[Ffmpeg-devel-irc] ffmpeg-devel.log.20120709

Tue Jul 10 02:05:04 CEST 2012

[01:02] <durandal_1707> michaelni: i cant find out what is wrong with rle sgi image, ffplay can't display it if DR1 is enable, (raw ones works fine)
[01:16] <ktdreyer> michaelni: thank you for looking at this. I'll try to get a hold of some xbmc devs
[01:18] <michaelni> durandal_1707, the out_end check should differ in behavior if linesize>width
[01:19] <michaelni> not sure if thats the issue
[01:19] <michaelni> iam just looking at the code not testing ..
[01:23] <durandal_1707> michaelni: actually out_end in raw case is changed, so it is pointless for function to have it as argument
[01:24] <michaelni> do you have a sample failing sgi file ?
[01:24] <durandal_1707> michaelni: any one created with our encoder, use -coder rle
[01:25] <ktdreyer> michaelni: you're absolutely right about this being fixed in xbmc's master. Guess I should've checked there :) Thank you.
[01:30] <michaelni> durandal_1707, its off by 1 error
[01:30] <michaelni> ill commit a fix
[01:30] <durandal_1707> hah
[01:32] <burek> can I suggest a bug in docs here, since it's too small
[01:32] <burek> so I don't create ticket instead
[01:33] <Compn> just commit it
[01:33] <durandal_1707> burek: what? where?
[01:33] <burek> in ffmpeg html doc -itsoffset doesn't explain the proper usage, i.e. should it be used before/after input(s) and how does it affect any subsequent inputs (if it does)
[01:34] <burek> Compn, I would, but I suck at git usage .. :/
[01:34] <burek> if it was a wiki, I would correct it already :))
[01:34] <michaelni> burek, commit it to cvs then ill merge with git-cvs
[01:35] <burek> lol :D
[01:36] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * r7bc155163e 10ffmpeg/libavcodec/sgidec.c: 
[01:36] <CIA-41> ffmpeg: sgidec: fix off by 1 error that tested a sample too far for oob.
[01:36] <CIA-41> ffmpeg: This led to decoding failure when linesize was too small.
[01:36] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[01:36] <burek> ok never mind :)
[01:36] <burek> but, bear in mind that these are (one of) the reasons the html docs are incomplete/innacurate/whatever
[01:38] <michaelni> someone has to write a wiki frontend to git
[01:38] <michaelni> but seriously, if you send a patch ia, happy to apply it
[01:39] <michaelni> but some vague reference that requires research is likely not going to lead to a commit from me as i dunno what where ...
[01:39] <burek> we were discussing that few days ago, it's not a problem for me, personally, to learn how to use git, it's the general problem because a lot of other "ordinary" users (non-devels) do not know it either
[01:39] <burek> and they might be able/willing to improve docs
[01:40] <burek> now, there is trac wiki, which is (I believe) integrated with git
[01:40] <michaelni> no integration between the 2 ATM
[01:41] <michaelni> isnt there some web frontend to git where you can edit wiki style and that produces a git patch ?
[01:41] <burek> well somebody (ubitux, saste, or somebody else) told me in that discussion that it is integrated
[01:41] <burek> I'll check that
[01:44] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r104ea00108 10ffmpeg/libavcodec/dpx.c: 
[01:44] <CIA-41> ffmpeg: dpx: add CODEC_CAP_DR1
[01:44] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[01:44] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r5d23043c12 10ffmpeg/libavcodec/exr.c: 
[01:44] <CIA-41> ffmpeg: exr: add CODEC_CAP_DR1
[01:44] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[01:44] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r0f65693529 10ffmpeg/libavcodec/sgidec.c: 
[01:44] <CIA-41> ffmpeg: sgidec: remove redundant argument from read_uncompressed_sgi()
[01:44] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[01:44] <CIA-41> ffmpeg: 03Paul B Mahol 07master * re2a6eb8da4 10ffmpeg/libavcodec/sgidec.c: 
[01:44] <CIA-41> ffmpeg: sgidec: add CODEC_CAP_DR1
[01:44] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[01:44] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r60351ec13c 10ffmpeg/libavcodec/g723_1.c: 
[01:44] <CIA-41> ffmpeg: g723_1: add CODEC_CAP_DR1
[01:44] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[01:49] <burek> michaelni, this perhaps? http://jkingdon2000.blogspot.com/2011/01/tracking-mediawiki-wiki-in-git.html
[01:53] <burek> https://github.com/github/gollum
[02:59] <durandal_1707> how do you read float numbers?
