[Ffmpeg-devel-irc] ffmpeg-devel.log.20130510

[burek]

[00:23] <ubitux> michaelni: http://pastie.org/7824488
[00:30] <michaelni> ubitux, is there a bug or you just want me to silcene the warnings ?
[00:31] <ubitux> i don't know if there is a bug, i just like it when compilation is not cluttered with all kind of warnings hiding others
[00:31] <ubitux> and those look recent
[00:31] <ubitux> maybe they were missed for some reason (already too much warnings?), so i was just pointing them out
[01:47] <cone-954> ffmpeg.git 03Michael Niedermayer 07master:8ef453ff830b: sws/output: silence warnings about uninitialized use of variables
[01:47] <cone-954> ffmpeg.git 03Michael Niedermayer 07master:500220a8e84a: mimic: Fix race condition
[02:43] <cone-954> ffmpeg.git 03Michael Niedermayer 07fatal: ambiguous argument 'refs/tags/n1.2.1': unknown revision or path not in the working tree.
[02:43] <cone-954> Use '--' to separate paths from revisions
[02:43] <cone-954> refs/tags/n1.2.1:HEAD: mimic: Fix race condition
[03:03] <ubitux> http://pastie.org/7824987  am i doing something stupid, or is it the compiler?
[03:24] <Skyler_> ubitux: does it get better if you use -ffast-math?
[03:25] <ubitux> mmh that's indeed better
[03:25] <ubitux> well, only with gcc though :)
[03:27] <ubitux> Skyler_: any idea why this optim is not done by default?
[03:27] <ubitux> maybe some precision issue or something?
[04:00] <Skyler_> because it's not a valid optimization
[04:00] <Skyler_> (x*x)*(x*x) != ((x*x)*x)*x
[04:00] <Skyler_> float arithmetic is not associative.
[04:00] <Skyler_> if I remember right, at least
[04:12] <iive> ubitux: what is the problem exactly?
[04:23] <iive> well, later...
[09:45] <cone-267> ffmpeg.git 03Luca Barbato 07master:ee9a4dff4ad7: mmf: K&R formatting cosmetics
[09:45] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:c475a58dd60c: Merge remote-tracking branch 'qatar/master'
[10:40] <cone-267> ffmpeg.git 03Stefano Sabatini 07master:eeb9a4928dcc: lavu/mem.h: fix typo
[10:40] <cone-267> ffmpeg.git 03Stefano Sabatini 07master:e1332ff324e9: lavfi/avfilter.h: fix typo
[10:40] <cone-267> ffmpeg.git 03Stefano Sabatini 07master:5c4c929878f4: lavfi/avfiltergraph: fix style
[10:41] <cone-267> ffmpeg.git 03Carl Eugen Hoyos 07master:0f05b72f19c0: Fix configure --help output for some autodetected features.
[10:53] <cone-267> ffmpeg.git 03Carl Eugen Hoyos 07release/1.2:33d699a4e73d: Fix configure --help output for some autodetected features. (cherry picked from commit 0f05b72f19c0a07cf48c912d2a71c3fb5f7bc325)
[10:55] <durandal_1707> saste: select filter have changed PICT_TYPE_I to just I at some point in time, is this expected?
[10:56] <saste> durandal_1707, no
[10:59] <durandal_1707> so should i add PICT_TYPE_I/.. back for backward compatibility?
[10:59] <saste> durandal_1707, if you want, yes seems a good idea
[11:02] <nevcairiel> whats with the lazy fate boxes
[11:04] <durandal_1707> perhaps some matchines have been disconnected/disabled
[11:15] <durandal_1707> i get this when seeking with ffplay with keys: [matroska,webm @ 0x29e37100] File is broken, keyframes not correctly marked!
[11:16] <durandal_1707> and same if i seek with mpv before the start ...
[11:18] <durandal_1707> saste: any comments on avectorscope?, i gonna apply it soon
[12:13] <ubitux> saste: s#"tcp://localhost:5555"#"tcp://localhost:5555" if len(sys.argv) < 2 else sys.argv[1]#
[12:14] <ubitux> well i could improve the script :p
[12:15] <saste> ubitux, any way to generate auto docs, like as in perl scripts
[12:15] <ubitux> yes
[12:15] <saste> (check tools/plotframes)?
