[Ffmpeg-devel-irc] ffmpeg-devel.log.20130309

[burek]

[00:11] <ubitux> is there some yuv pixfmt with 4 planes?
[00:11] <ubitux> i see this:
[00:11] <ubitux>     for (plane = 0; plane < desc->nb_components; plane++) {
[00:11] <ubitux>         int lines = plane == 1 || plane == 2 ? src_h >> vsub : src_h;
[00:12] <ubitux> i guess this is only relevant to handle both yuv and rgb*?
[00:12] <saste> ubitux, yuva?
[00:13] <ubitux> mmh.
[00:14] <ubitux> so luma and alpha, no shift, full scale all the time
[00:17] <saste> ubitux, why no shift?
[00:17] <saste> yuva420p is good as well
[00:17] <saste> note that the alpha channel is never subsampled
[00:18] <ubitux> yes that was my question
[00:18] <ubitux> no subsampling for luma and alpha
[00:18] <ubitux> in the current filter alpha won't be supported, so i guess i can change the condition to plane != 0
[00:19] <ubitux> can the 4th comp used for sth else than alpha?
[00:19] <saste> ubitux, right now, no
[00:19] <ubitux> ok
[00:19] <ubitux> thanks
[01:14] <saste> wow option handling in ff* tools is more and more convoluted
[01:15] <saste> it's finally "clean" after more and more refantoning
[02:46] <cone-317> ffmpeg.git 03Stephen Hutchinson 07master:98c4268ecac1: libutvideo: Prioritize compatibility with new versions
[11:31] <Cigaes> Hi.
[11:34] <michaelni> Cigaes, hi
[11:35] <Cigaes> Hi. Do you know if someone is working on af_aconvert?
[11:35] <michaelni> noone works on it
[11:38] <michaelni> ubitux, ive updated a few more filters & fixed a few bugs compared to yesterday
[11:39] <michaelni> Cigaes, did you had time to look at the buffer src/sink stuff ?
[11:40] <Cigaes> I am on it, but I have a hard time working on something that I can not test.
[11:40] <Cigaes> I have got af_aconvert to compile, how shall I submit it to you?
[11:41] <michaelni> do you have a github repo ?
[11:41] <michaelni> or any git repo? if so push there and ill merge
[11:41] <michaelni> otherwise git format-patch /send-email is fine too
[12:01] <cbsrobot_> hi Cigaes. Nice to see you on irc.
[12:08] <Cigaes> cbsrobot_: I am afraid I do not know who you are.
[12:09] <Cigaes> michaelni: my upload is seriously limited right now, and I am not good at creating push git repos, so I will go with send-email when I have something good.
[12:12] <michaelni> Cigaes, ok, btw iam constantly fixing things in my repo so dont forget to fetch from there from time to time
[12:22] <ubitux> hi Cigaes :)
[12:23] <ubitux> Cigaes: if we get the buffer sink / src working, we may have limited builds and tests possibles
[12:23] <ubitux> michaelni: nice, ok
[12:40] <Cigaes> ubitux: not only: all of ffmpeg*.c must be converted to remove the use of AVFilterBufferRef too.
[12:42] <ubitux> oh, ok.
[12:42] <ubitux> michaelni: added tinterlace
[12:45] <michaelni> ubitux, we have that duplicate now :)
[12:45] <ubitux> haha
[12:46] <ubitux> you can get alphaextract too then
[12:46] <ubitux> michaelni: ok, i'm taking removelogo/stereo3d/superx2sai/swapuv now
[12:47] <michaelni> ok, ubitux witgh swapuv be carefull we might need to swap more than what was previously swaped
[12:48] <ubitux> mmh ok
[12:48] <ubitux> i forgot some stuff in alphamerge
[12:49] <michaelni> ubitux, can you rebase so i can do a fast forward merge ?
[12:49] <wm4> will you actually merge the removal of the libavcodec resampler?
[12:49] <ubitux> michaelni: yeah right give me a moment
[12:51] <ubitux> michaelni: rebased; alphaextract and removelogo ok
[12:51] <michaelni> ill update deshake
[12:51] <ubitux> (brainfart for alphamerge, i'm not working on it)
[12:52] <ubitux> (rebased now for real)
[12:56] <michaelni> Cigaes, libav didnt update avconv.c in the commit that changed lavfi to AVFrame but then i did not check if that actually works
[12:57] <nevcairiel> the user-facing api supposedly has backwards compat
[12:57] <michaelni> it certainly does not have ABI compatibility though
[12:57] <Cigaes> michaelni: the I will have to look closer and add the compat functions.
[12:57] <nevcairiel> no, it was version bumped because of that
[12:59] <michaelni> deshake done
[12:59] <nevcairiel> are there still many filters to do, worth joining in for a while?
[13:01] <ubitux> michaelni: i've rebased and pushed stereo3d, superx2sai
[13:02] <michaelni> pushed
[13:02] <michaelni> nevcairiel, see https://github.com/michaelni/FFmpeg/tree/lavfi-merge-mess
[13:03] <wm4> nevcairiel: the backward compat worked well for me, even though I had no less than 2 custom get_buffers
[13:04] <ubitux> any idea what frame->base is?
[13:05] <nevcairiel> should be derpecated now, used to hold pointers to the allocated memory, iirc
[13:05] <nevcairiel> because data could be offset or something
[13:05] <michaelni> base was the allocated origin
[13:06] <michaelni> data could be offset for example by 16 pixels right and down 
[13:06] <michaelni> for edges 
[13:06] <ubitux> i guess i'll have to update them in swapuv then
[13:13] <ubitux> michaelni: added swapuv and a fixup for vf pp
[13:16] Action: ubitux working on life and cellauto now
[13:19] <nevcairiel> i pushed aconvert to https://github.com/Nevcairiel/FFmpeg/tree/lavfi-merge
[13:20] <ubitux> hopefully Cigaes was not on this one ;)
[13:20] <nevcairiel> oh, i tried to read up some history, but didnt see that
[13:20] <nevcairiel> oh well, was a nice excercise :p
[13:20] <ubitux> :)
[13:20] <nevcairiel> maybe he can just look at the change and sign off on it then ;)
[13:21] <nevcairiel> anyone doing the next audio things then, amerge, asetnsamples?
[13:22] <ubitux> i'm still on video for now
[13:22] <Cigaes> I am still on buffersrc / buffersink.
[13:23] <Cigaes> I have somethink that links and runs but do not convert anything.
[13:24] <ubitux> ffghost
[13:26] Action: ubitux working on histogram and kerndeint
[13:30] <ubitux> done
[13:31] <ubitux> michaelni: care to do a little merge?
[13:31] <michaelni> done
[13:31] <ubitux> thx
[13:33] <michaelni> nevcairiel, cherry picked & pushed
[13:37] <nevcairiel> i pushed amerge and asetnsamples
[13:37] <nevcairiel> rebased
[13:37] Action: ubitux going to do sendcmd
[13:37] <mateo`> which filters are not ported yet to avframe ?
