[Ffmpeg-devel-irc] ffmpeg-devel.log.20130709

burek burek021 at gmail.com
Wed Jul 10 02:07:48 CEST 2013


[01:04] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/0.10:f75964ad1fe7: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:04] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/0.11:c9f34c82604a: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:04] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/0.7:82c6b76d701a: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:04] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/0.8:f8c4dbe45ee2: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:04] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/0.9:d0a225d6c1a7: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:05] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/1.0:ec268c24a77e: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:05] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/1.1:18900381e2d3: mp3dec: detect CBR and use CBR axiom to seek
[01:05] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/1.1:0f84286677b5: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:05] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/1.2:2e75c455935d: mp3dec: detect CBR and use CBR axiom to seek
[01:05] <cone-757> ffmpeg.git 03Michael Niedermayer 07release/1.2:6cb33e076323: mpeg12dec: avoid reinitialization on PS changes when possible.
[01:56] <cone-757> ffmpeg.git 03Michael Niedermayer 07master:d7fbe926d312: ffmpeg: move creation_time for ffserver setting up
[01:56] <cone-757> ffmpeg.git 03Michael Niedermayer 07master:d1e8d4b055bd: ffmpeg: Allow overriding ffservers input
[02:32] <Foofie> just noticed that ffmpeg -h full does an infinite loop... is this a known issue?
[02:33] <durandal_1707> Foofie: doesn't here
[02:33] <durandal_1707> Foofie: what version?
[02:33] <Foofie> git head
[02:33] <Foofie> using latest git head of libvpx
[02:34] <Foofie> it appears to continuously loop on libvpx related options
[02:34] <durandal_1707> i don't have libvpx compiled
[02:35] <durandal_1707> but something changed recently with libvpx stuff
[02:35] <Foofie> it's very strange :/
[02:35] <durandal_1707> vp9 support have been added iirc
[02:36] <Foofie> yes
[02:36] <durandal_1707> Foofie: is this with MSVC?
[02:36] <Foofie> git clone http://git.chromium.org/webm/libvpx.git ... it's a pretty easy ./configure && make && make install
[02:36] <Foofie> nope, on mac osx
[02:41] <Foofie> I have found that some options end up freezing execution (so when running -v debug), it will just wait forever when it's parsing an option, such as -g (and some others).  My guess is that it's the same issue... some data pointing back to itself in a circular manner... ? (have not looked at the code though)
[02:42] <Foofie> like I can make it hang forever if I use the -g option, but when I get rid of it, it does go on through.
[02:43] <Foofie> afk for a bit
[03:06] <cone-757> ffmpeg.git 03Michael Niedermayer 07master:a20f049c2ab5: avcodec/libvpxenc: make class names unique
[03:07] <michaelni> Foofie, should be fixed (i hope)
[03:16] <cone-757> ffmpeg.git 03Matthew Heaney 07master:509642b4bd84: WebM muxer writes WebVTT subtitle track
[05:32] <cone-757> ffmpeg.git 03Michael Niedermayer 07master:00b1401df656: avcodec/h264_ps: try to support Ambarella AVC
[09:43] <wm4> wow this webvtt muxing stuff is ugly
[09:47] <superware> michaelni: hi. I wanted to thank you for fixing all issues with that hw stream!
[09:56] <michaelni> wm4, anything that could be done better ? if so please reply in the thread
[09:56] <wm4> it's probably too late and the format requires it
[09:57] <michaelni> also theres a demuxer patch, if you want to review it ...
[09:57] <michaelni> ubitux is busy AFAIK and wont review it quickly i suspect ...
[09:58] <michaelni> BuxiNess, there are some jpeg2000 patches on ffmpeg-devel if you have time to review ...
[10:07] <cone-395> ffmpeg.git 03Martin Storsjö 07master:a87a0acf9b5d: movenc: Make sure the RTP hint tracks have nondecreasing DTS
[10:07] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:848295c2eeac: Merge commit 'a87a0acf9b5d27aad032e61eef4973e62a4a6830'
[10:14] <cone-395> ffmpeg.git 03Martin Storsjö 07master:0677ddebc6e9: movenc: Rename the timestamp parameter to dts in write_hint_packets
[10:14] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:e914a21ada12: Merge remote-tracking branch 'qatar/master'
[10:36] <superware> is there anything I can try to reduce latency other than AVFMT_FLAG_NOBUFFER?