[12:16] <ubitux> basically you put some kind of doxy into scopes with ''' blabla '''
[12:16] <ubitux> then you can pydoc the thing
[12:29] <durandal_1707> ^ looks like parser issue to me (that code is nightmare)
[12:36] <durandal_1707> dint filter is useful?
[13:02] <cone-267> ffmpeg.git 03highgod0401 07master:63e8fc4d8f08: lavfi: modify names of deshake_kernel.h and unsharp_kernel.h
[13:04] Action: ubitux still thinks opencl prefix would have been better but well..
[13:09] <durandal_1707> move to separate directory so it can be removed with simple rm -rf opencl/
[13:17] <durandal_1707> there is no example how to use ffmeta files to add metadata when muxing?
[13:51] <ubitux> ok, got a better vignette filter.
[13:52] Action: durandal_1707 wants 3dlut
[13:53] <ubitux> durandal_1707: it's also done
[13:53] <ubitux> i'll submit soon, i just need to do something first
[15:17] <ubitux> any idea what could have cause such broken blocks? http://ubitux.fr/pub/pics/_broken-blocks.png (8x8, jpeg blocks)
[15:18] <ubitux> and if there is a way to fix them without denoising
[15:31] <saste> ubitux: do you have a command to show the chroma subsampling artifacts fixed by your patch?
[15:32] <ubitux> saste: it's mostly untested, so no :)
[15:32] <ubitux> ask durandal_1707, he did a few tests
[15:33] <saste> durandal_1707, ^^
[15:37] <ubitux> saste: we need more creepy tests like this btw
[15:40] <saste> ubitux, testsrc=s=223x103,format=yuv410p,hue=s=0,format=yuv410p
[15:40] <saste> but in general the effect is not very visible, and thus pass unnoticed
[15:42] <ubitux> heh.
[15:43] <ubitux> s/testsrc/mandelbrot/
[15:49] <ubitux> i'll apply the timeline patchset slightly differently soon
[16:15] <ubitux> http://lucy.pkh.me/old-scrat.webm
[16:15] <ubitux> ./ffmpeg -ss 10 -i ~/samples/GoneNutty.avi -vf "curves=vintage,vignette=PI/4+random(1)*PI/50:eval=frame,format=yuv420p,noise=c0s=5:c0f=t" -b:v 5000000 -t 20 -y old-scrat.webm
[16:15] <durandal_1707> saste: you sure that is testsrc bug? tried generating pngs and playing them with non-ffplay app?
[16:15] <ubitux> noise is not really doing what i'm expecting, but good enough for npw
[16:15] <ubitux> now*
[16:16] <durandal_1707> what kind of noise you want?
[16:16] <ubitux> i'm trying to make an "old" effect
[16:17] <durandal_1707> ubitux: stop using vp8
[16:17] <ubitux> no idea what kind of noise i should be trying
[16:17] <saste> durandal_1707, #2555 is a testsrc bug
[16:17] <ubitux> durandal_1707: playable in browsers, so handy
[16:18] <durandal_1707> saste: yes, you are correct, i thought it may be ffplay issue
[16:19] <durandal_1707> ubitux: mkv should be playable too
[16:19] <ubitux> durandal_1707: no
[16:19] <durandal_1707> and i do not play videos in brauzer
[16:20] <ubitux> mpv in a tty on your freebsd? :)
[16:20] <durandal_1707> saste: the ubitux's chroma subsampling patch should fix issues when width is not multiple of subsampled chrome values (log_chrome_v/h << 1)
[16:21] <durandal_1707> *width/height
[16:21] <ubitux> want me to apply it without extensive testing?
