[Ffmpeg-devel-irc] ffmpeg-devel.log.20120823

burek burek021 at gmail.com
Fri Aug 24 02:05:02 CEST 2012


[02:34] <Daemon404> cbsrobot, have you tried ffmbc
[02:56] <Compn> ffmpeg really should merge all of ffmbc
[02:56] <Compn> get bcourdier back in here :D
[02:57] <kierank> well ffmpeg can't
[02:57] <kierank> because ffmbc is gpl
[03:08] <Daemon404> fun factoid: ffmpeg compiles with msvc/converter too
[03:08] <Daemon404> (i rebased the patches i have on top of libav)
[03:10] <ubitux> you succeed in writing your converter using libclang?
[03:10] <kierank> what's the speed difference for h264 and the likes
[03:11] <ubitux> or you're refering to bbb/test.c ?
[03:11] <Daemon404> kierank, i want to wait till daniel's qpel patches get pushed to test that
[03:11] <Daemon404> ubitux, libclang
[03:11] <ubitux> :o
[03:11] <Daemon404> bbb wrote most of it
[03:11] <ubitux> awesome :)
[03:11] <ubitux> do you have a public repo for this?
[03:11] <kierank> i hope you don't get people to ship msvc builds
[03:11] <Daemon404> kierank, whys that
[03:11] <kierank> does msvc pass fate?
[03:11] <Daemon404> yes
[03:11] <kierank> Daemon404: because msvc stuff is slower
[03:11] <Daemon404> thats heresay
[03:12] <kierank> mingw binaries are fine
[03:12] <Daemon404> hearsay*
[03:12] <Daemon404> i doubt its more thah 2-3% max
[03:12] <kierank> well i doubt it's suddenly going to compile cabac wonderfully
[03:12] <Daemon404> and msvc generally does a betetr job optimizing C code from my experience
[03:12] <Daemon404> so we shall wait and see
[03:13] <kierank> you should present something about it at vdd :)
[03:14] <Daemon404> eh? nothing for me to present
[03:14] <Daemon404> my vote is still to ste up a fight club
[03:15] <Daemon404> let me download and apply daniel's qpel patches
[03:15] <Daemon404> and rebase my tree
[03:15] <Daemon404> ill run some simple tests
[03:15] <Daemon404> (late tonight)
[03:29] <Compn> kierank : yes, ffmpeg can, because ffmpeg can be compiled against gpl
[03:29] <kierank> Compn: think about it. how are you going to merge gpl patches into lgpl code
[03:29] <Daemon404> Compn, it would require relicensing existing files
[03:29] <Compn> just because debian/ubuntu cant figure how to distribute a gpl/lgpl hybrid binary doesnt mean squattttttt
[03:29] <Daemon404> authors wont be happy.
[03:30] <ubitux> Compn: it's a bit tricky with mixed code
[03:30] <Compn> is there mixed code ?
[03:30] <Compn> just ifdef gpl
[03:30] <Compn> :)
[03:30] <ubitux> also, ffmbc has no public source history btw
[03:30] Action: Compn joins others in ifdef hell
[03:30] <Compn> diff ffmpeg ffmbc, all new code = gpl
[03:30] <Compn> :P
[03:30] <ubitux> have fun doing it
[03:31] <Compn> bwahaha
[03:31] <Daemon404> ifdefing bugfixes?
[03:31] <Daemon404> thats silly.
[03:31] <ubitux> :)
[03:31] <Compn> thats reality
[03:31] <Compn> silly is yet another ffmpeg fork
[03:31] <Daemon404> man
[03:32] <Daemon404> configure on msys takes like the same amount of time as teh build
[03:32] <Daemon404> dat slow forking
[03:32] <Compn> yes, configure be slow on mingw
[03:32] <Daemon404> Compn, msys + msvc here
[03:32] <Daemon404> no mingw =p
[03:32] <Compn> wonder if you could cheat and use cygwin to run configure, then mingw to do the rest
[03:32] <Compn> oh
[03:32] <Compn> msvc...