[10:36] <JEEB> superdump, completely depends on what you're trying to do
[10:37] <JEEB> for example the encoder you might be using and its settings are relatively relevant :P
[10:39] <superware> JEEB: you are right. I can't touch the encoder, only the decoder. The stream is h.264/mpeg ts/udp..
[10:40] <b_jonas> yay, the ticket claims you fixed my bug! thanks for all the developers fixing it.
[10:41] <JEEB> so you're just decoding a random stream that you're getting? well, can't help much there except for maybe setting the container/video|audio formats and making sure ffmpeg doesn't try to probe it first?
[10:41] <JEEB> nothing else I can say :)
[10:41] <superware> JEEB: probing time isn't related to later video-latency, right?
[10:42] <BuxiNess> michaelni, yes I 'll get a look
[10:42] <BuxiNess> I think I found the problem of perf according the number of threads. 
[10:42] <michaelni> BuxiNess, thanks!
[10:42] <BuxiNess> I'll send a patch
[10:42] <michaelni> ok
[10:44] <JEEB> superdump, if you're just decoding the decoding should be going the usual speed, and it's mostly just latency in starting that decode?
[10:45] <superdump> JEEB: superw<tab>
[10:45] <JEEB> yes, mis-tabbed
[10:45] <nevcairiel> or get a irc client that prefers people that have talked last :p
[10:46] <superdump> although i just ruined such a feature
[10:46] <superdump> su<tab><tab> should work too
[10:46] <superdump> if your irc client cycles nicks
[10:46] <superware> JEEB, I don't really care about probing latency, only video latency
[10:47] <JEEB> yes, I just didn't cycle because I'm tired :P
[10:47] <superware> JEEB, with AVFMT_FLAG_NOBUFFER I think the latency is about 200ms
[10:48] <JEEB> I have no idea :P
[10:56] <wm4> so do you have to call avcodec_decode_audio4() multiple times per packet or not?
[10:56] <wm4> doc/examples/demuxing.c and ffplay.c disagree
[10:57] <wm4> ffmpeg.c seems to do the same as ffplay.c
[10:58] <wm4> basically, the question is whether this is still needed: http://git.videolan.org/?p=ffmpeg.git;a=blob;f=ffplay.c;hb=HEAD#l2167
[10:59] <wm4> (not counting the need to finish decoding with null packets)
[10:59] <nevcairiel> i believe you have to do that, it returns the number of consumed bytes and you have to handle that
[10:59] <wm4> so doc/examples/demuxing.c is wrong? (or I read it wrong)
[10:59] <nevcairiel> those examples are generally wrong or outdated :p
[11:00] <wm4> ok my guess is nobody really knows
[11:00] <nevcairiel> i'm quite sure i saw codecs where its mandatory
[11:00] <nevcairiel> als i think
[11:01] <nevcairiel> at least when playing a raw .als file where the frames arent pre-sliced
[11:09] <wm4> hm from what I can see ffmpeg seems to support als in mp4 only
[11:09] <nevcairiel> i'm certain i had a .als file once, but what do i know if it was secretly mp4 inside
[11:10] <wm4> any other formats that could be affected?
[11:11] <nevcairiel> realmedia codecs possibly
[11:11] <nevcairiel> cook or atrac
[11:12] <nevcairiel> i could quickly test a bunch of files if i were actually at home, but sadly that is not the case :d
[11:17] <wm4> nevcairiel: tried some random rm samples, no "luck"
[11:18] <wm4> randomly looking at some decoders in libavcodec I see how this could happen, though
[11:29] <wm4> durandal_1707: since you seem to know about audio decoding, any clue whether this is still needed? http://git.videolan.org/?p=ffmpeg.git;a=blob;f=ffplay.c;hb=HEAD#l2167
[11:31] <durandal_1707> its for "retardet" codecs/containers
[11:31] <durandal_1707> iirc used by shorten
[11:31] <wm4> any samples?
