[Ffmpeg-devel-irc] ffmpeg-devel.log.20150219

burek burek021 at gmail.com
Fri Feb 20 02:05:02 CET 2015

[01:59] <cone-30> ffmpeg.git 03Michael Niedermayer 07master:9d88be68733b: avfilter/vf_subtitles: Add () to protect the argument of the AA() macro
[02:26] <rcombs> \o/ another one!
[03:50] <cone-30> ffmpeg.git 03Michael Niedermayer 07master:f144e4ddff11: avcodec/wmalosslessdec: use sizeof() instead of a literal number
[03:58] <compn> andlabs : did you ask on ffmpeg forum ? they have lots of windows users there
[03:58] <compn> i've used regular pipes o nwindows but not named pipes as those are nasty.
[03:59] <compn> theres two for some reason , http://ffmpeg.gusari.org/ and http://ffmpeg.zeranoe.com/forum/
[04:00] <compn> not sure which is better :P
[04:00] <compn> i think this is what i was thinking of though, Using FFmpeg In Windows Software http://ffmpeg.zeranoe.com/forum/viewforum.php?f=15
[10:38] <pross> wm4: how could you forget .bin!
[10:42] <wm4> pross: the bintext demuxer has pretty bad effects when people try to naively open .bin (with anything that uses libavformat), so I blacklisted it in my code
[10:43] <wm4> I wish I hadn't to do these things
[10:45] <nevcairiel> extension matching on a name like .bin seems risky at best
[10:46] <pross> fair enough, it is a 'raw format' (to display you would just memcpy to the IBM PC textmode memory region)
[10:47] <pross> some files include metadata, so they could be probed
[10:54] <wm4> pross: then libavformat shouldn't open it, not with a probe score that indicates "reliable" (well I'm not so sure right now how this stuff works)
[10:55] <pross> agree
[10:56] <wm4> ok AVPROBE_SCORE_MAX/4 + 1 is the recommended "safe" score (derived from "#define AVPROBE_SCORE_RETRY (AVPROBE_SCORE_MAX/4)")
[10:56] <wm4> the numeric value is 26
[10:56] <wm4> and probing .bin returns a score of 50
[10:57] <wm4> that seems wrong...
[11:00] <cone-885> ffmpeg.git 03Himangi Saraogi 07master:0f532fd3f9b9: rv40: Return more meaningful error codes
[11:00] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:c1ff2cdbca2b: Merge commit '0f532fd3f9b9ca974cea0feb8ddcaf4a7f3b23c5'
[11:01] <pross> yep. i will investigate this on weekend.
[11:18] <wm4> anyway, it sets AVPROBE_SCORE_EXTENSION if the extension matches
[11:18] <wm4> this sounds wrong
[11:18] <wm4> AVPROBE_SCORE_EXTENSION hasn't been true to its name for years
[11:19] <wm4> except this code has been around since 2002 in this form
[11:19] Action: wm4 scratches head
[11:19] <wm4> it's handled differently depending on whether a read_probe callback is available
[11:20] <wm4> with read_probe -> extension generally sets prob score to 1 at most
[11:20] <wm4> without read_probe -> extension sets score to AVPROBE_SCORE_EXTENSION
[11:20] <wm4> seems very unbalanced
[11:20] <cone-885> ffmpeg.git 03Clément BSsch 07master:2280552057bb: avfilter/palettegen: raise cache size from 64k to 512k
[11:21] <nevcairiel> that kind of makes sense. if there is a (reliable) probe function, the extension should have much less value
[11:23] <wm4> in the end, the problem is that the .bin extension is too generic
[11:23] <wm4> there must be hundreds of file formats that use .bin
[11:24] <wm4> many of them without a header
[11:24] <wm4> so why would bintext get it?
[14:22] <compn> wm4 : i agree to remove .bin from extension list
[14:22] <compn> someone tell wm4 to submit patch
[14:23] <compn> or just commit fix
[15:43] <cone-885> ffmpeg.git 03Kevin Wheatley 07master:fb3fb1d0d4bb: avformat/mov: Add simple ACLR atom reading to set the color range of the incomming track for codec's like DNxHD that utilise AVID's proprietary atom.