[03:00] <burek> bit by bit? :)
[03:03] <durandal_1707> sample aspect ration is given as float number
[03:03] <durandal_1707> now i need to convert that to AVRational
[03:05] <burek> avctx->sample_aspect_ratio = (AVRational) {160, 99};
[03:05] <durandal_1707> you get 32bit float
[03:05] <durandal_1707> actually it is 64bit
[03:14] <burek> does this mean that the latest change in ffmpeg kinda broke the x264 way of using lavf in its encoder: http://pastebin.ca/2168922
[03:15] <burek> it's a config.log from ./configure for x264 with --enable-lavf
[03:16] <durandal_1707> burek: what x264 version?
[03:16] <burek> from yesterday
[03:17] <burek> same with current latest git
[03:17] <burek> (check #ffmpeg)
[03:19] <durandal_1707> burek: x264 compiles just fine here (lavf enabled)
[03:20] <burek> :)
[03:20] <wishie> durandal_1707: what options did you use to get x264 to compile with lavf support?
[03:21] <durandal_1707> nothing, i just used defaults
[03:22] <durandal_1707> and i really cant see from where is that file (reported in failure)
[03:22] <burek> durandal_1707, did you check after ./configure ... that "lavf... yes" is checked
[03:22] <durandal_1707> yes
[03:36] <burek> durandal_1707, everything is ok :) wishie was using svn :)))
[03:36] <wishie> lol
[03:36] <wishie> i was following a how to which didn't seem very old
[03:36] <wishie> yet it was telling me to get ffmpeg source from mplayerhq.hu using svn
[03:37] <durandal_1707> well svn server is still alive for some reasons....
[04:00] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r16a433d68b 10ffmpeg/libavcodec/tiffenc.c: 
[04:00] <CIA-41> ffmpeg: tiffenc: PIX_FMT_YUV440P support
[04:00] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[04:10] <durandal_1707> hmm, exr is going to have another - 2.0 version
[04:19] <michaelni> av_d2q() does float or double->rational
[04:20] <durandal_1707> michaelni: too late, i figured it already - patch on ml....
[04:43] <durandal_1707> michaelni: hmm i'm not sure for this division, it may be division by zero, i this division is not required at all
[04:44] <durandal_1707> s/this/think this/
[04:51] <CIA-41> ffmpeg: 03Paul B Mahol 07master * r07df939aa3 10ffmpeg/libavcodec/exr.c: 
[04:51] <CIA-41> ffmpeg: exr: support for reading pixelAspectRatio
[04:51] <CIA-41> ffmpeg: Signed-off-by: Paul B Mahol <onemda at gmail.com>
[08:47] <ubitux> saste: i worked a bit on showspectrum the other day, and got something weird with the buffer ref
[08:48] <dan_> hi, anyone familiar with ac3 / PTS?
[08:48] <saste> ubitux: what's the matter?
[08:49] <ubitux> saste: i get a buffer ref with write permission at the beginning, and since i'm using it for each incrementation, when pushing it to the next filter i send it a RO reference to this buffer
[08:49] <ubitux> this works, but it seems to leak
[08:49] <ubitux> i thought i had to unref this new RO reference after the start_frame/draw_slice/end_frame, but doing this crashes so i guess that's not the way
[08:50] <ubitux> see the leaking version: https://github.com/ubitux/FFmpeg/commit/df3f861a4f288de2b9ebd8374c0b9036cd151e21#L2R185
[08:52] <ubitux> i guess lavfi might not allow that kind of stuff, so maybe i'll have to create a new buffer ref and copy the local one into it
[08:52] <saste> ff_start_frame(outlink, avfilter_ref_buffer(showspectrum->outpicref, AV_PERM_READ));
[08:52] <ubitux> (or use & update a draw context or something each time)
[08:53] <saste> this leaks, since the receiving filter won't be able to change it, and it will have to copy it
[08:53] <saste> but the generating filter is not saving the reference itself, so cannot free it
[08:54] <ubitux> saste: right, but AVFilterBufferRef *ro_ref = avfilter_ref_buffer(showspectrum->outpicref, AV_PERM_READ); and then ff_start_frame(outlink, ro_ref); ff_draw_slice(); ff_end_frame(); avfilter_unref_buf(ro_ref); leads to a crash
[08:54] <ubitux> (double free iirc, or something like this)
[08:56] <saste> funny, seems contradictory...
[08:56] <ubitux> yeah
[08:58] <saste> when is the frame duplicated, later in the filterchain?