[11:32] <durandal_1707> and maybe adpcm in wav with block_align
[11:32] <wm4> it's notable that doc/examples/demuxing.c seems to assume that a packet has to be decoded only once
[11:43] <cone-395> ffmpeg.git 03Paul B Mahol 07master:3c5071db88b2: lavfi/psnr: rename 's' parameter to match documentation
[11:52] <durandal_1707> michaelni: there are some bugs reported that you never fixed and now are in libav?
[11:52] <durandal_1707> very "well" played
[11:55] Action: michaelni doesnt understand
[11:55] <durandal_1707> the security ones reported by google
[11:57] <michaelni> we fix(ed) all security bugs
[12:00] <durandal_1707> you must forgot some
[12:00] <durandal_1707> or they are not sending samples too
[12:04] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:0dd99628ea15: ff_lpc_calc_coefs: assert that the type is levinson or cholesky
[12:04] <michaelni> durandal_1707, what you talk about ?
[12:06] <durandal_1707> see http://lists.libav.org/pipermail/libav-devel/2013-July/048477.html
[12:17] <wm4> durandal_1707: btw. if you know some samples that requires decoding a packet multiple times, that would be nice, and might even lead to either the demuxing.c example to be fixed, or maybe simplifying the API itself
[12:18] <durandal_1707> wm4: you couldn't find shorted sample?
[12:19] <durandal_1707> *shorten
[12:19] <wm4> searching for it is kind of hard
[12:19] <durandal_1707> it must be in fate
[12:20] <durandal_1707> lossless-audio/luckynight-partial.shn
[12:20] <durandal_1707> and if samples.ffmpeg.org/lossless something...
[12:22] <wm4> hmmm
[12:23] <wm4> indeed, this sample is affected
[12:23] <wm4> thanks
[13:25] <cone-395> ffmpeg.git 03Carl Eugen Hoyos 07master:9a1ec1f2f223: Force correct extradata for AVID AVCI50 in mov.
[13:25] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:694550f18e02: Merge remote-tracking branch 'cehoyos/master'
[14:10] <durandal_1707> michaelni: apad is not documented at all
[14:57] <xlinkz0> god damn exit functions in ffmpeg tool
[14:58] <xlinkz0> is it written in such a way that if i #define exit(x) it doesn't break?
[14:59] <nevcairiel> read the code and find out? :)
[14:59] <nevcairiel> what you're trying to do is not supported in any way, so don't expect it to be :D
[15:00] <xlinkz0> well i got it compiled into a library
[15:00] <xlinkz0> and it works.. once
[15:00] <xlinkz0> because it then exit()s
[15:01] <xlinkz0> 36 exits in ffmpeg.c 
[15:01] <xlinkz0> fml
[15:56] <durandal_1707> michaelni: it looks like apad pads silence forever
[15:57] <michaelni> durandal_1707, its intended to be used with -shortest so a audio stream can be padded to the langth of video
[16:00] <durandal_1707> michaelni: shortest is ffmpeg only thing
[16:15] <michaelni> durandal11707, yes, the filter was writen for the purpose of extending the audio to the same length as video with ffmpeg
[16:16] <durandal11707> it should be extended to be less useless
[16:26] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:7d9cb126893d: avcodec/wmaprodec: Check block_align
[16:26] <cone-395> ffmpeg.git 03Luca Barbato 07master:87fb18c3e4fe: imc: Catch a division by zero
[16:44] <xlinkz0> ffmpeg tool only cleans up after itself if no errors occur?
[16:44] <xlinkz0> or does exit() by any chance trigger a callback?
[16:50] <durandal_1707> michaelni: perhaps it still use stuff when calling .uninit
[16:50] <durandal_1707> if i remove freep from it segv does not happen
[16:50] <durandal_1707> i know uninit works as it does not segv on normal exit
[16:52] <michaelni> xlinkz0, update your tree exit has been mostly replaced by exit_program() and yes ffmpeg  used atexit()
[16:52] <xlinkz0> was this in the last week?
[16:54] <durandal_1707> michaelni: so you get imc report or not?
