[Ffmpeg-devel-irc] ffmpeg-devel.log.20161019

[burek]

[01:40:48 CEST] <cone-095> ffmpeg 03James Almer 07master:47ee6f1a59b1: avformat/matroskadec: support parsing Chroma Location elements
[02:01:01 CEST] <cone-095> ffmpeg 03James Almer 07n3.0.4:HEAD: avformat/matroskadec: support parsing Chroma Location elements
[03:11:44 CEST] <cone-095> ffmpeg 03Aman Gupta 07master:f45d5e07ddbd: lavc/videotoolboxenc: skip SEI allocation when side data is not present
[03:29:23 CEST] <atomnuker> alright, how the hell does this work
[03:29:33 CEST] <atomnuker> ff_af_queue_remove gets called with an identical frame size
[03:30:11 CEST] <atomnuker> yet on the last frame there are samples left (this is okay, that's what's meant to happen)
[03:31:18 CEST] <atomnuker> but if I copt exactly what goes on in another encoder I get 0 samples left
[03:40:34 CEST] <atomnuker> nevermind, missing avctx->initial_padding
[04:10:54 CEST] <cone-095> ffmpeg 03James Almer 07master:ea201ad14cde: avformat/mov: pass the demuxer's AVFormatContext to avpriv_request_sample()
[04:58:56 CEST] <cone-095> ffmpeg 03James Almer 07master:f5cfc0cc54ce: avformat/matroskadec: clarify some Colour element defaults using enum values
[04:58:57 CEST] <cone-095> ffmpeg 03James Almer 07master:0ad71ed6f98e: avformat/matroskadec: fix BitsPerChannel element's default value
[09:10:43 CEST] <cone-500> ffmpeg 03Muhammad Faiz 07master:23b6f880d6d7: avfilter/firequalizer: add cubic_interpolate function on gain
[09:14:55 CEST] <Matador> Hi
[09:16:10 CEST] <Matador> Anyone else having issues on QSV (libmfx) ?  Couple bug reports already on it, hoping something can be done :|
[09:16:18 CEST] <Matador> https://trac.ffmpeg.org/ticket/5859
[09:16:25 CEST] <Matador> https://trac.ffmpeg.org/ticket/5848
[09:26:51 CEST] <Chloe> What's the status on the QSV revert & sync with libav?
[10:30:54 CEST] <jkqxz> Matador:  First one looks like user error; MSS 2017 does work.  At a guess, they need to look at their dynamic linker setup.
[10:31:10 CEST] <jkqxz> Don't know about the second one, I haven't used it on Windows.
[10:34:18 CEST] <BtbN> I also just wouldn't bother with QSV on Linux. It's bad and pointless.
[11:00:31 CEST] <cone-500> ffmpeg 03Matthieu Bouron 07master:b8c158a4eddb: lavc/mediacodec_wrapper: do not discard codecs reporting they do not support any profile
[11:00:32 CEST] <cone-500> ffmpeg 03Matthieu Bouron 07master:0f7fce87ea98: lavc: add vp8/vp9 mediacodec decoders
[11:00:33 CEST] <cone-500> ffmpeg 03Matthieu Bouron 07master:f62c54456db0: lavc: add mpeg4 mediacodec decoder
[11:00:34 CEST] <cone-500> ffmpeg 03Matthieu Bouron 07master:d5082a2ce776: lavc/mediacodec: use more meaningful filenames
[13:02:51 CEST] <cone-500> ffmpeg 03Sven C. Dack 07master:da4d0fa86b48: avcodec/nvenc: add test for Temporal AQ support
[13:02:52 CEST] <cone-500> ffmpeg 03Sven C. Dack 07master:aebbcb2706ab: avcodec/nvenc_hevc: Added missing option -temporal_aq
[13:02:53 CEST] <cone-500> ffmpeg 03Sven C. Dack 07master:1aa8fa418a82: avfilter/vf_scale_npp: select cubic and lanczos as alternative where super-sampling is not supported
[13:04:46 CEST] <thegeek_> I'm using libavcodec and I have a out-of-bounds access in ffv1enc.c, choose_rct_params()
[13:05:24 CEST] <thegeek_> I've enabled version 4 (experimental), which is why that code path is used
[13:07:20 CEST] <thegeek_> this is the offending statement: 
[13:07:21 CEST] <thegeek_> unsigned v = *((const uint32_t*)(src[0] + x*4 + stride[0]*y))
[13:07:53 CEST] <thegeek_> I'm using 16 slices on a 640x480 image, which means that each slice is 160x120
[13:08:32 CEST] <thegeek_> this function iterates through all pixels in the picture (y = 0; y<h) (x=0;x<w)
[13:09:06 CEST] <thegeek_> it always crashes when y==119 and x == 120
[13:10:04 CEST] <thegeek_> the buffers it operates on are not allocated by me as I do not pass in a reference-counted AVFrame
[13:10:27 CEST] <thegeek_> it seems to me that the code reads out of bounds on memory that I have not allocated
[13:10:37 CEST] <thegeek_> which could be a bug
[13:27:45 CEST] <cone-500> ffmpeg 03Jean Caille
 07master:6498549fd75f: lavf/mov: support gopro hero moments udta tag
[13:27:46 CEST] <cone-500> ffmpeg 03Michael Niedermayer 07master:18ad44d14548: avformat/movenc: Skip duration check for discontinuous fragments
[13:27:47 CEST] <cone-500> ffmpeg 03Derek Buitenhuis 07master:dfe7e5501d4a: FATE: Add test for muxing discontinuous MP4 fragments with large timestamps
[13:31:21 CEST] <michaelni> thegeek_, how can this be reproduced?