[08:58] <saste> that is, how are you using it?
[08:58] <ubitux> the filtergraph is pretty simple ATM, i just use showspectrum in standalone
[08:59] <saste> if there are no other filters which *writes* in the filter, then maybe they will free the reference after all
[08:59] <ubitux> i don't even split to listen at the same time since i don't set the PTS yet
[08:59] <saste> looking to traces should help
[09:00] <ubitux> right ok i'll look more into this when i have time, it was just in case you had an idea
[09:16] <dan_> :/
[13:47] <CIA-41> ffmpeg: 03Carl Eugen Hoyos 07master * r36b2bd9629 10ffmpeg/libavcodec/utils.c: Show hyphen in codec tag.
[14:34] <ubitux> Tjoppen: you like that NIH attitude? ;)
[14:34] <ubitux> this is ridiculous...
[14:38] <Tjoppen> wild mru appears
[14:39] <Tjoppen> I shouldn't have said anything..
[14:40] <ubitux> btw, the macro proposed by libav won't work in ffmpeg
[14:40] <ubitux> i mean, it won't do what you expect
[14:40] <Tjoppen> av_malloc(0) feels.. wrong
[14:40] <ubitux> it's ok in libav
[14:40] <ubitux> they return NULL
[14:40] <ubitux> but we don't
[14:41] <Tjoppen> you can rearrange it so that it works, but: macro. better to have an inline then
[14:41] <Tjoppen> of course, this is pointless bikeshedding
[14:41] <Tjoppen> I needn't participate at all. time to reveal a couple of cards
[14:48] <ubitux> it's broken in libav actually
[14:57] <ubitux> michaelni: where is it more interesting to return an allocated buffer with av_malloc(0) instead of NULL?
[14:58] <michaelni> ubitux, IIRC (maybe thats notz the only reason) its the error checking, that is
[14:58] <michaelni> if(!ptr) error after malloc()
[14:58] <michaelni> strictly one would need if(size && !ptr) everywhere if malloc(0) returned NULL
[14:59] <ubitux> oh i see, makes sense.
[15:00] <ubitux> i'm not sure it's that much a problem in most of the case though
[15:01] <merbanan> isn't malloc(0) undefined ?
[15:01] <ubitux> yes
[15:02] <merbanan> or NULL or pointer to zero sized buffer
[15:03] <merbanan> both are allowed
[15:04] <michaelni> merbanan, both are allowed in libc malloc()
[15:04] <michaelni> allowing just one in av_malloc() simplifies code
[15:04] <merbanan> yeah, makes sense
[15:31] <CIA-41> ffmpeg: 03Al at dneg.com 07master * rc88a832735 10ffmpeg/libavformat/movenc.c: (log message trimmed)
[15:31] <CIA-41> ffmpeg: movenc: Addtion of \251cmt field to udta
[15:31] <CIA-41> ffmpeg: Allows support of comment in udta for libquicktime, also currently \251des is used
[15:31] <CIA-41> ffmpeg:  which is ignored by mov.c.
[15:31] <CIA-41> ffmpeg: \251cmt IS currently already picked up by mov.c
[15:31] <CIA-41> ffmpeg: References
[15:31] <CIA-41> ffmpeg: libquicktime
[15:41] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * rcdb3f2f5f0 10ffmpeg/libavcodec/mpeg12enc.c: 
[15:41] <CIA-41> ffmpeg: mpeg12enc: switch to av_assert
[15:41] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[15:41] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * rbf5386385d 10ffmpeg/libavcodec/mpeg12.c: 
[15:41] <CIA-41> ffmpeg: mpeg12: switch to av_assert
[15:41] <CIA-41> ffmpeg: Signed-off-by: Michael Niedermayer <michaelni at gmx.at>
[15:50] <burek> guys, how do you feel about setting up a wiki for ffmpeg documentation, which would be backed up by git, so you can make offline changes too: http://jkingdon2000.blogspot.com/2011/01/tracking-mediawiki-wiki-in-git.html
[15:53] <michaelni> burek, in principle this sounds interresting (without reading  the link), id also would be in favor of moving from trak wiki to media wiki
[15:54] <ubitux> i personally like to have the documentation in the repository
[15:54] <burek> michaelni, I was thinking more of texinfo docs, which generates ffmpeg.html
[15:54] <michaelni> one big issue for anything is admin/maintaince time
[15:54] <burek> to join all the docs in one wiki
[15:55] Action: ubitux likes the split between user & dev doc
[15:55] <ubitux> although, a great part of the documentation could be integrated to the source code
[15:55] <ubitux> and generated
[15:55] <ubitux> especially the option descriptions
[15:57] <burek> the main issue I was aiming to solve with this is the case when users find some errors/bugs/missing_info in dev docs (like ffmpeg.html) they have to spend time to learn how to create a patch, send it to devel list and then devel(s) have to also spend time to review it and merge it into the docs
[15:57] <burek> that's like a double work
[15:58] <michaelni> burek, moving from git-texi to anything wiki means merges from libav will be harder due to merging between different system 
[15:59] <michaelni> if theres a volunteer to merge future changes from libav into a wiki based system,
[16:00] <michaelni> then ive no objections if everyone is fine with this, though ubitux didnt sound like he is 100% fine with it ...