[16:54] <durandal_1707> s/get/got
[16:57] <xlinkz0> michaelni: thank you so much, it is like god looked upon my feeble attempts and granted me with a speckle of light
[16:59] <xlinkz0> so i can just change exit(ret) from exit_program to pthread_exit 
[16:59] <xlinkz0> this.. can't be that easy
[17:21] <cone-395> ffmpeg.git 03Paul B Mahol 07master:c5a1b18f183d: lavfi/drawtext: add support for printing frame metadata
[17:23] <durandal_1707> now one can put psnr of videos with drawtext
[19:05] <Compn> we should combine our network code with vlc
[19:05] <Compn> so both projects dont have to reinvent the wheel every stream
[19:05] <Compn> :P
[19:10] <durandal_1707> Compn: everybody reinvents wheel
[19:16] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:d5de1231954e: avfilter/graphparse: Fix build with --enable-incompatible-fork-abi
[19:16] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:f9d7431272f4: avformat/AVFormatContext: Move fields down to match the fork
[19:30] <michaelni> buxiness, do you see a problem with the jpeg2000dec patches ?
[19:32] <durandal_1707> michaelni: what about questions I asked?
[19:33] <michaelni> durandal_1707, about the division by 0 ?
[19:33] <durandal_1707> about security issues that are still not fixed but reported
[19:33] <michaelni> its possible it didnt happen for us because some other check stoped it before
[19:33] <michaelni> iam not aware of any reported but not fixed issues
[19:39] <cone-395> ffmpeg.git 03James Almer 07master:1e5bbbfcf303: lavf/flacdec: Share the function to parse cover art blocks
[19:39] <cone-395> ffmpeg.git 03James Almer 07master:193782216fc8: vorbiscomment: Add support for embedded cover art
[19:49] <cone-395> ffmpeg.git 03Piotr Bandurski 07master:aa74810fc6b9: bitstream: fix a crash in case of oom
[19:54] <durandal_1707> saste: its just little confusing commit message
[19:55] <saste> durandal_1707, i'm basically replacing the original text, and merging two sections in one
[19:55] <saste> i have no problem to change the commit log
[19:56] <saste> but the content as is is greatly confusing, the part about "thou shall publish the code" is a false boast
[19:56] <durandal_1707> if they change something in lib itself, they MUST
[20:02] <saste> durandal_1707, redistribute != punblish
[20:02] <saste> you can modify and redistribute your copy only to the library user
[20:02] <saste> you are not required to *publish* the code
[20:03] <nevcairiel> if you make a build public, you need to make the changes in it public
[20:03] <nevcairiel> what you do on your own PC is only your own business
[20:03] <durandal_1707> no!!!!!
[20:03] <saste> but you can have a single user
[20:04] <durandal_1707> have GPLv3 fixed this?
[20:04] <saste> for example I'm not required to publish all my experimental patches
[20:04] <durandal_1707> saste: you must!
[20:04] <iive> durandal_1707: not in gpl.
[20:04] <saste> durandal_1707, that's absurd
[20:04] <durandal_1707> all patches must be published
[20:05] <saste> durandal_1707, no, that's not what LGPL states
[20:05] <saste> also it is practically absurd, since you don't define what is "to publish"
[20:05] <nevcairiel> you must publish source if you publish a binary
[20:05] <iive> e.g. google doesn't have to publish its own in house code, and they use a lot of linux
[20:05] <nevcairiel> this is basically all
[20:05] <iive> nevcairiel: exactly.
[20:05] <saste> you can encourage third parties who modifes the code to publish code (iff it is worthy to share that)
[20:06] <durandal_1707> ah, so explain it in doc if they make stuff public they must comply and publish code too
[20:06] <saste> comply??
[20:06] <saste> you can't enforce that
[20:06] <nevcairiel> well technically you can
[20:06] <nevcairiel> with lawyers
[20:06] <saste> you even don't want to do that, also what is to publish?
[20:07] <nevcairiel> its probably sufficient to supply the source on request, but IANAL
[20:07] <saste> that's not LGPL, but funky LGPL
[20:08] <saste> nevcairiel, so I add a change to print a count, I'm modifying the source code, so I must publish it?
