in SCC supposed to be in 1/24s or 1/30s of a second?
[12:28:18 CET] <kierank> Whatever the timecode rate is
[12:28:18 CET] <rcombs> and who the hell decided to use this batshit format for subtitle timestamps
[12:28:33 CET] <rcombs> is that the same as the frame rate or
[12:28:47 CET] <kierank> You don't know really
[12:29:00 CET] <kierank> This is some 1980s format on floppy disks
[12:29:07 CET] <kierank> 5 inch floppies
[12:29:31 CET] <rcombs> well for some reason it still exists in 2019 with HD video
[12:29:44 CET] <kierank> Yes, you have to know the frame rate a priori
[12:30:02 CET] <rcombs> so the timecode rate is the video's frame rate, then?
[12:31:00 CET] <rcombs> just, uh, rounded to the nearest integer, I guess????
[12:31:30 CET] <rcombs> trying to wrap my head around NDF timecodes on fractional-frame-rate video
[12:32:18 CET] <rcombs> I guess, like, 01:00:00:00 actually means "1*60*60*24 video frames after 00:00:00:00"
[12:32:52 CET] <JEEB> oh boy, SCC
[12:32:55 CET] <JEEB> I have looked into that once
[12:34:24 CET] <rcombs> which is sane for 24fps but for 24/1.001 that ends up actually meaning 01:00:03.06?
[12:34:53 CET] <rcombs> so& to get timestamps, you play back the video and just count frames?
[12:35:33 CET] <rcombs> and when you hit frame 24, whatever timestamp that has is timecode 00:00:01:00?
[12:35:49 CET] <rcombs> mmmmmmmmmmmm
[12:37:19 CET] <rcombs> I think I should only have to care about CFR, so I can _probably_ just convert to frame numbers by multiplying seconds by the nearest-integer rate + adding frames, then dividing by the real rate?
[12:37:56 CET] <rcombs> if I have this all right, would anyone object to a "video_rate" AVOption in sccdec.c
[12:40:47 CET] <rcombs> kierank: ^ sanity-check me?
[12:42:07 CET] <kierank> Afaik DF only matters for 29.97
[12:42:15 CET] <kierank> Not sure there is DF for 23.976
[12:43:28 CET] <kierank> But haven't looked at this for years
[12:43:46 CET] <kierank> Traditionally some content starts at 01:00:00:00 so you can have a pre slate
[12:43:52 CET] <rcombs> I'm dealing with NDF here
[12:44:18 CET] <rcombs> or at least the SCC formatting claims it's NDF
[12:44:43 CET] <rcombs> I _think_ if it was DF this would mostly just work, with a small amount of periodically-corrected drift
[12:44:57 CET] <kierank> NDF is just run as normal
[12:44:59 CET] <rcombs> (it'd be dumb, but it wouldn't substantially desync over time)
[12:45:19 CET] <kierank> You need to know the start timecode
[12:45:28 CET] <kierank> Usually 00:00:00:00 or 01:00:00:00
[12:45:38 CET] <kierank> Many places just assume the former
[12:46:16 CET] <rcombs> the problem I'm seeing is that when decoded using lavf's naïve "assume 30fps" demuxer, I get timestamps that desync from the video at a rate of .1%
[15:28:35 CET] <cone-210> ffmpeg 03Gyan Doshi 07master:1b78da449e20: doc/filters: complete and correct vmafmotion section
[16:07:25 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:c36e72ed279b: avfilter/af_aiir: check for stability
[16:07:26 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:e169d3756e92: avfilter/af_aiir: factor out response calculation
[16:10:15 CET] <rcombs> how should assenc handle multiple rects?
[16:10:58 CET] <rcombs> optimally it'd output multiple packets, but there's no way to do that with subtitles (which still, for some reason, use an entirely different API from audio and video)
[17:49:26 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:2f5fb9e60fa6: avfilter/af_aiir: make a/b coefficients array
[17:49:27 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:eecc45cea50f: avfilter/af_aiir: add missing normalization of biquads gains
[20:25:59 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:f2a01b4c8bf6: avfilter/af_aiir: fix biquads normalization
[21:07:12 CET] <cone-210> ffmpeg 03Andriy Gelman 07master:ccc83ff11e07: avcodec/v4l2_m2m_dec: Fix decoding on Odroid XU4
[21:07:13 CET] <cone-210> ffmpeg 03Andriy Gelman 07master:e759fbfbc477: avcodec/v4l2_context: Fix indentation
[21:16:03 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:f46b04c4c328: avfilter/af_biquads: add new normalize/n option
[21:16:04 CET] <cone-210> ffmpeg 03Paul B Mahol 07master:89aa1342b1e5: avfilter/af_aiir: normalize biquads only if divisor is big enough
[22:42:57 CET] <cone-210> ffmpeg 03Paul B Mahol 07n4.0.5:HEAD: avfilter/af_aiir: normalize biquads only if divisor is big enough
[23:30:50 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:0a4d2b5f674e: avformat/mp3dec: Check that the frame fits within the probe buffer
[23:30:51 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:c2f81cfc78fe: avcodec/sonic: Fix integer overflow in predictor_calc_error()
[23:30:52 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:5222756b36fe: avcodec/apedec: Fixes integer overflow of res+*data in do_apply_filter()
[23:30:53 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:0e96cffff6e4: avcodec/qdm2: The smallest header seems to have 2 bytes so treat 1 as invalid
[23:30:54 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:313af2294257: avcodec/vc1_block: Fix undefined behavior in ac prediction rescaling
[23:30:55 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:b385b3476141: avcodec/wmadec: Require previous exponents for reuse
[23:30:56 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:978aa2ae5087: avcodec/ralf: use multiply instead of shift to avoid undefined behavior in decode_block()
[23:30:57 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:eabece8639dd: avcodec/atrac3: Check for huge block aligns
[23:30:58 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:b9735a0271a9: avcodec/iff: Move index use after check in decodeplane8()
[23:30:59 CET] <cone-210> ffmpeg 03Michael Niedermayer 07release/3.4:e5baac772ccb: Changelog: update
[00:00:00 CET] --- Sat Nov 23 2019