[13:38] <ubitux> mateo`: git grep AVFilterBufferRef
[13:38] <ubitux> wow sendcmd looks easy..
[13:39] <nevcairiel> actually, i'm not sure asetnsamples is correct yet, let me revisit this
[13:40] Action: ubitux working on avf_concat
[13:41] <ubitux> arg, depends on buffersink, just like overlay, i can't..
[13:41] <nevcairiel> ah foudn the missing piece for asetnsamples
[13:41] <ubitux> Cigaes: do you have a compiling buffer sink / src? (even if it doesn't work)
[13:42] <ubitux> so we can test build for filters overlay, concat, etc
[13:42] <Cigaes> I have something that builds and links and does not work,
[13:42] <Cigaes> but it affects a lot of things.
[13:43] <Cigaes> I can produce a big diff, but not clean commits.
[13:43] <mateo`> is anyone working on vf_decimate ?
[13:43] <ubitux> mateo`: not me
[13:43] <ubitux> Cigaes: could it make sense to add it to the merge conflict branch?
[13:43] <nevcairiel> ok now there is two audio filters in my branch, but i need to prepare for going out now, stupid family event today .. if it happens that the decoder merge is tomorrow, i'll be around to help
[13:43] <ubitux> that might unblock a little the process
[13:44] <ubitux> nevcairiel: have fun
[13:44] <nevcairiel> i wont, but thanks
[13:44] <nevcairiel> :P
[13:44] <ubitux> :)
[13:45] <ubitux> wm4: is there a problem with the lavc resampler removal?
[13:46] <nevcairiel> some people dont seem to like the additional library that is the replacement
[13:46] <Cigaes> ubitux: I will try to produce something soon.
[13:46] <wm4> ubitux: just wondering... because then people _really_ will have to decide between libavresample/libswresample
[13:47] <ubitux> wm4: the merge process is kinda of slowed down right now because of the evil plan anyway..
[13:47] <mateo`> i take vf_decimate
[13:59] <ubitux> nice, github going down at the appropriate moment.
[14:04] <ubitux> yay it's back.
[14:07] <mateo`> michaelni: https://github.com/mbouron/FFmpeg/tree/lavfi-merge-mess, vf_decimate ported to AVFrame
[14:07] <nevcairiel> i had a few minutes left and converted atempo as well
[14:08] <ubitux> merging from libav, the last junkies addiction
[14:09] <nevcairiel> i actually like doing such work <.<
[14:09] <ubitux> :)
[14:09] <ubitux> it's good for ya they said
[14:14] <michaelni> nevcairiel, mateo` ubitux all merged
[14:16] <nevcairiel> pushed af_biquads as well, but now i really need to go, or i miss my train :d
[14:16] <nevcairiel> (rebased as well)
[14:17] <michaelni> nevcairiel, merged
[14:25] Action: ubitux working on earwax
[14:29] <michaelni> Cigaes, btw, dunno if i mentioned already but the buffer pool has been disabled and the buffer init memset too (theres some chance these could cause bugs as they are untested)
[14:29] <maister> michaelni, ever tested frequency response of ffmpeg resampler?
[14:30] Action: ubitux working on evalsrc
[14:30] <saste> Cigaes, hey :)
[14:30] <ubitux> aevalsrc*
[14:30] <Cigaes> saste: Hi.
[14:30] <mateo`> taking volumedetect
[14:30] <Cigaes> michaelni: the disabled buffer pool does not seem to be a problem. What buffer init memset are you refering to?
[14:31] <saste> i have ~2 hours to work on mergestuff, on what should i work?
[14:31] <michaelni> Cigaes,  memset(data[0], 128, i);
[14:31] <ubitux> saste: checkout the branch from michael on github
[14:31] <saste> ubitux, done
[14:31] <ubitux> saste: Cigaes is on buffersink/src, mateo` on volumedetect, i'm on aevalsrc
[14:32] <ubitux> just announce when you work on something and ask michael to merge your ports
[14:32] <ubitux> we only test build right now btw
[14:32] <saste> where should i send the patches?
[14:32] <ubitux> i think that's simpler for michael if you have a branch on your github or wherever
[14:32] <saste> ok
[14:33] <ubitux> saste: i ported earwax too btw
[14:33] <ubitux> (which is not yet pushed)
[14:34] <cone-370> ffmpeg.git 03Michael Niedermayer 07release/1.2:bdd2db60c2a6: MAINTAINERS: update for 1.2
[14:34] <cone-370> ffmpeg.git 03Michael Niedermayer 07release/1.2:1ea3248290a3: Update for FFmpeg 1.2
[14:35] <mateo`> volumedetect done
[14:35] <ubitux> michaelni: earwax and aevalsrc done
[14:36] <maister> michaelni, Ran an impulse through libswresample with ratio 2.0 and got this result: http://i.imgur.com/HsB02Lh.png. Doesn't seem right for cutoff to *start* at half BW.
[14:36] Action: ubitux working on vf blend
[14:38] Action: saste working on framestep
[14:39] Action: ubitux working on concat
[14:39] <michaelni> merged
[14:40] <Cigaes> Video transcoding working.
[14:40] <ubitux> yay :)
[14:40] <ubitux> michaelni: added blend
[14:40] Action: mateo` working on alphamerge
[14:41] <michaelni> maister, beastd did double blind listening tests with his friends to finetune the parameters
[14:42] <michaelni> maister, also you can chnage the lowpass point or use soxr (which is the SOX resampler which is also supported by swr)
[14:42] <beastd> hi maister, hi michaelni
[14:43] <michaelni> lowpass _cutoff_ point
[14:43] <michaelni> soxr uses a fft based resampler so its frequency response will be better
[14:43] <michaelni> if that translates to better for actual music with human ears i do not know
[14:44] <saste> framestep done, next is flite
[14:45] <maister> michaelni, I'll play around with cutoff point. Trying to compare lots of resamplers.
[14:46] <saste> we lost pos, right?
[14:46] <mateo`> vf_alphamerge done
[14:46] <ubitux> concat done
[14:47] <ubitux> it looks too easy to be true TBH
[14:47] <ubitux> i'm pretty sure i've messed some perms at some point
[14:47] <michaelni> maister, if you compare you probably want to try linear_interp, people could not hear a difference but it should show up in the graphs
[14:47] <Cigaes> It seems that audio transcoding is working in my tree too. Cool.
[14:47] <ubitux> hell yeah
[14:48] <michaelni> saste, we have pkt_pos and av_frame_... accessors
[14:48] <saste> michaelni, allright
[14:54] <michaelni> ubitux, merged
[14:55] <michaelni> saste, anything i should merge (git url ?)
