[Ffmpeg-devel-irc] ffmpeg-devel.log.20130908

burek burek021 at gmail.com
Mon Sep 9 02:05:02 CEST 2013


[05:24] <cone-859> ffmpeg.git 03Michael Niedermayer 07master:ea538b0d64cf: avcodec/snowenc: fix constness of the AVFrame argument in encode_frame()
[05:29] <cone-859> ffmpeg.git 03James Almer 07master:e3d34100e73c: doc: add *install-doc that covers all documentation
[09:42] <ubitux> wm4: ping
[09:43] <wm4> ubitux: hi
[09:43] <ubitux> wm4: can you test a vobsub seeking patch with mpv?
[09:44] <wm4> yes
[09:44] <ubitux> wm4: http://b.pkh.me/0001-avformat-vobsub-fix-seeking.patch
[09:50] <ubitux> actually, without the stream_index = 0 chunk (patch updated)
[09:51] <wm4> tested the first version, worked
[09:51] <ubitux> did you try with the different subtitles stream?
[09:51] <wm4> yes
[09:51] <ubitux> cool
[09:51] <ubitux> without the new version it should be more accurate
[09:52] <ubitux> especially in case the timestamps mismatch between the streams :p
[09:52] <wm4> should I test the updated patch?
[09:52] <ubitux> could be nice
[09:52] <ubitux> gonna do a few more tests myself and will push
[09:52] <ubitux> that seeking function could be improved to be a lot faster btw
[09:55] <wm4> ok tested the updated patch
[09:55] <wm4> also worked
[09:56] <ubitux> cool, thanks :)
[09:56] <wm4> and I think it rally improved accuracy, because in my first test there was a weird case where I got too many packets from before the seek target
[09:57] <wm4> will the patch go into the stable releases?
[09:57] <ubitux> it's a good candidate
[09:57] <wm4> awesome
[09:58] <ubitux> wm4: do you specify the vobsub stream while seeking btw?
[09:58] <wm4> my old code didn't, then I added a work around that does
[09:58] <ubitux> ok
[09:59] <ubitux> it should not be needed anymore
[09:59] <wm4> yes I'd like to remove it again
[10:00] <wm4> I'm not evrn sure why the API allows selecting the stream...
[10:00] <wm4> in general you just want to seek somewhere you can resume playback
[10:01] <wm4> and the demuxer knows which streams you want (via AVStream.discard or whatever it was)
[10:01] <ubitux> that is, for playback :p
[10:01] <ubitux> you may want an accurate seeking for video only if you're transcoding only the video
[10:01] <ubitux> i guess..
[10:01] <wm4> what would be helpful would be something for files with possible timestamp resets
[10:01] <wm4> maybe the ability to specify relative seeks or so
[10:02] <wm4> well, if you want video only, disable the non-video streams
[10:02] <ubitux> dunno, i'm not a seeking expert :p
[10:03] <ubitux> it's too bad bsearch() doesn't allow an opaque parameter :(
[10:03] <wm4> as it is, mplayer demuxers are still much better at seeking for weird reasons
[10:03] <wm4> wow there are people who use bsearch()?
[10:03] <ubitux> why not?
[10:04] <wm4> binary search is rather trivial to implement (though also easy to get wrong), and bsearch strikes me as one of those useless C legacy APIs
[10:04] <ubitux> :(
[10:04] <ubitux> ok ok
[10:04] <ubitux> we use it in 3 places in ffmpeg
[10:45] <zidanne> Doesn ffmpeg itself supports HE-AAC v2 by default?
[10:45] <zidanne> (for decoding)
[10:48] <ubitux> michaelni_: "If flags contain AVSEEK_FLAG_BACKWARD, it is ignored." // should i assume seeking callback are supposed to always seek backward, even if the forward difference is lower than the backward one? (and min/max ts will make it forward if necessary?)
[10:51] <nevcairiel> the new seeking API controls this based on the three times you give it
[10:52] <ubitux> right ok
[10:53] <wm4> can you express anything of practical importance with the new API you couldn't with the old?
