[Ffmpeg-devel-irc] ffmpeg-devel.log.20130530

[burek]

[00:19] <michaelni> funman, IIRC the way the code behaves is sort of a consequence of the external API, so the code can work with existing users of libavcodec without them needing to use a new API
[00:19] <michaelni> but it should be quite easy to change the buffering /delaying if someone wants
[00:22] <michaelni> ATM it should be delaying until its "num of threads" buffer is full and then block
[00:22] <michaelni> i suspect its a 1-2 line change to make it do something different
[00:58] <cone-452> ffmpeg.git 03Michael Niedermayer 07master:29e97cd6c3f2: avformat/mov: Try to detect and support files with invalid sample sizes
[01:06] <cehoyos> michaelni: Did you test the samples from ticket 2605 with ffplay? Timestamps appear to be off, afaict.
[01:20] <michaelni> cehoyos, off relative to what ?
[01:20] <michaelni> its a hex edited file
[01:20] <michaelni> its not valid mov
[01:20] <michaelni> does quicktime produce better timestmps ?
[01:25] <durandal_1707> ffprobe -show_format does not list chapters and chapters tags
[01:32] <cehoyos> QuickTime plays the file with a speed that seems likely, ffplay plays video in slow-motion (because audio timestamps do not increase iiuc)
[03:17] <cone-452> ffmpeg.git 03Ash Hughes 07master:fa30f4c24d10: fate: add smv test
[03:38] <cone-452> ffmpeg.git 03Michael Niedermayer 07master:455ae871a8c9: libavcodec/smvjpegdec: pass flags / idct algo into jpeg avctx
[03:38] <cone-452> ffmpeg.git 03Michael Niedermayer 07master:63e5e9f7c5ad: fate: fix smvjpeg test
[09:33] <cone-514> ffmpeg.git 03Luca Barbato 07master:91a6944e5623: vmd: use the PALETTE_COUNT constant uniformly
[09:33] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:6635b13411a8: Merge commit '91a6944e56236234f0a7ba162404665753cbcb51'
[09:41] <cone-514> ffmpeg.git 03Luca Barbato 07master:c8f3cb9119c2: vmd: return meaningful errors
[09:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:8ef89833d0a4: Merge commit 'c8f3cb9119c2183680d44a509a1b5a9817a3bee9'
[09:51] <cone-514> ffmpeg.git 03Luca Barbato 07master:676da248cad4: vmd: refactor the inner decode loop
[09:51] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:ab78e21ea285: Merge commit '676da248cad49debc40720baa13214f0b94dcc71'
[10:37] <cone-514> ffmpeg.git 03Luca Barbato 07master:701966730ce1: vmd: drop incomplete chunks and spurious samples
[10:37] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:103ffde5a391: Merge commit '701966730ce10290fd49c5ccedd73f505680f764'
[10:44] <cone-514> ffmpeg.git 03Diego Biurrun 07master:9b9b2e9f3036: build: arm: cosmetics: Place all OBJS declarations in alphabetical order
[10:44] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:1bbbbb0a3292: Merge commit '9b9b2e9f3036abfd42916bcf734af14b4cb686aa'
[10:46] <cone-514> ffmpeg.git 03Stefano Sabatini 07master:9f8937b5ef13: lavd: add xv output device
[10:46] <cone-514> ffmpeg.git 03Stefano Sabatini 07master:f7a267affe80: doc/encoders: remove duplicated "the"
[10:55] <cone-514> ffmpeg.git 03Diego Biurrun 07master:186599ffe0a9: build: cosmetics: Place unconditional before conditional OBJS lines
[10:55] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:3a0e21f037c0: Merge commit '186599ffe0a94d587434e5e46e190e038357ed99'
[11:02] <cone-514> ffmpeg.git 03Martin Storsjö 07master:36a7df8cf111: arm: Only build the FFT init files if FFT is enabled
[11:02] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:0b539da4c74b: Merge commit '36a7df8cf1115aa37a1b0d42324ecde5ab6c2304'
[11:07] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/1.1:2416eff5b927: ff_read_timestamp: check stream_index before using it as array index
[11:07] <burek> Is there a more detailed document about ffserver that can tell me which vcodec/acodec is being used between the ffmpeg (as a feed) and ffserver, and how the codecs are being chosen for the broadcast to the each client
[11:14] <cone-514> ffmpeg.git 03Martin Storsjö 07master:efb7968cfe8b: arm: Don't unconditionally build dsputil files
[11:14] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:e119d1b345c1: Merge commit 'efb7968cfe8b285ab4f27b363719b7c92d19ec74'
[11:22] <cone-514> ffmpeg.git 03Martin Storsjö 07master:86113667c0fc: arm: Include hpeldsp_neon.o if h264qpel is enabled
[11:22] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:1a7ae6be2a87: Merge remote-tracking branch 'qatar/master'
[12:12] <ubitux> we don't have a get_bits interface doing avio_read automatically when necessary, right?