[16:00] <ubitux> burek: i don't like the idea of having them modifying the doc without review
[16:00] <burek> https://mercure.iro.umontreal.ca/pipermail/gambit-list/2008-December/002874.html
[16:00] <burek> @michaelni 
[16:00] <ubitux> michaelni: saste also share my opinion on this
[16:00] <ubitux> it needs to be discussed
[16:00] <ubitux> IMO
[16:00] <michaelni> yes 
[16:00] <michaelni> or
[16:01] <burek> ok
[16:01] <michaelni> burek, fork the docs and if the wiki idea is successfull and ends better then there probably will be less objection
[16:02] <michaelni> and in that case we could drop the git-texi ...
[16:02] <ubitux> burek: then ideally a system would make allow them to create a patch on the fly
[16:02] <ubitux> but then they will need to edit it after review
[16:03] <ubitux> i can't think of anything simple for this except git
[16:03] <ubitux> and not having reviews for technical docs is wrong IMO
[16:03] Action: michaelni goes afk (will read death threads and love letters later) :)
[16:03] <ubitux> heh :)
[16:03] <burek> :)
[16:04] <burek> well, ok, I just thought it might be more simple, but I see your point
[16:08] <ubitux> burek: other simple alternative: for any trivial edit, ask here or on #ffmpeg hilighting me, and i'll fix it, giving some credits to the one who noticed it
[16:09] <burek> ok :)
[16:09] <ubitux> for more "complex" patches, passing through the review system is imo the right way
[16:09] <ubitux> they can also open an issue as task/request, "please document [...]"
[16:10] <burek> it's just my thought that anyone, who never used git, but spots some missing info in docs or similar, won't spend time to learn how to create patches and stuff, but will rather ignore the thing, which essentially is our loss
[16:10] <ubitux> or you (or anything other dev) could do that for them if the registering is a problem
[16:10] <ubitux> burek: saying "there is a typo in the doc" in not such a pain
[16:11] <burek> I did that kind of thing on the forum
[16:11] <burek> passing tickets from issues encountered
[16:11] <burek> but, obviously if we make things easier to be reported/fixed the more contributions we will have
[16:12] <burek> for example in ffmpeg.html I've spotted there is no usage example for the -f ffmetadata
[16:12] <ubitux> the thing is, we can't trust users for technical documentation except for typo
[16:12] <burek> and was willing to quickly add a simple use case to read/write it
[16:12] <burek> but, I didn't since I don't know how to properly create patch :)
[16:13] <ubitux> (you will have users documenting non consistencly, documenting libav because they thought it was ffmpeg, saying various wrong things, etc)
[16:13] <burek> wiki has a good thing, commits, which you can simply undo, if they are wrong/bad
[16:13] <ubitux> this is what review is for
[16:13] <ubitux> the other way is to have multiple developers acting as moderator on the wiki
[16:13] <burek> so, imho, it's easier to undo something, with the click of the mouse, than to review things and merge it
[16:13] <ubitux> and monitoring it closely
[16:13] <ubitux> but we don't have that manpower
[16:13] <burek> yes, I understand
[16:14] <ubitux> burek: i understand the issue, i'm not saying it's not a problem
[16:14] <ubitux> but i believe the current situation is better
[16:37] <cbsrobot> ubitux, burek : why not a mwdiawiki with the ffmpeg doc everyone can edit
[16:38] <cbsrobot> but it gets rebuild from ffmpeg every day, so the doc stays current
[16:38] <ubitux> arh not again
[16:38] <cbsrobot> the patches will be sent to the ml f.ex.