[03:33] <Daemon404> and cygwin would be just as slow
[03:33] <Daemon404> because it's sh.exe thats slow.. due to lolforks
[03:33] <kierank> 02:31:52 <Compn> silly is yet another ffmpeg fork --> it's not a proper fork
[03:33] <kierank> it's a specialised fork
[03:33] <Compn> kierank : that just means its a fork with a small set of features that makes it easier to merge back :)
[03:33] <kierank> doubt it
[03:33] <Daemon404> it means features that are not compatible with merging back to ffmpeg
[03:33] <Daemon404> read: CFR
[03:33] <kierank> and other broadcast stuff
[03:34] <kierank> if there was git history it would be nicer, yes
[03:34] <Compn> could always have two muxers/codecs
[03:34] <Daemon404> some can be merged back though
[03:34] <Compn> like we have now
[03:34] <Compn> prores_lgpl prores_kostya
[03:34] <kierank> well in an ideal world everyone would maintain their own forks and just push into a main branch
[03:34] <Daemon404> timestamp stuff is non-trivial
[03:34] <Compn> mxfdec_gpl mxfdec...
[03:34] <Compn> if you want to keep seperate files anyhow
[03:35] <Compn> and not ifdef
[03:35] <Compn> theres no technical nor legal barrier to do this
[03:35] Action: Daemon404 is really happy how easy it is to build with msvc now...
[03:35] <Daemon404> --cc="compilewrap cl"
[03:35] <Daemon404> to pre-conversion
[03:35] <Daemon404> no*
[03:36] <Compn> Daemon404 : happy using microsoft tools? heh
[03:36] Action: Compn trolls :P
[03:36] <Daemon404> ms debugger > gdb
[03:36] <Daemon404> ;)
[03:36] <ubitux> what about& gdb -tui!?
[03:36] <Daemon404> o u
[03:36] <ubitux> :X
[03:37] <Compn> but really, thered be a ffmpeg -f ffmbc_mpegts 
[03:37] <Compn> so it wouldnt change options or anything
[03:37] <Compn> just add some more formats...
[03:37] <ubitux> oh, it seems cgdb is still maintained :)
[03:38] Action: Compn talks to the wind
[03:39] <ubitux> Compn: i think a smarter move would be to talk to Baptiste about a potential relicensing of some parts and make him backport some stuff
[03:39] <ubitux> but it's likely he won't agree with that
[03:39] <Compn> heh
[03:39] <Daemon404> dammit i forgot --enable-runtime-cpudetect
[03:39] <Compn> he could do that himself at any time ubitux
[03:40] <ubitux> it's not in his interest to do so
[03:40] <Compn> of course
[03:40] <Compn> my logic is flawless :P
[03:42] <ubitux> feel free to try your theory with practice
[03:42] <ubitux> i think you'll prove yourself wrong quite soon ;)
[03:42] <Compn> hehe
[03:42] <Compn> my theories in the past have caused problems
[03:42] <Daemon404> he occasionally relicenses fixes to lgpl
[03:42] <kierank> Daemon404: does runtime cpudetect even do anything these days?
[03:43] <Compn> e.g. mplayer filters
[03:43] <Daemon404> kierank, supposedly
[03:43] <kierank> really?
[03:43] <Daemon404> i do it Just In Case
[03:43] <Daemon404> until someone tells me otherwise
[03:43] <Daemon404> not like it can do harm
[03:43] <Compn> in theory it should prevent crashes due to wrong cpu instruction when running same binary on diff cpu
[03:43] <Compn> if you dont have sse3 instruction on a p4...
[03:43] <Daemon404> it does that anyway
[03:44] <Daemon404> in many places
[03:44] <ubitux> qatar uses cpu detect only in ppc
[03:44] <ubitux> we use it in pp and vf mp
[03:44] <ubitux> (+ppc too)
[03:44] <kierank> eugh stop calling it qatar
[03:44] <kierank> it's libav
[03:44] <ubitux> sorry
[03:44] <Compn> qatar is easier to grep
[03:44] <Compn> :P
[03:45] <Compn> esp when half of your code starts with libav*...