[16:21] <durandal_1707> ubitux: that noise in sample you uploaded is little repeating too much
[16:22] <saste> ubitux: i'm fine assuming that FATE is fine with it
[16:22] <ubitux> saste: sure
[16:22] <saste> as I wrote it LGTM
[16:22] <ubitux> ok, will apply soon then
[16:23] <durandal_1707> you could try other flags: c0f=t+a+u:c0s=13
[16:25] <durandal_1707> well that combination needs higher strength
[16:25] <ubitux> yes i tried with average
[16:26] <ubitux> but it makes a gummy effect
[16:30] <ubitux> it would be awesome if testsrc could keep the circle round when playing with its ar
[16:31] Action: durandal_1707 wonders from where fish effect is
[16:32] <ubitux> you could adapt the vignette filter to make a fish effect
[16:32] <ubitux> well i guess i'll have to write the doc and submit it
[16:33] <saste> durandal_1707, VHOOK
[16:48] <durandal_1707> from where ami get all those game files...
[16:51] <kierank> Amiga stuff I guess
[17:00] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:a079ed63d3c0: avidec: Enable full parsing for potencoder mpeg1/2
[17:07] <ubitux> saste: my patch seems to fix the right border, but not the botton line
[17:07] <saste> ubitux, what's the problem with the bottom line?
[17:08] <ubitux> still green
[17:08] <durandal_1707> ubitux: what filter?
[17:08] <saste> in my test i couldn't notice any artifact in the bottom line...
[17:08] <saste> ubitux: may be an SDL issue
[17:08] <cone-267> ffmpeg.git 03Paul B Mahol 07master:d751a2526f9b: lavfi/anull: remove redundant lines
[17:08] <saste> uhm not so sure about that...
[17:10] <ubitux> only reproducible with ffplay though
[17:10] <ubitux> http://b.pkh.me/hue-old.png ’ http://b.pkh.me/hue-new.png
[17:10] <ubitux> (ffmpeg ^)
[17:12] <ubitux> saste: and ffplay: http://b.pkh.me/ffplay-hue-old.png ’ http://b.pkh.me/ffplay-hue-new.png
[17:12] <ubitux> but crop/pad might need fixing as well
[17:14] <durandal_1707> swscale may have own bugs, so use extractplanes
[17:15] <saste> ubitux, which command are you using?
[17:15] <ubitux> saste: yours, ffplay -f lavfi testsrc=s=223x103,format=yuv410p,hue=s=0
[17:17] <ubitux> well anyway, that's at least better with the patch
[17:17] <ubitux> but i think we'll need to have a deeper look into at least pad, crop, and possibly sws
[17:18] <ubitux> (and the other(s?) filter(s?) i mentioned)
[17:20] <saste> i cannot observe the issue with ffplay, only if i use ffmpeg
[17:20] <durandal_1707> ubitux: with your patch and ffmpeg -f lavfi -i testsrc=s=801x603,format=yuv410p,hue=s=0 /tmp/1.png its looks fine in mpv
[17:21] <ubitux> i'm going to push the patch in a moment so we can have a better basis
[17:22] <durandal_1707> same with ffplay
[17:23] <ubitux> i still have a bottom green line here
[17:24] <saste> ubitux, what if you convert to png and display it?
[17:25] <ubitux> saste: i can only reproduce with ffplay
[17:25] <durandal_1707> sdl bug probably
[17:25] <saste> ubitux, yes
[17:26] <saste> does SDL supports yuv410p?
[17:26] <ubitux> ./ffmpeg -f lavfi -i testsrc=s=801x603,format=yuv410p,hue=s=0,format=yuv420p -f sdl - 
[17:26] <ubitux> completely broken :))
[17:27] <durandal_1707> oh yes, i got that yesterday (forgot about it completely)
[17:27] <cone-267> ffmpeg.git 03Clément BSsch 07master:50e66726a237: lavfi: use ceil right shift for chroma width/height.
[17:28] <saste> is that another SDL bug? the same one affecting ffplay?