[14:56] <saste> michaelni, i have two patches, i'm currently figuring out how to track your repo from my local git repo
[14:57] <Cigaes> What is the plan after this merge fest, once this is working?
[14:57] <saste> Cigaes, a party?
[14:57] <ubitux> profit
[14:57] <Cigaes> Merge it directly into the main branch, or do it by steps, reordering/squashing patches to get something clean?
[14:58] <ubitux> would be nice to merge it so master don't get too much delay
[14:58] <michaelni> iam happy with anything but slightly in favor of what is least work
[15:00] <michaelni> mateo`, merged
[15:03] <Cigaes> I will clean up my commits as much as possible. Please do not consider any of them worthy of inclusion yet.
[15:05] Action: ubitux working on vf mp
[15:06] <ubitux> this one might be a bit tricky..
[15:11] <Compn> vf mp whatsit ?
[15:11] <Compn> what are you doing ?
[15:11] <ubitux> we are merging the evil plan
[15:12] <Compn> oh
[15:12] <Compn> the buffer situation
[15:12] <ubitux> yes
[15:12] <Compn> sorry for interruption then :)
[15:15] <maister> michaelni, played around with params a bit, got something odd. I set filter_size, kaiser_beta and cutoff. Strangely enough, the impulse response "stops" before it's finished (256 tap). Couple of plots: http://imgur.com/754SrR7,v2R8dQB
[15:17] <ubitux> is this independance day covers?
[15:18] <michaelni> maister, look at build_filter() if you think theres a bug in it
[15:19] <maister> michaelni, ok
[15:19] <saste> michaelni, git at gitorious.org:~saste/ffmpeg/sastes-ffmpeg.git
[15:19] <saste> branch lavfi-merge-mess
[15:19] <michaelni> saste, i dont think i will have access to git@
[15:20] <michaelni> i mean it should be something with git:// or https://
[15:22] <michaelni> git://gitorious.org/~saste/ffmpeg/sastes-ffmpeg.git works
[15:22] <saste> michaelni, I think http:// also should have worked
[15:25] <michaelni> saste, rebased and pushed, there was another unrelated commit in there that i skiped
[15:26] <michaelni> Cigaes, there are alot of fate failures still with your repo
[15:27] <Cigaes> michaelni: certainly, yes. I have several bugs yet to clean.
[15:28] <michaelni> ok
[15:31] <saste> what's still missing (apart buffersrc and buffersink)?
[15:50] <michaelni> saste, dunno if anyone is working on but i still see a warning from libavfilter/af_astreamsync.c, 
[15:50] <michaelni> libavfilter/af_astreamsync.c:125:5: warning: passing argument 2 of ff_filter_frame from incompatible pointer type [enabled by default]
[15:50] <saste> michaelni, i'm on it
[15:52] <saste> what happens on min_perms?
[15:52] <saste> *of min_perms?
[15:52] <saste> are we still supposed to keep it?
[15:53] <michaelni> id say drop when possible
[15:56] <saste> michaelni: the problem is i don't know "when possible", still need to figure out how the new buffer reference system works
[15:58] <saste> michaelni, astreamsync done, pushed on the same repo
[15:58] <saste> also fixed the previous commit on flite, hope i didn't mess up
[16:04] <michaelni> saste, your tree contains 2 "libutvideo: Prioritize compatibility with new versions"
[16:07] <michaelni> saste, cherry picked the 2 commits and pushed
[16:14] <saste> michaelni, thanks, no idea whence that libutvideo commit comes
[16:15] <saste> ah i rebased against master at some point...
[16:21] <cone-370> ffmpeg.git 03Carl Eugen Hoyos 07master:881684e27cf6: Add some fate dependencies to fix make fate with --disable-avdevice.
[16:21] <cone-370> ffmpeg.git 03Carl Eugen Hoyos 07master:982070c113d7: Add some fate dependencies to fix make fate with --disable-everything.
[16:21] <Cigaes> I have found the origin of some of the FATE failures:
[16:21] <Cigaes> The commit 759001c on the rawvideo decoder is needed.
[16:22] <Cigaes> (Otherwise, it returns a pointer to the buffer of the packet, and the packet is freed a few moments afterwards.)
[16:22] <Cigaes> Shall I try to apply it to my tree?
[16:22] <saste> michaelni, about the opencl patch, you want to split lavu and lavfi parts, or prefer a single patch?
[16:24] <michaelni> Cigaes, whatever is needed to make it work
[16:24] <michaelni> saste, seperate might be better
[16:26] <Cigaes> Humpf, this is THE HUGE PATCH :(
[16:27] <ubitux> erm, decoders patch is required?
[16:27] <Cigaes> ubitux: for rawvideo, yes.
[16:28] <ubitux> can't you pick only a subset?
[16:28] <Cigaes> I suppose I have to do that.
[16:28] <ubitux> we'll need to note all that stuff in the squashed merge commit at the end
[16:30] <michaelni> yes, can someone write a commit message for that, that is start collecting what needs to be in it ?
[16:32] <ubitux> everything relevant to my work is written in my commit descriptions; we'll need to consider them when squashing
[16:32] <ubitux> i'll squash all my commits when we are done
[16:34] <Cigaes> Some of my commits messages are a bit terse, but that is because they are temporary.
[16:35] <michaelni> Cigaes, should i merge anything in my github/michael tree ?
[16:35] <Cigaes> Not yet.
[16:35] <Cigaes> Does anyone know how to cherry-pick only a file from a big commit?
[16:36] Action: ubitux uses a lame git show branch:file.c > file.c
[16:37] <ubitux> otherwise maybe git cherry-pick -e might do the trick
[16:39] <michaelni> http://stackoverflow.com/questions/5717026/how-to-git-cherry-pick-only-changes-to-certain-files
[16:41] <ubitux>         if (i == 1 || i == 2)
[16:41] <ubitux>             h = -((-h) >> desc->log2_chroma_h);
[16:42] <ubitux> why the double -?
[16:42] <ubitux> overflow trick or sth?
[16:42] <Cigaes> michaelni: thanks, that worked. But the changes to rawdec.c are not trivial :(
[16:43] <maister> ubitux, probably integer division that rounds up.
[16:43] <michaelni> Cigaes, maybe its easier to just change rawdec to copy like normal decoders until the next step is mergeed
[16:44] <ubitux> maister: ah..
[16:45] <maister> michaelni, been debugging some more. Seems like the longest impulse response you get is 256 samples, no matter the filter length (here, 470-ish).
[16:48] <maister> maybe some data is stuck in an internal FIFO somewhere.
[16:50] <maister> michaelni, nvm, that was it.
[16:57] <Cigaes> Hum. I have got an invalid free on yesterday's ffmpeg :(
[16:58] <Cigaes> Forget what I just said.
[16:58] Action: michaelni forgets
[17:04] <ubitux> users with ffmpeg 0.5-svn make me dream
[17:08] <BBB> if you guys are taking a break, maybe someone can review my patches? I'm about to start working on hpel after that
[17:09] <ubitux> well done BBB, you scared saste :(
[17:09] <BBB> hm...