[03:45] <ubitux> qatar is a nice nickname :)
[03:45] <ubitux> and it avoid confusion with libav* indeed
[03:45] <ubitux> i could say "the fork" but it's like i'm trolling
[03:45] <ubitux> qatar sounds like a fancy nickname to me
[03:45] <Compn> ubitux : everything is trolling now
[03:46] <Daemon404> i failed at finding someone to draw my "bikeshed wars" tshirt for vdd
[03:46] <Daemon404> i wanted one so very much
[03:48] <ubitux> anyway, it doesn't look like cpudetection has much effect
[03:48] <Compn> wonder how many poor distro uses run into crashing builds :\
[03:48] <ubitux> at least as long as you don't use pp and ppc
[03:49] <Compn> do distros get crash reports, like on average how much does ffmpeg crash for users ?
[03:49] <kierank> windows probably does
[03:49] <kierank> unlikely that linux does
[03:49] <kierank> there's no reporting mechanism
[03:50] <Compn> opera and firefox have thier own crash reporting things for a while
[03:50] <Daemon404> if linux had crash reporting like windows, it'd be akin to a ddos
[03:50] <Compn> you'd think the distros would just have an aggregate information, just what cpu it crashed on...
[03:50] <Compn> derp
[03:50] <Compn> some distros have a generic reporter
[03:51] <kierank> vlc gets microsoft crash reports
[03:51] <Compn> yes, they pass them onto ffmpeg
[03:51] <Daemon404> that reminds me
[03:51] <Compn> when appropriate
[03:51] <Daemon404> the other thing this converter may be useful for is usign ms's arm compiler
[03:52] <Compn> its kinda weird having microsoft report crashes and help open source i tell you
[03:52] <ubitux> it's just to troll them with security issues
[03:53] <Daemon404> there are going to be at least 3 people using windows laptops at vdd this year... segregation time
[03:53] <Daemon404> (i bet)
[03:54] <Compn> nah, just time to put some windows vm and load it up with worms. make sure your fellow devels have thier windows patched up ;)
[03:54] <Compn> er windows vm on the wifi
[03:54] <Daemon404> (protip: one of them is me)
[03:55] <Compn> w7 or w8 ? :P
[03:55] <Daemon404> 7
[03:55] <ubitux> (2 ppl left to hunt)
[03:55] <Compn> why is it that i only like playing opengl games, and that valve and id software have stuck with opengl all these years? :P
[03:56] <Compn> well rip id software anywyas
[03:56] <Compn> re w8 and opengl benchmarks
[04:44] <xxthink> hello
[04:44] <xxthink> how to reduce the moov size of the mp4 file
[05:10] <Compn> xxthink : run qtfaststart on it ?
[10:27] <ubitux> btw, why "image2" instead of just "image"?
[10:37] <av500> why decode_audio4 :)
[10:39] <ubitux> ok, same thing again then :)
[10:39] <ubitux> Daemon404: did you see http://fate.ffmpeg.org/report.cgi?time=20120823001916&slot=x86_64-archlinux-gcc-valgrindundef ?
[10:39] <ubitux> (cllc tests)
[11:03] <saste> ubitux: for time manipulation effects the best thing would be to modify the duration
[11:04] <saste> that is through a setduration filter
[11:04] <saste> and fix the PTS accordingly with setpts
[11:04] <ubitux> mmh why not
[11:04] <ubitux> :)
[11:05] <saste> you could even change speed during playback through a GUI and reconfiguration commands
[11:06] <ubitux> how would you use it?
[11:06] <saste> use what?
[11:06] <ubitux> duration filter
[11:07] <saste> for doing silly things like doing slowmo effects
[11:07] <ubitux> yes i know, but with a simple usage, let's say
[11:07] <ubitux> you want to drawtext from 10 to 20 seconds, and from 15 to 25 you want to blur the pic
[11:07] <ubitux> what the cmd line would look like?
[11:08] <saste> how is that related to setduration?