[17:28] <ubitux> no, a different bug :)
[17:29] <durandal_1707> or avdevice bug
[18:45] <saste> ubitux: what I need is a lightweight dynamic buffer API
[18:46] <saste> a single function should be enough, and is more or less equivalent to the alloc_elem() function, but more flexible
[18:46] <saste> indeed I'm surprised we don't already have something similar
[18:49] <ubitux> we already have avbprint and avio_dyn_buf things
[18:49] <ubitux> for this purpose
[18:50] <ubitux> the dynamic array is useful for other purposes, and we already have various use cases
[18:52] <saste> uhm i'm confused
[18:59] <ubitux> ETOOMUCHFILTERS
[18:59] <durandal_1707> ?
[19:00] <ubitux> durandal_1707: 3 pending filters :)
[19:00] <ubitux> dctdnoiz, vignette and lut3
[19:00] <ubitux> lut3d*
[19:01] <ubitux> with avectorscope and zmq we're going to have quite a bunch of nice filters
[19:01] <ubitux> comments welcome on timeline btw
[19:03] <durandal_1707> how is lut3d used?
[19:03] <ubitux> lut3d=foobar.3dl
[19:04] <durandal_1707> and how i get foobar ?
[19:04] <ubitux> it's hard to find samples though
[19:04] <durandal_1707> what such file actually contains?
[19:04] <ubitux> parse error
[19:04] <ubitux> ah
[19:04] <ubitux> it contains... r,g,b values
[19:04] <ubitux> see the pandora example
[19:05] <ubitux> (not yet supported in this filter though)
[19:05] <ubitux> we need a slow motion filter :(
[19:06] <ubitux> anyway, i have a few hours for me right now
[19:06] <ubitux> should i port a mp filter?
[19:07] <ubitux> (if so, which one?)
[19:07] <durandal_1707> dint?
[19:07] <ubitux> i don't like the name
[19:07] <ubitux> reminds me bad nightmares
[19:08] <durandal_1707> ow
[19:09] <ubitux> okay
[19:09] <ubitux> durandal_1707: please suggest a name
[19:09] <ubitux> :)
[19:09] <ubitux> wldnoiz?
[19:10] <durandal_1707> ovrcmpltwvltdnisr
[19:10] <durandal_1707> ovrcmpltwvltdnsr
[19:11] <ubitux> hell yeah
[19:11] <durandal_1707> just do eq?
[19:12] <ubitux> too late i started ow
[19:12] <ubitux> feel free to do eq2
[19:12] <ubitux> so we can drop two filters at the same time
[19:13] <ubitux> nice, unused mode in ow
[19:13] <ubitux> and unused delta
[19:16] <durandal_1707> well i'm more interested in this http://docs.gimp.org/en/gimp-tool-levels.html (something like (auto)levels in avisynth) , deflicker and mergeplanes
[19:16] <ubitux> i think you got lost in your quest to kill mp wrapper
[19:17] <durandal_1707> i can remove non-functional softpulldown, crashing ilpack, and broken/flawed fil right now
[19:18] <ubitux> fine with me
[19:18] <ubitux> but i find it more fun when there is a new filter :)
[19:20] <ubitux> no one motivated to add gimp presets support to vf curves btw?
[19:24] <cone-267> ffmpeg.git 03Paul B Mahol 07master:005ee7a5867e: lavfi: avectorscope filter
[19:25] <ubitux> :)
[19:41] <cone-267> ffmpeg.git 03Paul B Mahol 07master:4868855b1305: lavfi/select: set forgotten VAR_PICT_TYPE_BI
[19:44] <durandal_1707> AV_PICTURE_TYPE_S is not set by any decoder/parser
[20:43] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:7ded0179832b: sws/output/yuv2gbrp_full_X_c(): silence warning about uninitialized variable
[20:43] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:b2cf655d6e3d: sws: add casts to silence pointer type mismatch warnings
[20:46] <ubitux> michaelni: thx ^
[20:48] <durandal_1707> michaelni: you fixing bogus warnings of stupid compilers
[20:49] <michaelni> all compilers are stupid
[20:49] <durandal_1707> some less
[20:52] <ubitux> durandal_1707: such as?
[20:52] <durandal_1707> anyone have amarok or banshee installed, or foobar2k?
[20:55] <llogan> why those?
[20:55] <durandal_1707> for id3v2 chapters
[21:07] <BBB> Yuvi: hey, want to write a vp9 decoder? ;)
[21:34] <ubitux> yay, ow ported.