[12:49] <michaelni> ubitux, no, for what would you need that ?
[12:49] <michaelni> also it would be inefficient
[12:50] <ubitux> i need it in a demuxer where i can't guess the number of bytes in advance
[12:50] <ubitux> and where i can't seek back
[13:04] <ubitux> michaelni: eb592f6af oh you :3
[13:15] <cone-514> ffmpeg.git 03Carl Eugen Hoyos 07master:b3af9242d013: Use -lX11 -lXext when linking Xv output device.
[14:19] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:8bedbb82cee4: j2k/jpeg2000: split data pointer in int & float.
[14:42] <saste> michaelni, I think mcdeint is reading uninited data, i.e. data in the padding
[14:42] <saste> CHECK(j) macro
[14:43] <saste>  ABS(srcp[-srcs-1+(j)] - srcp[+srcs-1-(j)]) + etc.
[14:43] <saste> srcp points to a pixel in the source image
[14:50] <michaelni> hmm like yadif 
[14:50] <michaelni> if so see yadif on how to possibly solve it
[14:57] <saste> michaelni, http://pastie.org/7983594
[14:57] <saste> ^^ this fix the diff issue
[14:57] <saste> now i'll look into yadif if there is a better way
[15:02] <michaelni> saste, that pastie patch is no fix
[15:02] <michaelni> its a bit better than return 0 but not much
[15:02] <saste> michaelni, i never claimed that, but it fixes the diffing issue
[15:02] <saste> with that i get consistent results
[15:02] <michaelni> hmm ok misunderstood
[15:04] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:84a6bc23570c: jpeg2000: tag_tree_size: replace check by assert
[15:04] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:efc2df2e27cc: jpeg2000: remove unneeded variable initialization
[15:04] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:09e03a84674f: jpeg2000: merges quantization fixes from j2k
[15:04] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:c2ac389bba24: jpeg2000: remove unneeded casts
[15:04] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:62d00aa22f28: jpeg2000/j2k: merge j2k/jpeg2000.c/h
[15:13] <saste> michaelni, i don't understand yadif right now, how many chances there are you can fix mcdeint yourself?
[15:14] <saste> or i think i'll just create a ticket
[15:41] <michaelni> saste, i dont think ill work on mcdeint bitexactness any time soon
[15:41] <michaelni> i dont remember when i last used the filter nor do i remember many users useing it
[15:42] <michaelni> and it works fine without bitexactness
[15:42] <michaelni> just not testable easily
[15:42] <saste> michaelni, should i just port it to lavfi and drop mp=mcdeint?
