[Ffmpeg-devel-irc] ffmpeg-devel.log.20130413

[burek]

[00:00] <cone-607> ffmpeg.git 03highgod0401 07master:8f99494c1ca5: avutil/opencl: merge opencl env to opencl context
[00:01] <saste> thanks
[00:28] <ubitux> durandal11707: i've just tried to de-telecined a telecined material with -vf telecine and it seemed to work fine, without the need for any settings
[00:29] <ubitux> saste: do you want to have a look to the ivtc filters before i push?
[00:29] <durandal11707> ubitux: change default pattern (23)
[00:30] <saste> ubitux, if you see no reply from me in a day you can push it safely
[00:30] <ubitux> ok
[00:30] <ubitux> saste: i can wait a little more if your priority is to deal with the graph parse issue ;)
[00:30] <ubitux> i'm not particularly in a hurry
[00:30] <ubitux> i'd like some desperate comments on the ass/ssa stuff though
[00:31] <saste> ubitux, not my area, maybe wait for nicolas?
[00:31] <ubitux> yeah i know :)
[00:31] <saste> which is the name of the patch anyway?
[00:32] <ubitux> "[PATCH] pain in the ASS" or something like that
[00:32] <ubitux> "subtitles: introduce ASS codec id and use it."
[00:33] <ubitux> btw, it seems our mkv demuxer doesn't deal that well with the reference mkv files
[00:42] <Compn> aurel was the ass and mkv maintainer :)
[00:42] <Compn> dont let it drive you insane ubitux!!!
[00:42] <durandal11707> so how that color issue in #2465 gonna be fixed?
[00:42] <durandal11707> ubitux: what reference files?
[00:43] <Compn> files on matroska.org ?
[00:44] <ubitux> durandal11707: test7.mkv to be specific
[00:44] <ubitux> from the official test suite from matroska.org indeed
[00:44] <durandal11707> what it have as codec(s)?
[00:44] <Compn> ubitux : got url is what we're asking for :)
[00:44] <ubitux> i don't remember but that's not the issue
[00:45] <durandal11707> l have it too?
[00:46] <Compn> http://www.matroska.org/downloads/test_w1.html
[00:46] <ubitux> durandal11707: it's not a regression or anything afaik
[00:46] <Compn> damaged mkv file
[00:46] <Compn> :P
[00:46] <Compn> is test7
[00:46] <durandal11707> huh, is this site nonsense, or my opera got hacked?
[00:47] <ubitux> https://sourceforge.net/projects/matroska/files/test_files/matroska_test_w1_1.zip/download
[00:47] <Compn> durandal11707 : whats it look like (i'm also using opera)
[00:47] <Compn> just looks like a wiki to me
[00:47] <cone-607> ffmpeg.git 03Clément BSsch 07master:dcea58502c9b: lavfi: drop now unused shorthand field from AVFilter.
[00:47] <durandal11707> Compn: stupid cursor when you hever over some underlined stuff
[00:48] <Compn> oh the double underlined links ?
[00:48] <durandal11707> this is 176.mb
[00:48] <Compn> thats some ad revenue generating script on the website. but no idea if they put it there or someone added it to the wiki :P
[00:48] <durandal11707> :)))
[00:50] <ubitux> so, when is anton going to send us a new storm? i'm getting bored
[00:51] <durandal11707> i searched for that file and found #545 bug report
[00:51] <saste> ubitux, you could implement filter reconfiguration, and then fix it when anton reimplement it differently
[00:51] <saste> or we can just wait
[00:51] <saste> i plan to work on 0mq, but i should fix sox (and splice) before
[00:52] <ubitux> it's funny how we are able to add so much filters given the constraints of lavfi, while l is doing the opposite trying to rethink the whole framework without having any testing filters yet
[00:53] <durandal11707> something like tile, but takes multiple inputs (better that StackVertical/Horizontal from vapour)
[00:53] <saste> worse is better
[00:54] <saste> then we should tackle scripting and high level binding
[00:55] <Compn> someone requested an audio filter to get rid of audio pops and glitches
[00:55] <Compn> like from vinyl album recordings
[00:55] <Compn> need to be de-popped
[00:55] <durandal11707> but stackh/v could be done with splice+tile, no?