[21:55] <Daemon404> 'ow' ?
[21:55] <ubitux> Overcomplete Wavelet denoiser.
[21:55] <ubitux> i'm sure you already love the name
[21:56] <Daemon404> it certainly makes me skeptical :)
[21:56] <ubitux> i'm not sure from where the "overcomplete" comes from though
[21:56] <ubitux> since there is only 3 option
[21:57] <durandal_1707> and one not implemented
[21:57] <ubitux> i counted only the implemented ones
[21:57] <ubitux> the two strength and the depth
[21:58] <durandal_1707> if depth is too big it cause crash iirc
[21:58] <ubitux> i've fixed that
[21:58] <durandal_1707> so is it better than dtcdntz?
[21:58] <ubitux> no idea
[21:58] <ubitux> (careful, "dtc" means "iya" in french)
[21:59] <durandal_1707> dctdnoiz, i'm not one who make up such names
[21:59] <ubitux> :)
[21:59] <ubitux> dctdnoiz really is slow
[22:00] <ubitux> at least if you don't play with overlap
[22:10] <ubitux> okay, ow submitted
[22:10] <ubitux> still have some time
[22:10] <ubitux> durandal_1707: which one now?
[22:11] <durandal_1707> qp?
[22:11] <ubitux> erh
[22:11] <ubitux> no thx
[22:11] <durandal_1707> phase filter looks nice (syntax)
[22:12] <ubitux> michaelni: you wanted vf spp at some point, no?
[22:12] <durandal_1707> i think this it is obsoleted by pp
[22:13] <ubitux> nope, it's more effective afaik
[22:13] <durandal_1707> beating complete libpostroc?
[22:13] <ubitux> it had a pretty decent output iirc
[22:14] <durandal_1707> sab looks trivial
[22:14] <ubitux> go for it
[22:15] <durandal_1707> the only big monster left is mcdeint
[22:16] <michaelni> ubitux, yes, i want spp at some point 
[22:16] <ubitux> michaelni: any other filter in priority?
[22:17] <michaelni> all that are popular with the movie/anime/pr0n  encoding communities
[22:17] <durandal_1707> that is none
[22:17] <durandal_1707> except pullup all others should be removed, and better filters written from scratch
[22:18] <michaelni> i didnt think of libmpcodecs here ...
[22:18] <ubitux> my question was porting to libmpcodecs
[22:18] <ubitux> since i want to get done with that stuff
[22:18] <ubitux> so we can focus on some more important stuff
[22:18] <michaelni> porting TO is a interresting idea
[22:18] <ubitux> TO?
[22:19] <michaelni> from libavfilter to libmpcodecs ;)
[22:19] <ubitux> ah haha sorry
[22:19] <ubitux> anyway, porting from mp
[22:19] <michaelni> but actually, some libavfilter wraper would make sense for mplayers libmpcodecs
[22:20] <ubitux> mpv has one
[22:20] <ubitux> mplayer too afaik, but broken iirc
[22:20] <durandal_1707> talk about that on #mplayer
[22:20] <ubitux> anyway, so next mp to port would be spp?
[22:20] <durandal_1707> or pp777
[22:21] <ubitux> no one wants to port some other? :(
[22:21] <durandal_1707> if i get really bored may write supereq
[22:21] <ubitux> michaelni: so if i port spp, you ok for dropping mp={fspp,pp7,spp,uspp}?
[22:22] <michaelni> i think some of these are faster than spp so i dont think they should be lost
[22:23] <ubitux> don't want to help porting them then? :(
[22:26] Action: michaelni doesnt see a major problem with using them through vf mp
[22:27] <michaelni> so i would probably rather work on adding things that can currently not be used at all
[22:27] <ubitux> mp is supposed to be a temporary solution :(
[22:27] <ubitux> michaelni: can you work on a fast dct 2d then? ;)
[22:28] <ubitux> so we can have a faster dctdnoiz
[22:29] <michaelni> is there no lgpl compatible fast dct2d code available that we could use ?