[17:09] <BBB> I blame italian-quality internet
[17:14] <ubitux> i prefer to think it's your fault
[17:15] <cone-370> ffmpeg.git 03Giorgio Vazzana 07master:0286b4254097: lavd/v4l2: correctly handle error conditions in mmap_read_frame()
[17:15] <cone-370> ffmpeg.git 03Giorgio Vazzana 07master:9222978aed8b: lavd/v4l2: simplify code
[17:16] <michaelni> BBB ill look at the patches in a moment 
[17:19] <Cigaes> I have now only two tests failing (from the basic set without samples) in my tree.
[17:23] <ubitux> maybe problems with ports then
[17:23] <ubitux> Cigaes: which ones?
[17:24] <Cigaes> lavf-pixfmt (problem in rawdec: my merge was bogus)
[17:25] <Cigaes> acodec-roqaudio (not looked into it yet)
[17:26] <ubitux> oh, ok that's nice.
[17:26] <cone-370> ffmpeg.git 03Ronald S. Bultje 07master:9d40f7106dd5: dsputil: make square put/avg_pixels functions local to h264qpel.
[17:27] <ubitux> mp is starting to work... a little.
[17:29] <Cigaes>     if (desc->flags & (PIX_FMT_PAL || PIX_FMT_PSEUDOPAL)) {
[17:29] <Cigaes> Is it just me, or should it be | instead of || ?
[17:30] <michaelni> probably |
[17:31] <Cigaes> Well, there's a bug in libav that ffmpeg will not have. Removing the extra | makes fate passes for this one.
[17:31] <cone-370> ffmpeg.git 03Ronald S. Bultje 07master:22cc8a103c66: x86/qpel: move fullpel and l2 functions to separate file.
[17:38] <BBB> hm I see someone committed my unfinished atomics patch
[17:38] <BBB> fun
[17:38] <BBB> I guess I'll have to fix that up to actually do something useful now
[17:45] <Cigaes> ubitux: FATE fails on filter-metadata-silencedetect and filter-metadata-scenedetect. I believe you worked on this/these one?
[17:46] <ubitux> yeah well, i guess the metadata code is somehow broken
[17:46] <ubitux> i've not done anything outside the filters
[17:47] <ubitux> metadata are injected in the AVFrame, but i'm not sure of the code behind this
[17:47] <Cigaes> ubitux: ok.
[17:47] <Cigaes> Right now, I have 4 failing tests out of the full set enabled in my tree.
[17:48] <ubitux> that's nice :)
[17:48] <Cigaes> These two, sub2video (obviously my responsibility) and delogo that I am investigating right now.
[17:48] <ubitux> i'm still messing with mp, 'might take a while
[17:49] <ubitux> (which is not tested anymore in fate because vf pp was ported)
[17:52] <Cigaes> ubitux: adding a test again may be a good idea at some point. But not right now.
[17:54] <ubitux> yeah sure
[17:55] <Cigaes> The delogo failure is quite strange: it is not an invalid memory access, premature free, assert failure, etc, it is actually the result of the filter that changed slightly. I am baffled.
[17:56] <ubitux> nothing in valgrind?
[17:57] <Cigaes> No, I checked that first of course.
[17:57] <Cigaes> Well...
[17:58] <michaelni> ubitux, silencedetect output changed too
[17:58] <michaelni> before -> silence_start: -0.0146939
[17:58] <michaelni> now -> silence_start: 2.09147e+14
[17:58] <ubitux> erk
[17:58] <ubitux> ok will look
[17:59] <ubitux> michaelni: i can look right now if you want to continue my work on vf mp
[17:59] <ubitux> which you're certainly more familiar with
[17:59] <michaelni> well i can try
[17:59] <michaelni> but mp isnt all that important anymore
[18:00] <ubitux> there is just a leak left to fix
[18:00] <ubitux> "just" :)
[18:00] <ubitux> will push the WIP in a moment
[18:00] <michaelni> ubitux, i see lots of  NOPTS in ashowinfo
[18:01] <michaelni> this could be related to the silencedetect stuff dunno
[18:01] <ubitux> michaelni: 0c96de32b8a79a27c2790a6e34d5f83827c7ea9e in my branch
[18:01] <ubitux> ok
[18:01] <ubitux> going to have a look
[18:01] <ubitux> btw
[18:01] <ubitux> i may be blind but...
[18:01] <ubitux> 7194330bcd6db8dc5c22e0c162a0992d519307f9
[18:02] <ubitux> i've been watching this for 2 minutes
[18:02] <ubitux> oh
[18:02] <ubitux> sorry.
[18:02] <ubitux> forget this
[18:09] <ubitux> ok get it
[18:09] <ubitux> that's because i don't set the pts out of movie
[18:09] <ubitux> that was one of my comments indeed..
[18:13] <ubitux> we'll need some fate updates because of the pos too
[18:16] <ubitux> huh... metadata per frame-side-data
[18:22] <BBB> michaelni: don't forget the dsputil selectable patch (http://permalink.gmane.org/gmane.comp.video.ffmpeg.devel/160075 - I can send a new version if it doesn't apply cleanly anymore)
[18:28] <ubitux> ok, lavfi needs update as well
[18:29] <ubitux> Cigaes: so what's the proper way to deal with the buffersink?
[18:31] <ubitux> it seems the current issue is that the compat layer is not copying the frame metadata, so the frame metadata from AVFrame (in filters) to AVFilterBufferRef (in buffersink) don't get copied
[18:31] <ubitux> so ffprobe don't get them
[18:31] <wm4> ah, somehow I thought your vf_scale would handle colorspace and color levels correctly - but apparently it doesn't?
[18:31] <ubitux> one way is to make lavd/lavfi.c use the new API, the other is to update the compat layer
[18:43] <michaelni> ubitux, the problem you run into is that NULL is reinterpreted by the new buffer api as av_free() so you need to use an actual donothing function
[18:44] <michaelni> also i suepect to be fully correct the buffer needs to be completely copied
[18:44] <ubitux> oh, i see, thanks
[18:44] <ubitux> by the way, if we want to merge this commit soon, i think it's ok to disable the metadata tests
[18:44] <ubitux> we'll loose the feature temporarly
[18:45] <michaelni> yes iam ok with that 
[18:45] <ubitux> it might need some re-thinking of the feature, and the merge commit might not be the best place to do such changes
[18:45] <michaelni> what about the pts ?
[18:45] <ubitux> the fix is easy for this one
[18:45] <ubitux> http://pastie.org/6431769
[18:45] <ubitux> this does the trick
[18:47] <michaelni> fatal: http://pastie.org/6431769/info/refs not found: did you run git update-server-info on the server?