[10:53] Action: wm4 fails to see the point
[10:54] <nevcairiel> in theory you can tell it how far back to look for the last keyframe, for example
[10:56] <nevcairiel> or stuff like "i want a keyframe at X, but i'll also take one between x-10s and x+1s"
[10:56] <nevcairiel> not that it really matters since nothing implements this
[10:57] <wm4> well, if you don't get a keyframe at X, you can either get one after that or one before that (which you can control with the backwards flag)
[10:58] <nevcairiel> i could see the new api as useful if its implemented, for example if seeking i would be fine with getting a keyframe which is up to 1s after my seek point, if thats closer to the target then the last keyframe before the seekpoint
[10:59] <wm4> add a AVSEEK_FLAG_CLOSEST...
[11:00] <wm4> and you could get that with the old api
[11:00] <wm4> and actually, maybe the ability to export the index to the user would be even better in some cases
[11:00] <nevcairiel> strictly the closest could still be too far into the future, in that case i don't want it
[11:00] <nevcairiel> also, you can get the index
[11:00] <wm4> you can?
[11:01] <nevcairiel> every AVStream has (nb_)index_entries
[11:01] <nevcairiel> many demuxers export the file index in there
[11:01] <nevcairiel> if the file only has one central index, they usually put it into the video stream
[11:02] <nevcairiel> from my experience its useful for mkv, avi and mp4
[11:02] <wm4> interesting
[11:02] <nevcairiel> maybe some other fringe formats
[11:02] <wm4> could be useful to handle some mp4 corner cases
[11:46] <cone-377> ffmpeg.git 03Luca Barbato 07master:85ac12587bfe: nuv: check ff_rtjpeg_decode_frame_yuv420 return value
[11:46] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:58e5f6a43320: oMerge commit '85ac12587bfef970d0e0e4abc292df346daf8478'
[12:08] <cone-377> ffmpeg.git 03Luca Barbato 07master:7ee191cab0dc: dv: Add a guard to not overread the ppcm array
[12:08] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:13e35a25f670: Merge commit '7ee191cab0dc44700f26c5784e2adeb6a779651b'
[12:30] <cone-377> ffmpeg.git 03Anton Khirnov 07master:26f027fba1c5: lavf: fix the comparison in an overflow check
[12:30] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:53fb52ac859c: Merge remote-tracking branch 'qatar/master'
[12:57] <cone-377> ffmpeg.git 03Clément BSsch 07master:f8678dcef3c5: avformat/vobsub: fix seeking.
[12:57] <cone-377> ffmpeg.git 03Clément BSsch 07master:1ca4bf930bab: avformat/subtitles: check lower bound for duration overlap seeking.
[12:57] <cone-377> ffmpeg.git 03Clément BSsch 07master:b1319e14e3b0: avformat/subtitles: binary search seeking.
[12:57] <ubitux> wm4: tests welcome ^
[12:59] <wm4> thanks for fixing this, will later
[12:59] <wm4> *will test later
[12:59] <ubitux> also faster! (maybe)
[13:06] <durandal_1707> what is max/min y/u/v value for 10/12/16 bit yuv ?
[14:01] <cone-377> ffmpeg.git 03Paul B Mahol 07master:0227b4292629: avfilter/vf_psnr: refactor subsampled format support
[14:41] <cone-377> ffmpeg.git 03Giorgio Vazzana 07master:7f6ec05f0952: lavd/v4l2: Improve debug message
[15:03] <cone-377> ffmpeg.git 03Paul B Mahol 07master:14851ca5f5a3: avformat/matroskaenc: remove bogus prores tag
[15:37] <ubitux> lol
[15:37] <ubitux> i just found a srt
[15:37] <ubitux> where there are no \n
[15:37] <ubitux> but only \r
[15:37] <ubitux> what. the. hell.
[15:38] <JEEB> lol
[15:39] <ubitux> and the frightening part is that it might be supported by chance in some player(s)
[15:46] <Daemon404> ubitux, so it was made on a mac.
[15:46] <ubitux> Daemon404: can you explain what was going on here?
[15:47] <Daemon404> macs use CR for linebreaks
[15:47] <Daemon404> end of story?