[22:31] <michaelni> It would be less work to reuse existing code than write it new
[22:31] <ubitux> possibly
[22:32] <ubitux> well anyway, ETOOMUCHWORK today
[22:32] <ubitux> i'll go watch some animes waiting for reviews before i do anything else
[22:32] <ubitux> cya
[22:32] Action: ubitux &
[22:36] <michaelni> the different pp and wavelet filters could also be compared before porting, that is if any is always worse qiality & speed it could be skiped
[22:56] <cone-267> ffmpeg.git 03Luca Barbato 07master:72fbc9685c08: avconv: use current syntax for the scale filter
[22:56] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:fc94b68eb53f: Merge commit '72fbc9685c08baae7ab9996642707eaab873b6f1'
[23:04] <cone-267> ffmpeg.git 03Anton Khirnov 07master:16a4a18db089: af_asyncts: fix offset calculation
[23:04] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:2a7d71984805: Merge commit '16a4a18db089af8c432f1cdec62155000585b72c'
[23:09] <Daemon404> so uh
[23:09] <Daemon404> can someone regenerate the site docs
[23:09] <Daemon404> ffmpeg.org/filters.html is 5 months old
[23:13] <durandal_1707> can't be, force refresh?
[23:13] <ubitux> ffmpeg-filters.html
[23:13] <Daemon404> o
[23:13] <ubitux> filters.html should not exist anymore
[23:13] <Daemon404> well it does
[23:13] <ubitux> then it should be removed, or symlinked 
[23:14] <ubitux> the documentation page should not reference it anymore anyway
[23:18] <michaelni> filters.html deleted
[23:19] <michaelni> i can setup a redirect if something off site points to it still ?
[23:20] <ubitux> did you delete codecs.html etc?
[23:20] <ubitux> (if any)
[23:21] <cbsrobot> one thing that could be improved is the site title
[23:21] <cbsrobot> it's always "FFmpeg documentation : : "
[23:21] <cbsrobot> even if it's ffplay ...
[23:21] <michaelni> ubitux, didnt delete anything else
[23:21] <cbsrobot> but I know I should send a patch - so ...
[23:23] <michaelni> ubitux, if you want a directory listing to check the files there, i can provide one 
[23:23] <ubitux> michaelni: ls *.html welcome
[23:23] <durandal_1707> ubitux: ow is not LGPL
[23:24] <ubitux> durandal_1707: oh oups
[23:24] <ubitux> will fix asap
[23:25] <michaelni> ubitux, list sent
[23:25] <ubitux> thx
[23:26] <cone-267> ffmpeg.git 03Christophe Gisquet 07master:2c299d4165cd: x86: sbrdsp: implement SSE2 qmf_pre_shuffle
[23:26] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:5e1278c6401a: Merge commit '2c299d4165cd9653153e12270971c2368551b79e'
[23:27] <ubitux> michaelni: 'seems like a nice mess :))
[23:28] <ubitux> michaelni: i think decoders.html bitstream_filters.html eval.html protocols.html muxers.html encoders.html demuxers.html indevs.html syntax.html outdevs.html metadata.html can be dropped
[23:29] <ubitux> michaelni: according to the owner/group and date, they are not updated anymore
[23:29] <ubitux> and likely replaced by the ffmpeg-*.html
[23:32] <cone-267> ffmpeg.git 03Janne Grunau 07master:94235f2ba2eb: indeo3: avoid writes without necessary alignment in copy_cell()
[23:32] <cone-267> ffmpeg.git 03Michael Niedermayer 07master:496f2df9084b: Merge remote-tracking branch 'qatar/master'
[23:34] <michaelni> ubitux, done
[23:34] <ubitux> :)
[23:35] <ubitux> michaelni: so what lavc wavelet code in particular were you thinking about?
[23:37] <michaelni> *dwt*.{c,h}
[23:38] <ubitux> no dwtutils? :
[23:38] <ubitux> :(
[23:50] <cone-267> ffmpeg.git 03Clément BSsch 07master:a9705e4de9c0: lavfi/curves: avoid ctx pointer indirection.
[00:00] --- Sat May 11 2013