[20:08] <nevcairiel> it comes down to one major rule, if you distribute a binary, you must also make the source used to build this binary available
[20:08] <saste> that's nonsense
[20:08] <nevcairiel> thats how the (L)GPL works
[20:09] <saste> i can modify the code and distribute it to my (single) customer
[20:09] <saste> i'm required to provide him/her with the modified sourcecode, so he can change it
[20:09] <saste> but i'm not required to publish the code to the known universe
[20:10] <saste> nevcairiel, yes
[20:10] <durandal_1707> saste: i'm going to hunt your patches
[20:12] <nevcairiel> of course you're not required to publish it to the world, but anyone that has access to the binary can also ask for the source, more or less
[20:12] <iive> and then he have the right to publish the source, if he wants.
[20:13] <saste> nevcairiel, iive: yes
[20:13] <michaelni> saste, about "thou shall publish the code", its not about legel req but what is the right thing to do, aka you improve it you should give back so all benefit
[20:13] <saste> if i distribute a binary, i have to provide its source code
[20:13] <saste> it's not like, i do a change, i must publish it => nonsense
[20:14] <saste> michaelni, but don't confuse the license with ethics
[20:14] <iive> I think that MPL is like that (Mozilla)
[20:14] <saste> users are *encouraged* but not *enforced* to publish the modified source code (only if it makes sense)
[20:15] <nevcairiel> you can't really enforce that and still be open-source, its also a bit about freedom of the developers
[20:39] <johnsmith> Hi, I was wondering if anyone here has compiled/installed the latest version of FFMPEG onto CENTOS 5 (Work related reasons)
[21:17] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:325411a687c4: snowenc: check minimum resolution
[22:06] <wm4> how do I build the ffmpeg examples? FAQ says they are installed by default, but it installs the .c files
[22:08] <saste> wm4, less README, make
[22:08] <llogan> saste: ping on the PPC
[22:09] <saste> llogan, ? i know nothing about PPC
[22:10] <llogan> i guess you did time out then. i asked in IRC if you still wanted a PPC. i have an older, but still usable powerbook if you want it (if shipping isn't prohibitive).
[23:16] <wm4> lol
[23:17] <wm4> demuxing.c doesn't handle planar audio formats
[23:17] <durandal_1707> example?
[23:17] <wm4> how do you expect anyone to use this library if even the examples are highly buggy
[23:30] <saste> wm4, demuxing was added before planar was introduced
[23:30] <saste> so it was not even required back then
[23:31] <wm4> does Libav not have examples?
[23:31] <j-b> planar audio transition was the worse change ever done in libavcodec
[23:31] <j-b> in terms of transition, not technical reason
[23:32] <wm4> let's break all applications for fun!
[23:32] <wm4> but the cause of the suck is deeper: decoders output whatever they want by design
[23:32] <saste> wm4, the soluton is to use libswresample
[23:32] <saste> in case of demuxing
[23:32] <wm4> instead of returning the data in a format the user asks for
[23:32] <wm4> i.e. too low level API
[23:32] <saste> yet i feel the libswr API is still too much complex
[23:33] <saste> wm4, i don't know we want that
[23:33] <saste> at the same way decoders don't require the use of libswscale
[23:35] <saste> eh two weeks is not "old" in the standard of ffmpeg
[23:51] <durandal_1707> saste: using , as separator for delays need \\ escaping
[23:51] <durandal_1707> i could use ' ' or '|'
[23:51] <saste> durandal_1707, yes
[23:51] <durandal_1707> so what separator?
[23:52] <saste> | seems fine
[23:52] <saste> not sure for the space, since it could imply other escaping issues
[23:58] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:e9be4675b88a: doc/ffmpeg: document -override_ffserver
[23:58] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:776155ca29e7: mjpeg_decode_scan: reset restart state
[23:58] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:1c032e94d0a5: jpegdec: give handle_rstn() a return code that indicates if restart did happen
[23:58] <cone-395> ffmpeg.git 03Michael Niedermayer 07master:233177a4ec68: jpegdec: clear EOBRUN on restarts
[23:59] <saste> cehoyos: "fixed by you" is pretty useless
[00:00] --- Wed Jul 10 2013


More information about the Ffmpeg-devel-irc mailing list