[16:15] <ubitux> michaelni: you can upload without commiting right?
[16:15] <ubitux> i mean you can just commit locally and git archive to make a tarball
[16:15] <ubitux> which you can upload
[16:15] <ubitux> (re: coverity model file)
[16:26] <michaelni> ubitux, you are coverity admin too, i know as much as you do, and AFAIK the model file is just set in the web interface and i guess probably used for the next build, theres also a limit on the number of builds IIRC like 1 per day but i might misremember or it might have changed
[16:30] <michaelni> If someone knows of a way to test a model file without affecting the main ffmpeg at coverity, id like to know
[16:47] <cone-885> ffmpeg.git 03Nicolas George 07master:b491751409bc: doc: set documentencoding on toplevel texi files.
[16:47] <cone-885> ffmpeg.git 03Nicolas George 07master:c49c42a4a3a9: doc/faq: explain DAR/SAR preserving.
[16:47] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:65da611d3415: Merge remote-tracking branch 'cigaes/master'
[16:55] <ubitux> michaelni: do we specify configure options?
[16:56] <ubitux> we can't have a CONFIG_COVERITY to plug into libavutil/mem.c?
[16:56] <ubitux> how about the align?
[16:56] <michaelni> we could add a CONFIG_COVERITY
[16:58] <ubitux> as a toolchain or something maybe
[17:00] <michaelni> toolchain, possible
[17:00] <michaelni> not sure i understand what you mean by align ?
[17:02] <ubitux> michaelni: coverity doesn't need to know that the alloc is aligned?
[17:02] <ubitux> is this really going to help anything?
[17:03] <ubitux> i mean, it can follow that av_malloc is calling malloc
[17:03] <michaelni> ive no idea
[17:03] <ubitux> and it seems to find the leak
[17:03] <ubitux> +s
[17:12] <wm4> ubitux: coverity uses "models" for this
[17:13] <wm4> you probably can write a model for av_malloc/av_free
[17:13] <ubitux> well it already detects leaks
[17:13] <ubitux> so i'm wondering how this model is going to help anything
[17:14] <wm4> ah I see there's a patch
[17:16] <wm4> ubitux: one argument: coverity could get confused by the padding, and consider writing to it valid
[17:17] <wm4> at least with some of those crazy code paths
[17:40] <ubitux> wm4: well that's the point of the padding, reading and writing to it
[17:42] <wm4> ubitux: no
[17:42] <wm4> you're thinking of INPUT_PADDING something, but that memory is _really_ allocated by the padded amount
[17:43] <Daemon404> solution is to rewrite in rust with strict bounds checking on every array access
[17:45] <kierank> #shithnsays
[18:31] <cone-885> ffmpeg.git 03Michael Niedermayer 07fatal: ambiguous argument 'refs/tags/n2.2.13': unknown revision or path not in the working tree.
[18:31] <cone-885> Use '--' to separate paths from revisions
[18:31] <cone-885> refs/tags/n2.2.13:HEAD: Merge remote-tracking branch 'cigaes/master'
[20:36] <cone-885> ffmpeg.git 03Anton Khirnov 07master:5b1d9ceec715: pixfmt: add a pixel format for QSV hwaccel
[20:36] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:e3755119fae7: Merge commit '5b1d9ceec715846a58fe029bc3889ed6fa62436a'
[20:49] <cone-885> ffmpeg.git 03Anton Khirnov 07master:31d2039cb426: h264_parser: export video format and dimensions
[20:49] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:560eb7179a7c: Merge commit '31d2039cb42668ebcf08248bc48bbad44aa05f49'
[21:18] <aetasx> hey guys, is there a reason why the movie filter doesn't do the same amount of error recovery from stream errors?
[21:18] <aetasx> were they just never implemented or was there a reason behind it?
[21:19] <nevcairiel> same amount as compared to what?