[18:47] <michaelni> for mp you need: static void dummy_free(void *opaque, uint8_t *data){}
[18:47] <michaelni> i can update your mp commit too if you prefer
[18:47] <ubitux> i'll commit properly :)
[18:48] <ubitux> whatever you prefer for mp
[18:48] <michaelni> lazy
[18:48] <ubitux> ok, i'm doing it
[18:48] <michaelni> thx :)
[18:48] <ubitux> i'll commit the pts fix in a moment
[18:50] <ubitux> michaelni: you can pick eefeb3ffce438fb9386ba29122a6bf72e6fa7287
[18:50] <ubitux> and 91f9c824a776f17d96d55ebc4c2132133376e115 eventually
[18:51] <ubitux> mmh 91f9c824a776f17d96d55ebc4c2132133376e115 depends on the work from Cigaes though
[18:52] <michaelni> both picked
[18:53] <ubitux> nice, the dummy free indeed improved things
[18:53] <ubitux> still get a leak though
[18:54] <michaelni> dont double ref
[18:54] <ubitux> i've removed it
[18:56] <ubitux> oh it's ok
[18:56] <ubitux> the leak is not in mp
[18:56] <ubitux> it's the compat layer, reproducible with the other filters as well
[18:56] <michaelni> silencedetect works now with your paztches :)
[18:57] <ubitux> without metadata, right?
[18:57] <michaelni> i get the same output with a test mp3 as with old ffmpeg
[18:58] <michaelni> that is silcence positions
[18:58] <ubitux> yeah right, but the fate test will still fail
[18:58] <ubitux> since metadata are not raised properly from lavfi device anymore
[18:59] <ubitux> michaelni: what did you mean by buffer fully copied in vf mp?
[19:00] <jojva> Anyone can explain to me how mpegts splits packets ? I need to get the non base-view packets for MVC. mpegts is doing a good job but it's not perfect.
[19:01] <michaelni> ubitux, like memcpy width*height
[19:04] <michaelni> also pkt_pos is lost somewhere
[19:04] <michaelni> and fixing av_frame_copy_props() isnt enough
[19:06] <ubitux> the AVBuffer is created with the stride data and linesize; isn't it enough?
[19:10] <michaelni> i think the avbuffer can live longer than the mpi source is guranteed to be valid
[19:10] <michaelni> but i could be wrong, its also not so important, vf_mp is going to be removed anyway sooner or later
[19:11] <ubitux> i don't know what to do, so feel free to pick and adjust :p
[19:11] <ubitux> Cigaes: what's left on your side?
[19:12] <Cigaes> ubitux: the sub2video failure is strange, but I leave it for now.
[19:12] <Cigaes> Right now I am cleaning-up my commits to make them mergeable.
[19:12] <wm4> hm, testing with Libav, frames returned by the decoder never seem to be writable... although this was possible with the old DR
[19:15] <michaelni> [ffplay_buffer @ 0x7fc6e031c900] Changing frame properties on the fly is not supported by all filters.
[19:15] <michaelni> *** glibc detected *** ./ffplay: invalid fastbin entry (free): 0x00007fc6e031de50 ***
[19:16] <ubitux> michaelni: i've tested with Cigaes' branch
[19:16] <ubitux> michaelni: btw, did you have a look to the revert of d7dcd96a in vf pad?
[19:16] <michaelni> Invalid read of size 8
[19:16] <michaelni> ==32590==    at 0xB6E70C: av_frame_move_ref (frame.c:311)
[19:16] <michaelni> ==32590==    by 0x482D0A: av_buffersrc_add_frame (buffersrc.c:130)
[19:16] <michaelni> ubitux, the revert is ok
[19:17] <michaelni> but i get crashes all over the place with cigaes branch
[19:17] <Cigaes> Right now, there is a design problem. We had av_buffersrc_add_frame(bufsrc, frame, flags); they added av_buffersrc_add_frame(bufsrc, frame) without the flags argument. But the flags argument is needed. How do we call the good function? Right now, I used av_buffersrc_add_frame_flags(). Any better idea?
[19:17] <Cigaes> michaelni: with ffmpeg, or with ffplay?
[19:18] <michaelni> both
[19:18] <michaelni> the exmaple above is ffplay only though
[19:18] <Cigaes> Can you give me a failing command line?
[19:23] <michaelni> Cigaes: ./ffmpeg -i matrixbench_mpeg2.mpg -t 1 test.avi ; ./ffplay test.avi
[19:23] <michaelni> Assertion frame->format == link->format failed at libavfilter/avfilter.c:744
[19:24] <michaelni> with threads 1 it ends with *** glibc detected *** ./ffplay: invalid fastbin entry (free): 0x00007f2ff80bfab0 ***
[19:25] <ubitux> michaelni: you might want to pick afb68bca01b9475f4e788d4f4a6ae5716f620b4c
[19:25] <ubitux> anyway, all my commits can be squashed into the merge commit, except the last two vf hue
[19:25] <ubitux> (direct path + reindent)
[19:26] <Cigaes> michaelni: crash reproduced.
[19:28] <michaelni> Cigaes, can i merge your changes?
[19:29] <ubitux> Cigaes: btw, i suspect the delogo changes are due to the switch to the direct path (or the other way around)
[19:29] <michaelni> its quite time consuming rebase back and forth every time i want to push a commit
[19:30] <Cigaes> michaelni: not yet, but hopefully soon.
[19:31] <michaelni> :(
[19:31] <Cigaes> Do you want me to work on cleaning up my commits or finding the cause of the crash first?
[19:31] <ubitux> is the cleanup important since everything is going to be squashed in the merge commit anyway?
[19:31] <michaelni> there are 100s crashes
[19:32] Action: ubitux looking at the delogo changes
[19:32] <Cigaes> michaelni: this particular crash seems to be related to a bufref inserted where a frame is expected.
[19:32] <Cigaes> I go back to the cleanup.
[19:33] <ubitux> make sure these kind of crashes are not fixed in a later commit
[19:33] Action: michaelni wants to work but cant because he cant push the changes
[19:33] <ubitux> :D
[19:33] <michaelni> because of cleanup :/
[19:37] <ubitux> mmh seems both head and the evil branch are using the direct mode for delogo, so that's something else
[19:44] <ubitux> the problem might be elsewhere... :p
[19:48] <michaelni> ubitux, do you have a vf_mp fix i can merge ?
[19:49] <michaelni> b467ff13dadc65dcfc0026594f4883d8f6ab0b2f ?
[19:50] <ubitux> a fix for what?
[19:50] <ubitux> this is my final commit yes
[19:50] <ubitux> but i haven't changed the memcpy you were asking for
[19:50] <michaelni> ok will merge
[19:50] <ubitux> this port works though
[19:50] <ubitux> you can take afb68bca01b9475f4e788d4f4a6ae5716f620b4c too
[19:50] <Cigaes> michaelni: running FATE before publishing my cleaned-up patches.