[15:43] <saste> note that my naive fix doesn't seem to generate visual problems
[15:43] <michaelni> if the differences are confined to the edges then it should be fine to drop mpcodecs mcdefint
[15:44] <saste> michaelni, the differences propagate to the center of the image
[15:44] <saste> but usually are just a few units, so it's makes no visual difference
[15:44] <michaelni> hmm
[15:44] <saste> ffplay -i matrixbench_mpeg2.mpg -vf "interlace,split[a][b]; [a]unsharp,showinfo,mp=mcdeint[a1]; [b]unsharp,showinfo,mp=mcdeint[b1];[a1][b1]blend=all_mode=difference,lutyuv=y='if(val,maxval,0)'"
[15:44] <michaelni> if you want you can also fix it, it shouldnt be hard
[15:44] <saste> so you get an idea
[15:45] <michaelni> i mean take the pixel closest to the edge if  outside pixel would have been used
[15:45] <michaelni> just dont skip any output pixels from being calculated
[15:45] <saste> ok
[15:52] <microchip_> michaelni: any idea why mpeg2video doesn't support MPEG-style quantizers?
[16:12] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:6f7de3d5c8a5: mpegvideo_enc: allow mpeg_quant to be set in mpeg2video
[16:51] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:310f9dd6988c: j2k/jpeg2000: merge pix_fmt setting code
[17:54] <cone-514> ffmpeg.git 03Clément BSsch 07master:0ab9362fcb77: lavfi: add vignette filter.
[18:01] <wm4> it appears Libav commit 9b6f47c (which was already merged) changes what wavpack packets the wavpack decoder accepts
[18:02] <wm4> and it seems old and new packet format are incompatible
[18:02] <wm4> libavcodec version was not bumped
[18:02] <wm4> is there any way around this so I can be compatible with both old and new libavcodec?
[18:14] <cone-514> ffmpeg.git 03Clément BSsch 07master:4a85390d29b6: lavfi/vignette: make sure a sane sar is set.
[18:22] <microchip_> michaelni: is mpeg_quant 1 compatible with -intra_matrix/-inter_matrix (ie, custom matrices) ?
[18:24] <michaelni> microchip_, should work but i didnt try
[18:24] <microchip_> ok, thanks
[18:24] <michaelni> wm4, i can bump minor version if that solves the issue ?
[18:24] <wm4> michaelni: I think there's already a minor bump in another unrelated commit after that
[18:30] <wm4> sigh, checking version number it is, then...
[18:39] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:783658393406: j2kdec: s/j2k_/jpeg2000_/
[18:39] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:9ccc9f30f7b6: jpeg2000: drop red-blue swap
[18:39] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:941a8efae2ac: j2k:decode_tile() pass picture as argument
[18:39] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:3f714d1ffd96: jpeg2000/j2k: merge non functional changes
[18:39] <cone-514> ffmpeg.git 03Michael Niedermayer 07master:826b3a75cd29: h264_cavlc: fix reading skip run
[18:40] <michaelni> wm4, if you have a better idea or patch ...
[18:41] <wm4> well, I don't know why exactly the "native" (or the _new_ "native") wavpack format doesn't work with the old decoder, so I better shut up
[18:57] <saste> michaelni, ugly patch sent
[19:17] <cehoyos> ubitux: Did you see the new "tiling" mail on ffmpeg-user? Is it ok to call -vf twice?
[19:18] <ubitux> calling -vf twice will ignore one of them iirc
[19:19] <ubitux> nicolas replied to the tiling mail afaict
[19:23] <wm4> so, is mcdeint really so great?
[20:12] <cehoyos> michaelni: I wanted to close ticket 244 ("rv10 / rv20 only support width and height that are multiples of 16 / 4") as fixed by ami_stuff in 388241ef and fef75ef but you have added an analysis that doesn't fit the original description or that I don't understand. Could you elaborate?
[20:24] <durandal_1707> why is there no way to disable xv?
[20:25] <wm4> why is xv needed in libavdevice at all?
[20:26] <durandal_1707> because libavdevice is meant to replace libao2 / libvo 
[20:27] <wm4> lol
[20:28] <durandal_1707> actually libavdevice is copy pasted libavformat with formats that did not look right to be in libavformat....
[20:34] <michaelni> cehoyos, my comment is about the issue described in the tickets title
[20:35] <michaelni> if the text of the ticket describes a different issue then well ...