[11:09] <ubitux> ooh i misunderstood
[11:09] <saste> using setduration the yesterday problem would be
[11:12] <saste> setduration=st((gt(t, 5)*lt(t, 10), 0); if(ld(0), DURATION*3) + ifnot(ld(0), DURATION), setpts=PREV_OUTPTS+DURATION
[11:12] <saste> ifelse(...) may help also
[11:12] <saste> but that needs to be implemented
[11:13] <ubitux> mmmh
[11:14] <ubitux> ok
[11:32] <japjap> I am loosing RTPs when using RTSP via UDP. How to cope with this ?
[15:01] <CIA-56> FFmpeg: 03Michael Niedermayer 07master * r599a4e0 10/ libavcodec/mpeg4audio.c : mpeg4audio: dont take the SBR flag too serious. (+14 more commits...) - http://git.io/wBnVZA
[15:38] <TimNich> c2271f1a  make is failing with a "ffmpeg_opt.o and ffmpeg_filter.o no such file" error after a make clean/distclean, ffplay is OK, anyone else had any problems?
[15:39] <ubitux> 'c2271f1a': unknown revision or path not in the working tree.
[15:40] <TimNich> oops c2271fa
[15:41] <ubitux> sounds totally unrelated
[15:42] <ubitux> maybe use git clean (carefully)
[15:49] <TimNich> Hmmm this file appeared (6c180b) since my last build, before going away, but I don't see a corresponding addition to the Makefile..
[15:50] <ubitux> you re-run configure right?
[15:53] <TimNich> yes, make clean && make distclean && ./configure,,, all by the usual script
[15:55] <ubitux> check git status
[15:57] <CIA-56> FFmpeg: 03jamal 07master * r732dfc7 10/ libavformat/ape.c : lavf/ape: Fix stream duration calculation - http://git.io/57ofeA
[15:57] <TimNich> "nothing added" one untracked file which is  .patch leftover
[16:37] <Daemon404> ubitux, k
[16:37] <Daemon404> i had trouble properly checkign cllc because GET_VLC made valgrind freak out
[16:37] <Daemon404> and kill itself
[16:37] <Daemon404> oh wait
[16:38] <Daemon404> thats exactly what that is.
[16:38] <Daemon404> line 166 is GET_VLC(code, bits, gb, vlc->table, 7, 2);
[16:49] <TimNich> curious, a fresh git clone falls over in exactly the same way...
[17:23] <michaelni> TimNich, which platform, which make version is this happening at ?
[17:31] <TimNich> michaelni: Running on Suse 12.1. I think I have narrowed  it down. It falls over if you set --progs-suffix
[17:33] <TimNich> my last good build was Aug 08, I know this because I use --progs-suffix to easily identify versions...;P
[17:33] <michaelni> TimNich, look at DOPROG in Makefile and surroundinh code, this is related to the ffmpeg_*.o stuff
[17:33] <michaelni> maybe theres a bug in it
[17:34] <TimNich> OK I will try and do a bisect to also narrow it down
[17:36] <michaelni> its probably the commit that split the stuff off ffmpeg.c ...
[17:37] <TimNich> Yes I reckon that, so will start either side
[18:42] <TimNich> michaelni: Yes its  in 11a1033c9f
[18:54] Action: TimNich of home now
[19:13] <CIA-56> FFmpeg: 03Michael Niedermayer 07master * r2b3b52d 10/ libavcodec/lagarith.c : lagarith: fix alignment on buffers passed into prediction dsp code. - http://git.io/tzsmug
[19:51] <saste> ubitux: why 9a19341e6e7965818898bf624dbf8d5bb418f3d3 ?
[19:51] <saste> or in other words, what's the difference between LDFLAGS and LDLIBS?
[19:52] <ubitux> LDFLAGS ’ -Wl,..., LDLIBS = -lxxx
[19:53] <ubitux> look at the make -p output
[19:53] <ubitux> the implicit rules use LDFLAGS and LDLIBS differently
[19:53] <saste> uhm ok... where can I find some official documentation for them?
[19:53] <saste> I always used LDFLAGS for library linking flags
[19:54] <ubitux> gmake documentation i guess
[19:55] <saste> that means that environment LDFLAGS is completely ignored by examples/Makefile?