[21:21] <aetasx> for instance, if I take a video and use it as input to ffprobe or something, it'll be parsed regardless of stream errors whereas using -f lavfi -i movie... can die off
[21:21] <aetasx> using the same set of options naturally
[21:24] <aetasx> the movie filter code seems much less forgiving about stream errors
[21:26] <wm4> the movie filter is a completely different implementation
[21:26] <wm4> lavfi nonsense
[21:27] <cone-885> ffmpeg.git 03Anton Khirnov 07master:4e08c821106f: lavc: add an Intel libmfx-based H.264 decoder
[21:27] <cone-885> ffmpeg.git 03Anton Khirnov 07master:11c22dfd495b: configure: add missing h264_qsv hwaccel deps
[21:27] <cone-885> ffmpeg.git 03Anton Khirnov 07master:0fbb27131889: qsv_api: fix building with libmfx disabled
[21:27] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:841e9f431cf4: Merge commit '4e08c821106fc1d6d358864abf0d8488b12f38c6'
[21:27] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:ac649bb775b9: Merge commit '11c22dfd495bbd3f38df639772aa4bb9a1be882d'
[21:27] <aetasx> it looks like for the most part that it expects the streams to be pretty well intact
[21:28] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:99bcfe62278f: Merge commit '0fbb271318899a0fb1fbcbb3db8292e909b91e23'
[21:28] <aetasx> or it just bails out
[21:31] <aleb> Hi. Why does remuxing an MTS to MKV fail (with "Can't write packet with unknown timestamp") but remuxing to MP4 works fine? Where do the timestamps come from in case of MTS->MP4? https://trac.ffmpeg.org/ticket/3339
[21:31] <nevcairiel> matroska may simpler be stricter. I know that mkv does not allow a missing timestamp at all. Maybe MP4 does, i'm not sure about that
[21:32] <compn> mkv and mp4 have different timestamp .... things
[21:32] <compn> whats it called
[21:32] <compn> timestamp base ?
[21:32] <aleb> (Also, remuxing the obtained MP4 to MKV then works fine)
[21:32] <compn> try -vf genpts or fixpts or whatever its called, aleb
[21:32] <compn> or theres -copytb 
[21:32] <compn> maybe
[21:33] <compn> too many options to remember :D
[21:33] <aleb> but what should that filter do? :)
[21:33] <aetasx> thought it was -fflags +genpts
[21:33] <aleb> looking to see what those do!
[21:44] <cone-885> ffmpeg.git 03Oleksij Rempel 07master:c56b9b1eb278: lavc: Add DSS SP decoder
[21:44] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:dfe5a9f022b3: Merge commit 'c56b9b1eb278c5ef89d3f0832a56dfe4732cb68b'
[21:53] <cone-885> ffmpeg.git 03Oleksij Rempel 07master:062cd5a975ff: lavf: Add DSS demuxer
[21:53] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:e821910a2151: Merge commit '062cd5a975ff7bd6fb91f9b4d1d9d102a7545499'
[22:01] <cone-885> ffmpeg.git 03Vittorio Giovara 07master:b727b0eeb73c: fate: Add tests for DSS
[22:02] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:29d2197d8805: Merge commit 'b727b0eeb73cd78533a7fd387ceccd505e48fa6c'
[22:23] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:b2bbe8298ba5: avformat/dss: Check sscanf() return value in dss_read_metadata_date()
[22:23] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:be0cb7e05385: avformat/dss: Use av_freep() to avoid leaving stale pointers in memory
[22:32] <aleb> compn, there is -vf setpts  but using filters does not work with -c copy. 
[22:32] <aleb> I tried "ffmpeg  -i x.MTS  -vf showinfo" and it looks messy to me: http://fpaste.org/187837/14243807/  
[22:33] <aleb> how much work/money would it take to fix https://trac.ffmpeg.org/ticket/3339 - maybe I can run a fundraiser for it
[23:02] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:4f1c1b26f234: avcodec/dss_sp: Protect DSS_SP_FORMULA() with a set of outside brackets
[23:02] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:31fb029f2d9e: avcodec/dss_sp: de-obfuscate index inversion
[23:02] <cone-885> ffmpeg.git 03Michael Niedermayer 07master:b11edbd289e4: avcodec/dss_sp: Avoid a slow division and modulo in the pitch_lag calculation
[00:00] --- Fri Feb 20 2015

More information about the Ffmpeg-devel-irc mailing list