[00:57] <durandal11707> Compn: it should be possible, just port relevant sox filters
[00:57] <durandal11707> the only thing i'm not sure is possible, but is required is writing to file
[00:58] <durandal11707> so filter would scan audio and output data ....
[00:58] <durandal11707> A->?
[00:58] <durandal11707> than another filter takes data and audio and produce de-popped audio: A?->A
[00:59] <durandal11707> i don thing how good/useful/reliable would be to make it single pass: A->A
[01:27] <Compn> i dont think theres any actual technical reason we cant do 2pass video encoding in one pass
[01:27] <Compn> just that no one does it
[01:27] <Compn> like reverse playback. its possible , but one of those things 
[01:33] <BBB-> Compn: do 2 pass encoding in 1 pass?
[01:33] <BBB-> what does that even mean
[01:34] <BBB-> you mean ffmpeg -i file -vcodec mpeg4 -pass 1 and 2 at the same time out.file ?
[02:40] <Compn> BBB- : yep
[02:40] <Compn> just needs more buffer or something
[02:42] <BBB-> just seek back after pass1
[10:10] <cone-7> ffmpeg.git 03Janne Grunau 07release/1.2:f0dc5c04194f: vaapi: fix argument for ff_vaapi_common_end_frame call (cherry picked from commit b066d90211072c7532e17c0c54d8475f10fc97ad)
[10:36] <kurosu> I'm about to add a test for aac; I can send a patch for the makefile part, but what about the 3MB worth of data files?
[10:56] <kurosu> michaelni: regarding r5q = countq, I guess I need to allocate  1 more reg for UNIX64 and set NOISE_TABLE to r6q then ?
[10:56] <kurosu> I can't test that
[11:05] <kurosu> bah, I lost my current tree somewhere between a rebase and a cherry pick... sigh :/
[11:26] <michaelni> kurosu, if you want to add a fate test just post a link or upload to the files for fate to our ftp, also smaller files are preferred if possible
[11:27] <michaelni> about unix/pic/x86_64, yes another register seems the obvious solution
[11:28] <kurosu> michaelni, ok, but there are issues (license, big difference between conformant output and current output) so I mailed alex converse for clarifications
[11:28] <kurosu> so I'll postpone that part
[11:28] <michaelni> ok
[11:29] <kurosu> sorry for the fumbling, I'm not very proficient at git, and managing reviews on every side and merging the result is not easy for me
[11:29] <kurosu> regarding ac3, I prefer not being in charge of anything, my time being limited - it just happens that I have 2-3 weeks where I can have time off / enough motivation now
[11:31] <michaelni> well, you can just review the ac3 stuff as long as you have interrest/motivation 
[11:39] <burek> we discussed some time ago something regarding yuv444p in libx264, which confuses people, since most players expect yuv420p, and we concluded that if the output is yuv444p then the logical behavior would be to just copy that to output, but this: http://pastebin.com/bWbWgurP
[11:40] <burek> shows that the input is bgr0 and the output is still yuv444p (default i guess), when in fact it should (imho) be yuv420p
[11:40] <nevcairiel> why
[11:40] <nevcairiel> bgr is a 4:4:4 format
[11:40] <nevcairiel> it should use the format as close to the original as possible
[11:40] <burek> oh that's the idea.. ok
[12:03] <ubitux> < kurosu> bah, I lost my current tree somewhere between a rebase and a cherry pick... sigh :/ // git reflog can save you
[12:38] <durandal_1707> saste: why i cant limit fps for sdl muxer?
[12:39] <saste> durandal_1707, don't know
[12:58] <kurosu> ubitux, thanks I was indeed pointed to that solution
[13:36] <cehoyos> durandal_1707: Do you think that there really is a xv driver that supports anyhting else than yuv420p ?
[14:12] <saste> durandal_1707, why telecine is documented amongst audio filters?
[14:14] <cone-86> ffmpeg.git 03Stefano Sabatini 07master:87dd62e14198: lavfi/testsrc: make use of AV_OPT_TYPE_DURATION
[14:46] <burek> does anyone know where can i ask questions about strange diff output? :)
[14:46] <burek> is there an official website for that tool or something
[14:48] <ubitux> man diff
[14:48] <ubitux> ’ GNU diffutils home page: <http://www.gnu.org/software/diffutils/>
[14:49] <ubitux> what's the "strange diff output"?