[19:51] <ubitux> michaelni: want a commit to disable metadata tests?
[19:52] Action: ubitux starting to write a description for the squashed merge
[19:52] <ubitux> i'll need a comment from you about the buffer sink/src Cigaes :p
[19:52] <michaelni> ubitux, git tells me that the afb68bca01b9475f4e788d4f4a6ae5716f620b4c commit is empty
[19:53] <michaelni> it already has been merged
[19:53] <ubitux> michaelni: ah, ok, sorry then :)
[19:53] <Cigaes> I believe you can merge the following patches from my tree:
[19:53] <Cigaes> 23c721e rawdec: use AVPALETTE_SIZE instead of magic constants.
[19:53] <Cigaes> 8acc118 rawdec: work with refcounted frames.
[19:53] <Cigaes> b455d13 lavfi/audio: add missing include.
[19:53] <Cigaes> b73b1f5 lavu/frame: copy the channels field.
[19:53] <Cigaes> 634cdc6 lavfi/audio: set the channels field on allocated buffers.
[19:53] <Cigaes> d1f9180 lavfi/buffersrc: port to AVFrame.
[19:53] <Cigaes> 9739d2a lavfi: remove obsolete functions and code.
[19:53] <Cigaes> eb2360a lavfi: disable buffersinks from the fork.
[19:53] <Cigaes> 13dbfd5 lavfi: move obsolete declarations to reduce diff with the fork.
[19:54] <Cigaes> 13e3fc5 lavfi/buffersink: upgrade to AVFrame.
[19:54] <Cigaes> That should be all there is in the branche, but better make sure I did not mess up with git.
[19:56] <michaelni> ubitux, we will have to either disable the metadata test or fix it
[19:57] <ubitux> yes, i'd suggest to disable them unless someone see an obvious fix
[19:57] <ubitux> but i think we'll have to mess a little with lavfi device
[19:57] <michaelni> ubitux, well theres no real hurry i cant commit the code to mainline as long as its crashing
[19:57] <ubitux> ok
[19:58] <michaelni> but disable is perfectly fine too
[19:58] <Cigaes> michaelni: we may want to let people test for a few days on this separate branch before merging to mainline anyway.
[19:58] <ubitux> since everything is blocked, i guess i'll look into the metadata
[19:58] <ubitux> Cigaes: maybe, but we should not stop picking from libav in this branch
[19:59] <ubitux> we've more to merge..
[20:00] <michaelni> Cigaes, cherry picked & pushed your commits
[20:00] <michaelni> ill be afk for 10min need to eat
[20:00] <Cigaes> Do not eat too fast!
[20:01] <ubitux> come eat behind the PC, you'll waste less time
[20:01] <ubitux> s/behind/in front of/
[20:02] <Cigaes> michaelni/ubitux: revert b0555d61ec827f560440d191630de2aaaf8b65c0. It is the one causing crashed in ffplay (at least the reported crash).
[20:02] <ubitux> oh ok
[20:02] <ubitux> it was a temporary hack to make it build
[20:02] <ubitux> sorry about that
[20:03] <Cigaes> ubitux: eating can be done in front of the PC, but not preparing.
[20:03] <ubitux> that's because you're not skilled enough
[20:04] <Cigaes> I do not think even an AMD CPU is powerful enough to cook something.
[20:05] <thegeek_> cooking on cpus is a well established art
[20:14] <michaelni> Cigaes, ./ffmpeg -i matrixbench_mpeg2.mpg -t 1 test.y4m ; ./ffmpeg -i  test.y4m test2.y4m <-- also segfaults
[20:15] <michaelni> looks like near null ptr in rawdec
[20:16] <ubitux> Cigaes: in c204eb08b345d6577ea75f75b0169332a16d5142, you may have dropped the metadata copy
[20:16] <michaelni> pushed revert of b0555d61ec827f560440d191630de2aaaf8b65c0
[20:16] <michaelni> and can confirm that fixes the ffplay crash
[20:19] <Cigaes> michaelni: confirmed: this is interlacing stuff that I badly merged with Anton's changes.
[20:22] <Cigaes> michaelni: 50f765e from my tree may fix the crash with y4m, but I do not know interlacing well enough to be sure it is correct.
[20:24] <Cigaes> ubitux: you are right. I am adding it back and testing. pos is in the same situation.
[20:26] <ubitux> pos is a bit complicated
[20:26] <ubitux> it's dropped from AVFrame afaik.
[20:27] <michaelni> Cigaes, another failure: ./ffmpeg -f lavfi -i "amovie=fate-suite/wavpack/num_channels/eva_2.22_6.1_16bit-partial.wv,asplit=3[out1][a][b]; [a]showwaves=s=340x240,pad=iw:ih*2[waves]; [b]showspectrum=s=340x240[spectrum]; [waves][spectrum] overlay=0:h [out0]"   out.avi
[20:27] <michaelni> Assertion ref->buf->refcount > 0 failed at libavfilter/buffer.c:99
[20:27] <Cigaes> f1d57c6 from my tree fixes the metadata (and pos) problem. Apparently metadata-scenedetect and metadata-silencedetect now work, I do not know if it is related.
[20:27] <ubitux> great
[20:27] <Cigaes> michaelni: there is something fishy with overlay, I am looking into it in the short time I have left.
[20:27] <ubitux> i'll share the merge commit description in a few minutes
[20:28] <ubitux> you'll be welcome to complete it
[20:28] <michaelni> Cigaes, i still have many crashes with files related to security fixes but i belive iam not allowed to share them
[20:31] <michaelni> fate-filter-delogo still fails
[20:32] <ubitux> this one might be a problem in the decoders
[20:32] <ubitux> or not at filter level at least
[20:33] <Cigaes> michaelni: I just fixed the one you reported with eva_2.22. There was a big FIXME in my commit. 
[20:34] <Cigaes> 6eaa21b lavfi/sink_buffer: fix unrefing of remaining frames.
[20:39] <michaelni> Cigaes, ./ffmpeg -i matrixbench_mpeg2.mpg -deinterlace test.avi crahses too
[20:40] <michaelni> also cherry picked and pushed your fixes
[20:41] <ubitux> http://pastie.org/private/wvosnzgzcbdaw0thzlpxpw  adjustements welcome
[20:41] Action: ubitux going to |sort a bit
[20:42] <ubitux> http://pastie.org/private/idh5sqd2eehlndigzoyta
[20:42] <ubitux> a bit better
[20:43] <Cigaes> michaelni: this one does not seem related to parts of the code I know.
[20:44] <michaelni> Input stream #0:0 frame changed from rate:48000 fmt:fltp ch:6 chl:5.1(side) to rate:48000 fmt:fltp ch:6 chl:6 channels (FC)
[20:44] <michaelni> Assertion frame->channels == link->channels failed at libavfilter/avfilter.c:750
[20:44] <michaelni> Aborted (core dumped)
[20:44] <Cigaes> michaelni: the -deinterlace crash happens with HEAD avconv too. They broke it.