[20:39] <michaelni> cehoyos, its possibly the example in the ticket had 2 issues and when i debuged it i stoped at the first and missed the second
[20:39] <michaelni> still AFAIK the first is not resolved
[20:53] <cone-514> ffmpeg.git 03Clément BSsch 07master:cf6f1790444e: lavfi/vignette: fix scale with aspects smaller than one.
[20:53] <cone-514> ffmpeg.git 03Clément BSsch 07master:bfafcc4ebe63: lavfi/vignette: simplify dmax definition.
[20:53] <cone-514> ffmpeg.git 03Clément BSsch 07master:33620e243fb6: lavfi/vignette: add some scaling debug.
[20:53] <cone-514> ffmpeg.git 03Clément BSsch 07master:2886e8065e63: lavfi/vignette: add aspect option.
[21:14] <ubitux> <+wm4> so, is mcdeint really so great? // it seems to have upset someone when saste said it was going to be delayed, so i guess yes
[21:14] <ubitux> also microchip_ was whining for it the other day
[21:14] <ubitux> since it seems to fix some yadif artifacts at times
[21:14] <microchip_> hey! :p
[21:14] <ubitux> :)
[21:14] <durandal_1707> well its only such filter anyway currenty here...
[21:15] <durandal_1707> if there is better filter, tell us
[21:15] <llogan> what kind of yadif artifacts? i haven't noticed any.
[21:15] <llogan> of course my sources are from 1994 on rotting tape
[21:15] <ubitux> llogan: http://guru.multimedia.cx/deinterlacing-filters/ i guess?
[21:16] <ubitux> maybe yadif behaves better now though
[21:16] <microchip_> nnedi3 needs to be ported too :p
[21:16] <durandal_1707> anything else?
[21:16] <llogan> yeah. an external filter without such a hard name to remember.
[21:16] <ubitux> microchip_: http://code.google.com/p/vapoursynth/source/browse/trunk/src/filters/eedi3/eedi3.c
[21:16] <ubitux> go ahead
[21:17] <ubitux> should be relatively easy
[21:17] <ubitux> most boring part will be writing the documentation, as usual
[21:17] <ubitux> microchip_: maybe you can start writing the documentation actually, doesn't require the-c-skill-you-don-t-want-to-learn ;)
[21:18] <microchip_> too lazy atm, watching Indiana Jones and chilling :p
[21:19] <ubitux> ah, so i guess that filter is not a priority then and we can delay it by several months/years
[21:20] <thegeek> are there any benchmark results for the new opencl filter implementations?
[21:21] <ubitux> durandal_1707: cehoyos will insist on keeping mp wrapper unless someone port pullup, so you could consider it a good candidate, because i'm not motivated into doing it
[21:23] <ubitux> and saste is likely traumatized by mcdeint to port this one btw
[21:23] <durandal_1707> i lost 99% of my brain cells by libswscale trip
[21:24] <kierank> lol
[21:26] <ubitux> mmh we have more than 160 filters
[21:26] <ubitux> that's starting to get pretty
[21:27] <microchip_> port spp too :p
[21:27] <ubitux> it's in progress
[21:27] <microchip_> good :)
[21:27] <ubitux> but i'm messing with something else right now
[21:28] <microchip_> the only thing that's still keeping me to mencoder, is mencoder's mpeg muxer for pulldown
[21:30] <durandal_1707> soft telecine
[21:30] <microchip_> yeah
[21:30] <microchip_> but
[21:30] <microchip_> mencoder is currently broken with AC3 audio, so i use ffmpeg instead
[21:36] <durandal_1707> there is no way to get exacst number of samples put in matroska?
[21:39] <cehoyos> michaelni: You changed the ticket title (244), the original issue reported by ami_stuff was that Realplayer and FFmpeg cannot decode samples encoded with -vcodec rv10/rv20 that have incrrect dimensions
[21:39] <cehoyos> I don't understand the issue described by you and don't know how to reproduce.
[21:43] <cehoyos> durandal_1707: Did you test --disable-outdev=xv
[21:44] <durandal_1707> cehoyos: no, but there should be way to disable xv, same like libcaca and sdl
[21:50] <cehoyos> You mean like --disable-indev=alsa ?