[13:40:57 CEST] <thegeek_> michaelni: my current repro is tied to my custom code/application
[13:41:12 CEST] <thegeek_> I shall try to create a repro with ffmpeg
[13:41:35 CEST] <thegeek_> but I was wondering if this code is sound or very experimental
[13:41:50 CEST] <thegeek_> I'm using v4 as it allocates much less memory 
[17:41:38 CEST] <jacereda> Looks like avienc.c doesn't write the "strf" tag for AVMEDIA_TYPE_DATA... any reason for that? if that tag is skipped, bad things will happend for data streams in avidec.c (like not setting codec_type). Or am I missing something?
[17:56:18 CEST] <jacereda> OTOH, the MKTAG('s', 't', 'r', 'f') handler in avidec.c has an 'if (!size) break;' that will also avoid setting codec_type.
[18:01:08 CEST] <jacereda> The commit message for that bit says "Do not try to parse empty strf tags.", but I fail to see why the empty strf should be problematic.
[20:56:07 CEST] <Matador> qsv is a huge fail now on Windows + Ubuntu
[20:56:10 CEST] <Matador> it worked months ago
[21:57:22 CEST] <Compn> Matador : use months ago ffmpeg
[21:57:30 CEST] <Compn> i heard the merges may have borked it ?
[21:59:07 CEST] <JEEB> not sure if it was the merges that borked it, but IIRC the current state of it @ Libav was better than FFmpeg's (except for the nobody-cares MJPEG decoder)
[21:59:14 CEST] <JEEB> or was it encoder as well?
[21:59:54 CEST] <JEEB> in any case it IIRC was decided that since the maintainer didn't pop up here we'd reset to Libav's version during the next merges
[22:00:06 CEST] <JEEB> that's just my understanding of things, though
[22:03:24 CEST] <Chloe> JEEB: I asked about the status of that yesterday but no response.
[22:04:04 CEST] <rcombs> the MJPEG decoder might be useful if lavc didn't fall apart on moderately large frame sizes
[22:04:35 CEST] <JEEB> right
[22:07:49 CEST] <rcombs> lol ints
[23:15:00 CEST] <cone-736> ffmpeg 03Andreas Cadhalpun 07master:a5b4476a602f: avformat: prevent triggering request_probe assert in ff_read_packet
[23:25:19 CEST] <tmm1> is there a filter that does the opposite of h264_mp4toannexb
[23:26:02 CEST] <nevcairiel> no
[23:26:09 CEST] <JEEB> I think the mov muxer did something like that
[23:26:19 CEST] <nevcairiel> the opposite is trivial
[23:26:30 CEST] <nevcairiel> which is why muxers that need it usually just do it internally
[23:28:47 CEST] <nevcairiel> i think w e have some helper function they all call
[23:29:46 CEST] <tmm1> cool i'll check it out, thanks
[23:40:29 CEST] <cone-736> ffmpeg 03Carl Eugen Hoyos 07master:d8716864f967: lavf/riff: Document the interleaving of VSM4 video.
[00:00:00 CEST] --- Thu Oct 20 2016