[20:45] <Cigaes> michaelni: do you have a command line for this one?
[20:45] <michaelni> its just decode file to -f null -
[20:45] <michaelni> iam not sure where i get the file from so i am not sure aboutsharing i need to check
[20:53] <michaelni> Cigaes, i asked if iam allowed to share the files, but AFAIK the awnser was no 1-2 years ago
[20:54] <michaelni> also matrixbench_mpeg2.mpg -vframes 2 -aframes 2 -f framecrc - 
[20:54] <michaelni> produces different output 
[20:54] <michaelni> 2 frames less
[20:59] <Cigaes> michaelni: is it the same file that you wrote me about two weeks ago?
[20:59] <michaelni> probably not
[21:01] <Cigaes> Regarding the -vframes 2 -aframes 2: all [av]frames stuff is fragile. Changes are to be expected.
[21:01] <Cigaes> IMHO, -[av]frames should only be used when there is only a single stream in the file.
[21:02] <Cigaes> Regarding the file you are not allowed to share:
[21:02] <Cigaes> "ch:6 chl:6 channels (FC)"
[21:03] <Cigaes> looks to me like a bug in the decoder.
[21:03] <ubitux> can i help in anything?
[21:03] <wm4> you should add the header libavutil/channel_layout.h
[21:03] <wm4> it's the same as audioconvert.h (I think) except it's not deprecated in l.
[21:04] <michaelni> its very possible the decoder is buggy but it shouldnt cause a crash/abort
[21:04] <ubitux> wm4: i have it locally here
[21:04] <ubitux> but i was asking for TEP merge
[21:04] <ubitux> michaelni: any bug i could look for?
[21:09] <michaelni> with: http://www.ffmpeg.org/~michael/restest.h264 i get Assertion frame->format == link->format failed at libavfilter/avfilter.c:744 using:
[21:09] <michaelni> ./ffmpeg  -i ~/videos/restest.h264 test.mpg test2.mpg
[21:11] <michaelni> Cigaes, see above for another bug (or ubitux but i think its more Cigaes area)
[21:11] <michaelni> ill test more
[21:11] <ubitux> i can work on whatever you want to accelerate that merge
[21:13] <michaelni> Cigaes, "-i ~/videos/matrixbench_mpeg2.mpg -vframes 3 -aframes 2  -f framecrc -" is much slower than before
[21:13] <michaelni> it seems decoding the whole file
[21:14] <michaelni> ubitux, you can pick any of the issues above and work on it if you like or just test a few random files maybe you find issues or crashes yourself
[21:14] <ubitux> ok, gonna run some tests myself
[21:18] <michaelni> ubitux, maybe try fuzzing files, the crashes i see are generally not from valid files
[21:20] <Cigaes> michaelni: the problem with two outputs should be fixed by the following commit:
[21:20] <Cigaes> dfe7cf7 ffmpeg: do not give away our reference to the frame.
[21:21] <Cigaes> I suspect the problem with -vframes 3 -aframes 2 will be fixed when AV_BUFFERSRC_FLAG_PUSH is restored in ffmpeg.c.
[21:22] <michaelni> Cigaes, thanks alot for dfe7cf7, i confirm it fixes it
[21:29] <michaelni> Cigaes, ubitux  there are still 2 fate failures left (sub2video & delogo)
[21:29] <ubitux> ok, i guess i'll go back on delogo
[21:31] <Cigaes> I will have to stop soon, probably more or less for tonight. I have the failure of sub2video in mind, but I do not understand it yet.
[21:32] <michaelni> Cigaes, ok, you have time again tomorrow ?
[21:36] <michaelni> ill look into the crashes as i cant share the files that cause them ...
[21:36] <Cigaes> I will have time tomorrow, yes.
[21:36] <michaelni> great, thanks
[21:38] <ubitux> i wonder if the delogo changes are not because of overread in the padding
[21:38] <michaelni> ubitux, could be anything 
[21:38] <ubitux> it's not the decoding at least
[21:39] <michaelni> i think its not happening of the source is converted to lets say huffyuv IIRC
[21:40] <michaelni> could also be changes to the initial buffer content or something maybe if any of it leaks through
[21:42] <ubitux> btw there is a memleak
[21:44] <ubitux> mmh my bad
[21:44] <ubitux> it's just the opengl init..
[21:48] <ubitux> huh still a sscanf in delogo
[21:49] <michaelni> fixed first crash
[22:13] <michaelni> ubitux, Cigaes i think the crashes are fixed, but ill do more testing
[22:14] <ubitux> nice
[22:14] <ubitux> OTOH i still have no clue about delogo
[22:14] <ubitux> linesizes and w/h are identical within the filter
[22:15] <ubitux> between* the filters*
[22:15] <ubitux> same cmd line replacing delogo with vf copy produces the same output
[22:17] <ubitux> and of course still no vf diff when we need it
[22:17] <ubitux> i guess relying on luca was a bad idea
[22:18] <ubitux> oh but we have the blend filter!
[22:18] <michaelni> gimp can be used too with 2 layers
[22:18] <michaelni> but its work
[22:20] <ubitux> hehe that's interesting
[22:21] <ubitux> http://lucy.pkh.me/diff.webm
[22:21] <ubitux> some difference shows up on the top left, and around the delogo zone
[22:22] <ubitux> though the video doesn't show the top left messed up
[22:24] <michaelni> ubitux, delogo passes if you force delogo not to use direct=1
[22:25] <ubitux> i thought it was related to this
[22:26] <ubitux> but both master and evil branch had direct=1
[22:27] <ubitux> but indeed..
[22:29] <michaelni> ubitux, the premissions are wrong or something
[22:29] <ubitux> oh wait
[22:29] <ubitux> yeah
[22:29] <michaelni> at the end: if(av_frame_is_writable(in))
[22:29] <michaelni>         memset(in->data[0], 0, 50*in->linesize[0]); results in artifacts
[22:29] <michaelni> aboce with direct forced to 0
[22:41] <ubitux> the frame is not writable in libav
[22:44] <michaelni> ubitux, http://pastebin.com/iF2YnFbS
[22:44] <michaelni> does that look correct to you ?
[22:45] <ubitux> mmh yes indeed
[22:47] <ubitux> but that doesn't seem to be the case in libav
[22:47] <ubitux> looks like it's done differently somewhere
[22:47] <ubitux> maybe the tools force RO or something
[22:49] <ubitux>     return avpriv_atomic_int_add_and_fetch(&buf->buffer->refcount, 0) == 1;
[22:49] <ubitux> erm
[22:50] Action: ubitux wonders why not avpriv_atomic_int_get()
[23:00] <wm4> ubitux: it follows other naming conventions, I think
[23:00] <wm4> like gcc's builtins and whatnot
[23:01] <wm4> or wait
[23:01] <wm4> you mean avpriv_atomic_int_get actually exists?