[14:49] <burek> it creates such a diff where it marks all the lines with - and then marks new changes with + (for the whole file) when there are just 2 new lines added..
[14:49] <burek> http://ffmpeg.gusari.org/paste.txt
[14:50] <burek> i'll try with that url, thanks
[14:50] <iive> burek: whitespaces ?
[14:50] <burek> -b?
[14:51] <iive> more like -w
[14:51] <nevcairiel> sounds like EOL differences
[14:51] <iive> e.g. it could be difference in cr/lf
[14:52] <burek> oh, i see
[14:52] <ubitux> tab vs spaces
[14:52] <ubitux> maybe :)
[14:52] <burek> yes yes i got it, thanks a lot for that tip :) i'll check it :)
[14:59] <burek> thanks guys :) it was windows/unix crlf issue :)
[15:08] <cone-86> ffmpeg.git 03Michael Niedermayer 07master:9b672d4017bb: vf_mp: mp buffers are not compatible with the reference count system
[15:54] <nevcairiel> really, how many problsm can be in one 10 line function
[16:09] <cone-86> ffmpeg.git 03Hendrik Leppkes 07master:cac309611cf3: lavf/mxf: fix parsing of the month from mxf timestamps
[16:10] <mateo`> michaelni: i didn't saw your reply :( obviously the bit masks are wrong :(
[16:11] <mateo`> forgot an extra F :(
[16:12] <michaelni> mateo`, want to send a patch ?
[16:13] <ubitux> why the 0xf are wrong?
[16:14] <mateo`> it does not cover 8bits
[16:14] <ubitux> oh, 8 bits, ok
[16:14] <ubitux> year is on 16 bits?
[16:15] <ubitux> mateo`: you could use an union btw
[16:15] <michaelni> other software had a year 2000 bug, we have a 16th hour and 16th minute bug
[16:17] <mateo`> ubitux: i used mxfenc as reference, i better check the specs
[16:17] <mateo`> sorry about this mistake guys
[16:24] <nevcairiel> i noticed the 0xf stuff only after i sent the patches, or i would've fixed it directly as well
[16:24] <nevcairiel> i only looked at the month line, where it worked =p
[16:26] <nevcairiel> anyhow, the other patches are more important, they actually fix msvc, this one i just found weird :p
[16:32] <mateo`> nevcairiel: shouldn't we issue a warning if month and day are not valid ?
[16:33] <nevcairiel> you tell me how the mxf spec describes those values
[16:34] <nevcairiel> the file in fate has one date with the timestamp 0
[16:35] <nevcairiel> and month -1 or day 0 cause errors in strftime
[16:40] <mateo`> from S377M: mxf timestamp is year (16), month (8), day (8), hour (8), min (8), sec (8), msec (8). A value of 0 for each field means timestamp 'unknown'
[16:46] <nevcairiel> so what do you do if the month is "unknown"? :)
[16:46] <nevcairiel> only print a time?
[16:47] <BBB-> kurosu: hey you're online
[16:50] <mateo`> nevcairiel: from what i understand, "unknown" is the value of the time if all field are set to 0
[16:51] <mateo`> in theory month/day/hour/... fields should be checked ...
[16:52] <nevcairiel> zero hours/minutes is at least valid
[16:53] <mateo`> i was thinking about values above 60 for these fields
[16:53] <nevcairiel> oh right
[16:53] <nevcairiel> not a problem right now
[16:53] <nevcairiel> :P
[16:54] <mateo`> sure ...
[16:54] <nevcairiel> cant even reach 16 :)
[16:57] <cone-86> ffmpeg.git 03Hendrik Leppkes 07master:8da2a3aba3a7: lavf/mxfdec: use more widely supported time format string
[17:02] <mateo`> nevcairiel: i'll double check the ts values with other mxf tools
[17:07] <kurosu> BBB-: yeah, been so for a week
[17:07] <kurosu> trying to optimize my still inefficient patch submission
[17:17] <mateo`> nevcairiel: maybe year should be checked as well ?