[21:54] <durandal_1707> cehoyos: no, like --disable-zlib
[22:02] <michaelni> cehoyos, take ANY high bitrate video and mux it in rm, it will fail
[22:04] <michaelni> cehoyos, ./ffmpeg -i matrixbench_mpeg2.mpg -vf scale=512:512,noise=100:100 -qscale 2 test.rm
[22:06] <ubitux> triggers an assert here
[22:06] <michaelni> i know
[22:06] <ubitux> but requires assert level 2
[22:07] <michaelni> at lower assert level you get a broken file as >64k packets are not support 
[22:07] <michaelni> which is what i said in the ticket
[22:07] <Daemon404> do people really use ffmpeg to mux rm?
[22:08] <cehoyos> ami_stuff used "200 kb/s" which I guess does not allow to reproduce this issue - thank you for explaining!
[22:17] <Compn> ffmpeg rm demuxer needs help, i cant imagine what the rm muxer looks like...
[22:21] <Daemon404> it cant be THAT bad
[22:21] <Daemon404> the gf watches copious amounts of evil chiense realmedia with it
[22:22] <Daemon404> without issue
[22:22] <durandal_1707> with realmedia player?
[22:23] <Daemon404> no
[22:23] <Daemon404> lav filters
[22:23] <Daemon404> aka ffmpeg
[22:25] <cone-514> ffmpeg.git 03Paul B Mahol 07master:c4e0e3142488: tta: remove pointless code
[22:31] <mateo`> durandal_1707: you asked me if i was registered to the ffmpeg-devel mailing: yes i am. Is there something wrong ?
[22:32] <durandal_1707> no, you put yourself into cc, and that is usually done when someone is not subscribed....
[22:33] <mateo`> durandal_1707: strange, i'll check my mailer
[22:34] <ubitux> durandal_1707: isn't this automatically done by git send-email?
[22:35] <durandal_1707> ubitux: not here
[22:35] <ubitux> mine does it
[22:39] <llogan> beastd: how was LinuxTag?
[22:41] <beastd> llogan: Was quite nice in Berlin. All went well and we had a nice booth with the xbmc.
[22:42] <llogan> who attended the booth with you?
[22:49] <saste> beastd, hey o/
[22:50] <saste> michaelni, ping on mcdeint
[22:52] <beastd> Hey saste. Everything went smoothly on your flight back to sardegna?
[22:53] <saste> beastd, sure
[23:07] <durandal_1707> what you were doing on LinuxTag?
[23:11] <saste> durandal_1707, sitting, looking around, talking with people, doing stuff
[23:12] <saste> durandal_1707, some more mp filter you want to port?
[23:19] <durandal_1707> maybe much later, currently no
[23:20] <beastd> llogan: Oops, I forgot to answer your question. It was Carl Eugen, Thilo, Stefano, Reimar and me. We shared a booth with another 4 people from XBMC.
[23:28] <durandal_1707> anything interesting to share?
[23:32] <cone-514> ffmpeg.git 03Matthieu Bouron 07master:f468325d3489: lavf/id3v2enc: fix cover art display on some software
[23:38] <beastd> durandal_1707: I did not fully process the event myself yet. Others may also have had other inputs I don't know of. But I think it was mostly the more usual user questions. I can't remember anything that would be interesting for you right now.
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/0.10:02923b8c781c: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/0.11:5fd83f29f7f2: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/0.7:746c7480703c: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/0.8:d04dc7b5a78b: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/0.9:8e7e12dcbfda: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/1.0:d70043ea0f47: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/1.1:a98775026765: h264_cavlc: fix reading skip run
[23:41] <cone-514> ffmpeg.git 03Michael Niedermayer 07release/1.2:96e6d4da374f: h264_cavlc: fix reading skip run
[23:44] <saste> michaelni, what about the perl doc stuff?
[00:00] <michaelni> saste, iam not a big fan of replacing 1 like awk scripts by 20line perl scripts but iam not docs maintainer
[00:00] --- Fri May 31 2013