[15:47] <nevcairiel> didnt they stop doing that like centuries ago?
[15:48] <Daemon404> i bet some apps still do
[15:48] <Daemon404> or teh srt is old as time
[15:48] <ubitux> srt is pretty recent, a few days ago
[15:53] <JEEB> Apple stuff before OS X did \r only
[15:54] <JEEB> OS X afaik does \n
[16:01] <ubitux> so we have \n, \r, and \r\n
[16:01] <ubitux> anyone knows another brilliant system with \n\r now?
[16:02] <durandal_1707> that is very old story....
[16:03] <durandal_1707> srt files are text files
[16:03] <nevcairiel> i dont think \n\r was ever used officially by anything
[16:03] <ubitux> good
[16:03] <ubitux> well i'll just fix that shit in a moment
[16:05] <ubitux> LF+CR: Acorn BBC and RISC OS spooled text output.
[16:05] <ubitux> ohgodno.
[16:06] <nevcairiel> lol
[16:07] <JEEB> lol
[16:09] <ubitux> ZX80 and ZX81, home computers from Sinclair Research Ltd used a specific non-ASCII character set with code NEWLINE (0x76, 118 decimal) as the newline character.
[16:09] <ubitux> i love this one ^
[16:10] <durandal_1707> dualinput eats last frame
[16:12] <durandal_1707> in some strange case...
[16:16] <kierank> Daemon404: srt is pissing me off
[16:16] <kierank> people are using it as a dodgy pro subtitle interchange format
[16:17] <Daemon404> Nothing New (TM)
[16:17] <kierank> and then somehow it has to fit into two bytes per frame
[16:17] <kierank> s/it/the captions have
[16:34] <durandal_1707> actually the dialinput bug can be reproduced with psnr and single images
[16:49] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:c914c99d4b81: swscale/utils: use memcpy instead of loop in sws_cloneVec()
[17:14] <cone-377> ffmpeg.git 03Paul B Mahol 07master:7d7b4e8d5cbc: avcodec/png: remove obsolete comments
[18:14] <durandal_1707> this libpng is awful function args in header are not explained at all
[18:56] <cone-377> ffmpeg.git 03Clément BSsch 07master:cfcd55db164e: avformat/srtdec: skip initial random line breaks.
[18:56] <cone-377> ffmpeg.git 03Clément BSsch 07master:90fc00a623de: avformat/subtitles: add a next line jumper and use it.
[18:56] <cone-377> ffmpeg.git 03Clément BSsch 07master:378a830e7b9a: avformat/subtitles: support standalone CR (MacOS).
[18:56] <cone-377> ffmpeg.git 03Clément BSsch 07master:860a0810583f: avcodec/assenc: fix potential overread.
[18:56] <cone-377> ffmpeg.git 03Clément BSsch 07master:3a54c221d574: avcodec/srtdec: fix potential overread.
[18:56] <ubitux> ok so, still not perfect, but better.
[19:07] <durandal_1707> michaelni_: is there some macro for 'rounding to nearest'?
[19:08] <ubitux> ROUNDED_DIV() ?
[19:10] <durandal_1707> there is no documentation
[19:11] <durandal_1707> i need to study macro to know what it does
[19:11] <durandal_1707> and it nowhere says it rounds to what...
[19:12] <durandal_1707> i will just switch everything to double
[19:16] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:8586d8dda201: avformat/avidec: fix signedness of pointer type in get_stream_idx()
[19:38] <durandal_1707> hmm it appears to me that this "none" color thing in rotate filter is nonsense
[19:38] <durandal_1707> it would left uninitialized data
[19:41] <durandal_1707> lol i get nice effect if i use 'none'
[20:38] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:c54aa2fb0f86: avcodec/truemotion2: Fix av_freep arguments
[20:45] <durandal_1707> lol, i looked at that code but failed to notice it
[20:52] <durandal_1707> why is there av_calloc and av_mallocz_array?
[22:18] <ubitux> beastd: what regression are you talking about?
[22:19] <ubitux> ow.