[17:17] <nevcairiel> -1900 is at least valid for msvcs strftime
[17:18] <nevcairiel> there is a upper limit of something like 8900
[17:18] <nevcairiel> not sure how much checking is needed there
[17:19] <nevcairiel> i guess the limit means years 0 to 10000 are supported
[17:19] <nevcairiel> or something
[17:24] <mateo`> what is the behaviour of strftime with invalid values in msvcs ?
[17:25] <nevcairiel> it looks like its trying to call some error handler, but for some reason that causes it to crash
[17:25] <mateo`> doh !
[17:25] <nevcairiel> may only be crashing on my system because i enforce no-execute on all apps
[17:25] <nevcairiel> which isnt the default
[17:26] <nevcairiel> and the error lists that as the reason
[17:30] <RobertNagy> why does video_get_buffer use pic->format to get the AVPixFmtDescriptor,  and in all other places use s->pix_fmt?
[17:30] <RobertNagy> seems inconsistent...
[17:32] <mateo`> nevcairiel: i guess the patch is ok then
[17:37] <mateo`> nevcairiel: isn't the mday check superflous since the value (timestamp >> 32 & 0xFF) is of type uint64_t ?
[17:37] <cone-86> ffmpeg.git 03Matthieu Bouron 07master:c8b364449424: lavf/mxf: fix parsing of timestamps
[17:37] <nevcairiel> but a day of 0 is not valid
[17:37] <nevcairiel> month starts at 1
[17:38] <mateo`> nevcairiel: i confused month/day ...
[18:12] <cone-86> ffmpeg.git 03Hendrik Leppkes 07master:423089e96441: lavf/mxfdec: validate month/day before date formatting
[18:13] <nevcairiel> yay that should make fate happy again
[18:21] <ubitux> possibly not only fate
[19:58] <cone-86> ffmpeg.git 03Clément BSsch 07master:73180f5b0efd: lavfi: flag more dynamic i/o filters.
[19:58] <cone-86> ffmpeg.git 03Clément BSsch 07master:2b44195fc9db: cmdutils: print if filters are sources or sinks.
[20:09] <cone-86> ffmpeg.git 03Clément BSsch 07master:8137ea358c71: doc/filters: use common options introduction string.
[20:19] <cone-86> ffmpeg.git 03Michael Niedermayer 07master:ded2f187a272: avcodec: fix ignorecrop AVOption offset
[20:23] <ubitux> michaelni: seems there are more in v4l2, drawtext, and fps
[20:23] <ubitux> michaelni: is it fixing a bug?
[20:28] <michaelni> no, not that i know
[20:29] <michaelni> also having the offsets equal to the parent field of a constant isnt that wrong having the offset being equal to a random value is wrong
[20:31] <ubitux> sure
[20:31] <ubitux> ok
[20:34] <cone-86> ffmpeg.git 03Michael Niedermayer 07master:536baf6cc5d1: oggenc: Fix pref_duration AVOption type
[20:44] <ubitux> michaelni: do we have a fate instance with assert level >= 2?
[20:45] <michaelni> good question
[20:47] <cone-86> ffmpeg.git 03Nicolas George 07master:6fbb21d6858b: lavfi/buffersink: factor checks for lists sizes.
[20:47] <cone-86> ffmpeg.git 03Nicolas George 07master:76c8060654f0: lavu: add av_pure to av_int_list_length_for_size.
[20:47] <cone-86> ffmpeg.git 03Nicolas George 07master:2a1d7ea5f8ba: lavu: add parens to macro argument.
[20:47] <cone-86> ffmpeg.git 03Nicolas George 07master:9e8e03de38b3: lavu/opt: check int lists length for overflow.
[20:47] <cone-86> ffmpeg.git 03Michael Niedermayer 07master:38e66af43b06: Merge remote-tracking branch 'cigaes/master'
[21:24] <cone-86> ffmpeg.git 03James Almer 07master:572e38a51313: log: Remove redundant va_list creation from missing_feature_sample()
[23:53] <cone-86> ffmpeg.git 03Clément BSsch 07master:28ac8426e646: lavu/dict: cosmetic realign.
[00:00] --- Sun Apr 14 2013