[19:55] <ubitux> no, you can set it, the implicit rules will make use of them
[19:55] <ubitux> try maybe LDFLAGS=-Wl,--as-needed make and see if that's honored
[19:56] <ubitux> if you want too add libs, LDLIBS=-lfoobar LDFLAGS=-Wl,--as-needed make should do the trick as well
[19:59] <ubitux> saste: do you think i made a mistake?
[20:00] <CIA-56> FFmpeg: 03Georg Lippitsch 07master * r15acfa2 10/ libavcodec/ffv1.c : ffv1: more than 8 bit per RGB channel - http://git.io/FX2YRQ
[20:00] <saste> ubitux: no but it looked weird, and it was apparently ignoring LDFLAGS
[21:11] <CIA-56> FFmpeg: 03Joseph Artsimovich 07master * r628e6d0 10/ libavcodec/dnxhdenc.c : Fix partially hidden macroblocks for 10-bit DNxHD. - http://git.io/ssF9SA
[21:30] <Compn> is there any filter to say ... skip every 3rd frame ?
[21:31] <Compn> er, i guess skipping decoding of every 3rd frame wont work
[21:31] <Compn> how about decoding only keyframes ?
[21:32] <Daemon404> you can do that in avisynth for 10 years ago
[21:32] Action: Daemon404 runs
[21:32] <Daemon404> from*
[21:34] <Compn> yes vdub has show keyframes only while seeking too
[21:34] <Compn> theres probably a lavc decoder option for it
[21:34] Action: Compn digs into manual
[21:34] <ubitux> < Compn> is there any filter to say ... skip every 3rd frame ? // -vf 'select=not(mod(n\,3))' ?
[21:36] <ubitux> if you want only the keyframes sth like select=eq(pict_type\,I) or something
[21:37] <Daemon404> (btw thats still horribly ugly and a massive pita to script from the cmd line)
[21:37] <Compn> oh ok it is -lavdopts skipframe ;)
[21:37] <Daemon404> i had to resort to wine + avisynth today to: stack 2 videos and add the text "original" and " transcoded" n top of them
[21:37] Action: Compn using mplayer
[21:41] <ubitux> Daemon404: you can do that with libavfilter without much trouble
[21:42] <Daemon404> frame-accurately?
[21:42] <Compn> ubitux : we need ability to start on frame 33224 :P
[21:42] <Compn> stupid frame accuracy
[21:42] <retrosnub1> when using v4l2 as input ffmpeg seems to hang in libavformat/utils.c in function avformat_find_stream_info looping endlessly because of AVERROR(EAGAIN) & any ideas?
[21:43] <Daemon404> Compn, even linearly, i doubt it is
[21:43] <Daemon404> e.g.
[21:44] <Daemon404> in avs: stackvertical(clip1.assumefps(24).bicubicresize(704,400), clip2.assumefps(24).bicubicresize(704,400))
[21:45] <Daemon404> ive struggled with ffmpef adding/dropping frames a lot in my time..
[21:45] <Daemon404> and such.
[22:13] <Compn> Daemon404 : for a devel, it looks kinda like you use windows tools more than ffmpeg....
[22:13] <Daemon404> i use both a lot
[22:13] <Daemon404> but i use What Works For The Task
[22:13] <Compn> fair enough :)
[22:14] <Daemon404> just cause i work on foss doesnt mean in a fossie
[22:14] <Daemon404> s/in/im/
[22:14] <ubitux> :))
[22:14] <Compn> fossil :P
[22:14] <Compn> would have been also acceptable
[22:14] <Daemon404> nah...
[22:15] <Daemon404> fossil's in teh ground generally have less dirt on them than foss hippies
[22:15] <Daemon404> bah grammar.
[22:15] Action: Compn shakes some dirt off himself
[22:15] Action: Compn sweeps it under CIA-56
[22:18] <CIA-56> FFmpeg: 03Michael Niedermayer 07master * r168ddcd 10/ libavcodec/mjpegdec.c : mjpeg: fix fliping with emu edges. - http://git.io/4lPGMw
[00:00] --- Fri Aug 24 2012


More information about the Ffmpeg-devel-irc mailing list