[23:01] <wm4> durp it does
[23:01] <wm4> where's the undo function in this IRC program
[23:02] <ubitux> ask everyone to /clear
[23:02] <ubitux> sometimes it work
[23:02] <ubitux> but well i guess this code was written before avpriv_atomic_int_get() was introduced
[23:02] <iive> wm4: i think it was recent commit by BBB
[23:15] <ubitux> michaelni: in libav, each buffer has a refcount of 3
[23:15] <ubitux> while we have a refcount of 1
[23:15] <ubitux> (all the frame->buf[0..2])
[23:16] <ubitux> having only one refcount triggers the writable case
[23:16] <ubitux> so it seems we might be missing a ref somewhere
[23:16] <ubitux> (3 looks a bit strange though)
[23:17] <michaelni> ubitux, i intend to push my fix for delogo, the code doesnt look right as is
[23:17] <wm4> ubitux: wait what, frame->buf[n] all point to the same buffer?
[23:18] Action: wm4 shuts up
[23:18] <ubitux> nope
[23:19] <ubitux> that was a good question actually
[23:19] <ubitux> but no, different pointers in both projects
[23:19] <michaelni> ubitux, showspectrum starts with a green background 
[23:20] <ubitux>         memset(outpicref->data[0], 0, outlink->h * outpicref->linesize[0]);
[23:20] <ubitux> yes :p
[23:21] <michaelni> -f lavfi -i "amovie=/home/michael/videos/matrixbench_mpeg2.mpg,asplit=2[out1][b]; [b]showspectrum=s=340x240[out0]" -vcodec huffyuv test.avi
[23:21] <michaelni> actually uninitialized id say
[23:21] <ubitux> mmh that's because divVerent changed it to yuv
[23:21] <michaelni> it was black before
[23:21] <ubitux> it was rgb before
[23:22] <ubitux> so the memset was correct
[23:22] <michaelni> its not solid green
[23:22] <michaelni> its uninitialized
[23:22] <michaelni> ill guess its not a new bug though
[23:22] <ubitux> yes, the memset is not correct anymore
[23:23] <michaelni> maybe we just initialized the buffers before
[23:23] <michaelni> somewhere else
[23:23] <ubitux> the memset was correct with rgb24
[23:23] <ubitux> put passing it in yuv + planar broke two things
[23:24] <ubitux> first the memset data[0] only init luma
[23:24] <ubitux> and secondly, it inits it in green
[23:24] <ubitux> (chroma remains uninit)
[23:24] <michaelni> no need to explain just push a fix :)
[23:24] <ubitux> master?
[23:24] <michaelni> probably
[23:38] <michaelni> ubitux, saste how should we stash the commits together for pushing to mainline ?
[23:38] <saste> michaelni, how can i help with the evil merge?
[23:38] <saste> michaelni, you are the merge expert ;-)
[23:39] <saste> I suppose something which doesn't break git bisect
[23:39] <ubitux> michaelni: did you see my pastebin?
[23:39] <ubitux> i think you can squash everything in the merge commit
[23:39] <ubitux> (except the two last hue patches)
[23:40] <ubitux> http://pastie.org/private/idh5sqd2eehlndigzoyta  and add this in the description of the commit
[23:40] <ubitux> any improvements to the text welcome, that's just a pattern
[23:40] <ubitux> and i may have made mistakes
[23:40] <michaelni> ok, ill do a bit more testing and bugfixing maybe and then probably push it if its bug count is negligible
[23:41] <wm4> wow that sure was fast
[23:41] <michaelni> ill probably have to disable sub2video
[23:41] <michaelni> fate-sub2video that is
[23:41] <ubitux> michaelni: that's fun i can reproduce the invalid reads only with the merge branch
[23:42] <ubitux> it looks like the buffer was previously initialized at some point
[23:42] <michaelni> wm4, its just the lavfi part of the evil plan, lavc is then stil left
[23:42] <wm4> oh :(
[23:43] <saste> that's just the start of the fun
[23:43] <ubitux> the good news is that lavfi had huge differences
[23:43] <ubitux> in comparison to lavc, IMO
[23:43] <ubitux> OTOH, we may have some trouble with pp with lavc i guess
[23:43] <wm4> you mean vf_pp?
[23:43] <ubitux> (or maybe not?)
[23:44] <ubitux> no i don't know, actually it might be ok
[23:44] <michaelni> iam not sure how well pp worked, the qp passing code was not very well tested
[23:44] <ubitux> yes :p
[23:44] <saste> if fate already passing?
[23:44] <michaelni> though in projects like mplayer pp worked long ago at least
[23:44] <ubitux> saste: except sub2video
[23:45] <saste> awesome
[23:45] <wm4> AFAIK l. stopped setting qscale_table
[23:46] <ubitux> at least the field exists
[23:47] <wm4> I suspect the reason might be that you'd need to ref-count the qscale data too?
[23:47] <wm4> which is annoying, slow, needs memory etc.
[23:48] <ubitux> michaelni: 
[23:48] <ubitux> -        memset(outpicref->data[0], 0, outlink->h * outpicref->linesize[0]);
[23:48] <ubitux> +        memset(outpicref->data[0],   0, outlink->h * outpicref->linesize[0]);
[23:48] <ubitux> +        memset(outpicref->data[1], 128, outlink->h * outpicref->linesize[1]);
[23:48] <ubitux> +        memset(outpicref->data[2], 128, outlink->h * outpicref->linesize[2]);
[23:48] <ubitux> does this help?
[23:48] <ubitux> (looks like to here)
[23:49] <michaelni> ubitux, should be ok (didnt try but i dont think this needs to be tested by 2 people ...)
[23:49] <ubitux> i'm not sure of the values ;)
[23:49] <ubitux> i get black and valgrind is happy so i guess that's ok
[23:50] <michaelni> yes 0,128 is good enough
[23:50] <BBB> iive: ?
[23:53] <cone-370> ffmpeg.git 03Clément BSsch 07master:ee0a8bcba145: lavfi/showspectrum: fix outpicref initialization.
[23:55] <ubitux> michaelni: so basically delogo has an issue we can now spot because direct output works, and libav didn't notice it because they have a bunch of extra frame copy?
[23:55] <michaelni> probably
[23:55] <ubitux> that's fun
[23:56] <ubitux> michaelni: you're fixing it?
[23:56] <michaelni> ive 2 bugfixes locally that i need to push but i want to test a bit more
[23:56] <michaelni> delogo is fixed by the first
[23:59] <michaelni> ubitux, saste doc/examples/filtering_audio.c needs a update 
[23:59] <michaelni> doc/examples/filtering_audio.c:210:17: error: too many arguments to function av_buffersrc_add_frame
[00:00] --- Sun Mar 10 2013