[22:19] <ubitux> yeah i guess i didn't test enough the intermediate one
[22:19] <beastd> ubitux: hi
[22:20] <beastd> I think you found it
[22:20] <ubitux> i actually wrote the final version first
[22:20] <ubitux> and split the code
[22:20] <beastd> yeah, it is easy to mess up that way
[22:20] <ubitux> so the 2-line function was written at the end, as an intermediate commit
[22:21] <beastd> same goes for many git rebase -i actually
[22:21] <ubitux> in practice it worked, since it makes a +1 all the time
[22:21] <ubitux> but that's not fixing the issue the commit was all about
[22:21] <ubitux> well anyway, upstream is fixed
[22:21] <beastd> yes, it doesn't help with overread
[22:23] <beastd> ubitux: yes. i actually did think about not posting about it
[22:23] <ubitux> it's good you noticed
[22:23] <ubitux> because the "commit fix" needs to be backported, while the following one adding the feature is not essential
[22:24] <ubitux> so the backport needs to be done properly ;)
[22:24] <beastd> but than I thought I would rather say sth if someone decided to backport sec fix and not improvement ( CR only support)
[22:24] <ubitux> :)
[22:24] <beastd> same thought :)
[22:24] <ubitux> now you won the right to backport it yourself while i put my brain at sleep
[22:25] <beastd> i am already over my time for today. remind me tomorrow if you want me to do the backport.
[22:26] <ubitux> i already forgot
[22:26] <ubitux> what were we talking about?
[22:26] <beastd> subtitle line skip overread by 1 char
[22:27] <ubitux> :(
[22:30] <beastd> ubitux: gn8 (I will try to remind myself)
[22:32] <ubitux> 'night :)
[22:41] <ubitux> durandal_1707: can you please write a thunderbird tutorial plz plz
[22:46] <durandal_1707> here is evil one: start->programs->thunderbird->uninstall
[22:51] <cone-377> ffmpeg.git 03Paul B Mahol 07release/1.0:8b282d9d8b78: avformat/matroskaenc: remove bogus prores tag
[22:51] <cone-377> ffmpeg.git 03Paul B Mahol 07release/1.1:8715ca0b82c1: avformat/matroskaenc: remove bogus prores tag
[22:51] <cone-377> ffmpeg.git 03Paul B Mahol 07release/1.2:8f87e75c6c92: avformat/matroskaenc: remove bogus prores tag
[22:52] <cone-377> ffmpeg.git 03Paul B Mahol 07release/2.0:f5ae34250a22: avformat/matroskaenc: remove bogus prores tag
[22:56] <cone-377> ffmpeg.git 03Paul B Mahol 07master:cd6241b5dfb1: avcodec/truemotion2: use av_calloc() & av_malloc_array()
[23:14] <wm4> ubitux: that vobsub seek commit works fine
[23:14] <wm4> ubitux: backport pls
[23:18] <cone-377> ffmpeg.git 03Michael Niedermayer 07master:d8fb170da2e7: avcodec/ratecontrol: give some trivial tips in case of buffer underflows
[23:31] <durandal_1707> michaelni_: i think 2 prores decoders could be merged into one
[23:33] <michaelni_> durandal_1707, you talked with the maintainer(s) of the 2 decoders ?
[23:34] <michaelni_> from what i remember elvis was in favor or merging them but the other side wasnt, but alot of time has passed so maybe its different now or i even misremember
[23:35] <durandal_1707> what other side? since when we need to listen other side?
[23:36] <Compnn> because then our prores will be different than libav prores
[23:36] <durandal_1707> i will inspect differences in depth
[23:36] <Compnn> you want to maintain it ? 
[23:36] <Compnn> (i dont mind if its merged or not. just trying to help)
[23:36] <Compnn> brb
[23:38] <durandal_1707> i see no prores in MAINTAINERS
[23:40] <michaelni> if theres no entry in maintainers then their authors are the next best persons to talk to
[23:41] <durandal_1707> i've lost elvis phone number
[23:44] <cone-377> ffmpeg.git 03Paul B Mahol 07master:43497d8a838e: avcodec/proresdec2: return meaningful error codes
[00:00] --- Mon Sep  9 2013


More information about the Ffmpeg-devel-irc mailing list