[16:38] <ubitux> cbsrobot: i don't care about mediawiki or anything, what i say is that we need to keep the doc split between dev & user
[16:38] <burek> well mediawiki actually creates some kind of diff files daily
[16:39] <burek> so those could be auto-sent to ml and reviewed
[16:39] <ubitux> also, we use the trac wiki because we already have it for free (because of the trac bug tracker)
[16:39] <burek> but I don't want to complicate things, if people are not interested enough in it
[16:39] <cbsrobot> ubitux: I prefer the git aswell, but at least the users get a chance to help
[16:40] <cbsrobot> the main doc will always stay in git
[22:43] <CIA-41> ffmpeg: 03Justin Ruggles 07master * rdb0e7bd0bd 10ffmpeg/libavresample/audio_mix.c: lavr: mix: validate internal sample format in ff_audio_mix_init()
[22:43] <CIA-41> ffmpeg: 03Justin Ruggles 07master * r6410397600 10ffmpeg/libavresample/ (resample.c resample_template.c utils.c): 
[22:43] <CIA-41> ffmpeg: lavr: resampling: add support for s32p, fltp, and dblp internal sample formats
[22:43] <CIA-41> ffmpeg: Based partially on implementation by Michael Niedermayer <michaelni at gmx.at> in
[22:43] <CIA-41> ffmpeg: libswresample in FFmpeg. See commits:
[22:43] <CIA-41> ffmpeg: 7f1ae79d38c4edba9dbd31d7bf797e525298ac55
[22:43] <CIA-41> ffmpeg: 24ab1abfb6d55bf330022df4b10d7aec80b3f116
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r5b7a88f0ba 10ffmpeg/libavformat/matroskadec.c: matroskadec: honor error_recognition when encountering unknown elements.
[22:43] <CIA-41> ffmpeg: 03JULIAN GARDNER 07master * r0dd283faca 10ffmpeg/libavcodec/libx264.c: 
[22:43] <CIA-41> ffmpeg: libx264: support aspect ratio switching
[22:43] <CIA-41> ffmpeg: Signed-off-by: Anton Khirnov <anton at khirnov.net>
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * rcd99146253 10ffmpeg/libavfilter/ (15 files): lavfi: add error handling to filter_samples().
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r1ffb645662 10ffmpeg/libavfilter/af_resample.c: 
[22:43] <CIA-41> ffmpeg: af_resample: fix request_frame() behavior.
[22:43] <CIA-41> ffmpeg: Make sure that an output frame has really been produced before returning
[22:43] <CIA-41> ffmpeg: 0.
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r6f834293d3 10ffmpeg/libavfilter/af_asyncts.c: 
[22:43] <CIA-41> ffmpeg: af_asyncts: fix request_frame() behavior.
[22:43] <CIA-41> ffmpeg: Make sure that an output frame has really been produced before returning
[22:43] <CIA-41> ffmpeg: 0.
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r8d18bc550e 10ffmpeg/tests/fate-run.sh: fate-run: make avconv() properly deal with multiple inputs.
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r083d0f6be8 10ffmpeg/libavfilter/split.c: asplit: don't leak the input buffer.
[22:43] <CIA-41> ffmpeg: 03Anton Khirnov 07master * r8a645bfc31 10ffmpeg/avconv.c: avconv: propagate fatal errors from lavfi.
[22:43] <CIA-41> ffmpeg: 03Michael Niedermayer 07master * rf8911b987d 10ffmpeg/: (log message trimmed)
[22:44] <CIA-41> ffmpeg: Merge remote-tracking branch 'qatar/master'
[22:44] <CIA-41> ffmpeg: * qatar/master:
[22:44] <CIA-41> ffmpeg:  mss3: use standard zigzag table
[22:44] <CIA-41> ffmpeg:  mss3: split DSP functions that are used in MTS2(MSS4) into separate file
[22:44] <CIA-41> ffmpeg:  motion-test: do not use getopt()
[22:44] <CIA-41> ffmpeg:  tcp: add initial timeout limit for incoming connections
[22:44] <CIA-41> ffmpeg: 03Diego Biurrun 07master * r8e85ba8d13 10ffmpeg/configure: 
[22:44] <CIA-41> ffmpeg: configure: Change the rdtsc check to a linker check
[22:44] <CIA-41> ffmpeg: A compile-only test could succeed with an implicit function declaration.
[22:44] <CIA-41> ffmpeg: Also move the check to the x86 section; rdtsc is x86-only.
[22:44] <CIA-41> ffmpeg: 03Jordi Ortiz 07master * r5427df0186 10ffmpeg/libavformat/tcp.c: 
[22:44] <CIA-41> (13 lines omitted)
[00:00] --- Tue Jul 10 2012