[Ffmpeg-devel-irc] ffmpeg-devel.log.20180823

burek burek021 at gmail.com
Fri Aug 24 03:05:03 EEST 2018

[00:23:32 CEST] <durandal_1707>  did anyone received last patch from me?
[00:24:05 CEST] <durandal_1707> ah it appeared
[05:33:01 CEST] <TimothyGu> It takes so much work to roll FFmpeg in Chromium, the instructions are almost funny to read
[05:33:02 CEST] <TimothyGu> https://docs.google.com/document/d/14bqZ9NISsyEO3948wehhJ7wc9deTIz-yHUhF1MQp7Po/edit
[05:35:19 CEST] <TimothyGu> "It's much less stressful to do the roll earlier. If youre familiar with the process and get lucky (nothing breaks), you can finish the FFmpeg roll within a day."
[06:56:09 CEST] <kierank> atomnuker: they don't need to buffer frames to do that
[06:56:19 CEST] <kierank> just lines
[06:58:43 CEST] <Compn> dur : 
[06:58:46 CEST] <Compn> where he go
[07:10:36 CEST] <Compn> does someone still run a server box for devs to test things ?
[07:10:46 CEST] Action: Compn needs to build mplayer somewhere....
[11:40:39 CEST] <durandal_1707> ubitux: this site: http://www.arri.com/camera/alexa/tools/lut_generator/lut_generator/ can generate various 3d luts, some are unsupported by lut3d filter
[11:46:09 CEST] <ubitux> yeah i just picked the most common formats i was sent by people
[11:46:16 CEST] <ubitux> it's incomplete
[11:46:36 CEST] <ubitux> it's a mess of random text files
[11:46:53 CEST] <ubitux> (reminds you sth, doesn't it..)
[11:47:14 CEST] <ubitux> the only sane lut3d format is haldclut
[11:47:20 CEST] <ubitux> which is binary/an image
[11:47:26 CEST] <ubitux> ppl should just use that
[11:48:09 CEST] <JEEB> didn't you first poke at subtitles?
[11:48:24 CEST] <JEEB> which is one of the "random stuff in text" things
[11:48:24 CEST] <JEEB> :D
[12:04:06 CEST] <ubitux> JEEB: yeah that's what i was refering to :(
[13:18:33 CEST] <cone-207> ffmpeg 03Jun Zhao 07master:c9ed7f002492: lavc/encode: remove redundant av_init_packet after av_packet_unref.
[15:22:50 CEST] <durandal_1707> ubitux: why you didn't do tricubic interpolation in lut3d?
[15:30:48 CEST] <ubitux> too slow i suppose, i looked at the most common interpolation for lut3d
[16:01:31 CEST] <cone-207> ffmpeg 03Michael Niedermayer 07master:4251a44e7d9e: configure: do not add fsanitize cflags with ossfuzz if coverage is tested
[16:01:32 CEST] <cone-207> ffmpeg 03Dale Curtis 07master:e7080d1fe29b: avformat/mov: Correct opus-in-mp4 pre-skip to be uint16_t versus int16_t.
[16:01:33 CEST] <cone-207> ffmpeg 03hwren 07master:b8da7ba5ac99: lavc/extract_extradata_bsf.c: add AVS2
[16:46:27 CEST] <durandal_1707> Who is Dav1d ?
[17:18:43 CEST] <Gramner> dav1d is dav1d
[17:22:07 CEST] <j-b> Gramner: :D
[17:31:07 CEST] <durandal_1707> dav1d is j-b super secret then, he do not want to share it with anyone except himself
[17:31:31 CEST] <j-b> durandal_1707: just wait a couple of weeks and come to VDD
[17:32:00 CEST] <durandal_1707> only when VDD come to me
[17:35:36 CEST] <j-b> Where are you?
[17:35:45 CEST] <durandal_1707> dav1d = b1veb
[17:35:46 CEST] <atomnuker> you still have 2 days left to register
[17:36:47 CEST] <durandal_1707> am against registering
[17:37:20 CEST] <j-b> necause?
[17:38:07 CEST] <durandal_1707> i'm not gonna share my personal info
[17:38:15 CEST] <j-b> personal info?
[17:38:15 CEST] <JEEB> the most you are sharing is your name
[17:38:25 CEST] <JEEB> and that's kind of required wrt venues
[17:38:36 CEST] <j-b> You don't even need to share your name, if you really need that.
[17:38:55 CEST] <j-b> I can work-around it for one person.
[17:39:37 CEST] <atomnuker> you can always say you're elvis presley too
[17:40:35 CEST] <JEEB> durandal_1707: in any case, whatever negative thoughts you have about VDD, try to put them aside. People I've met at VDD have been nice unrelated to any internet personas
[17:41:10 CEST] <durandal_1707> i'm awaiting VDD to come to me, because I can not move out
[17:41:54 CEST] <gnafu> atomnuker: {:-]
[17:41:56 CEST] <atomnuker> its a 2 hour flight from where you are, that's nothing, I've been on a 14 hour flight a day after I did an 8 hour flight
[17:42:33 CEST] <durandal_1707> atomnuker: some super secret agency is looking for me
[17:44:24 CEST] <atomnuker> disguise yourself, a long brown coat, a fedora and glasses with a nose and a mustache
[17:45:04 CEST] <durandal_1707> you just described me!!
[17:46:22 CEST] <BradleyS> i totally read that "long brown cat"
[17:46:38 CEST] <atomnuker> ah, well, its easy then, put on a shiny white suit, pomade on your hair, then you'll pass for elvis presley
[17:47:21 CEST] <atomnuker> agents won't be looking for you because they don't look for dead people, conspiracy theorists will however, but they're friendly, only take pictures for underground newspapers
[17:47:31 CEST] <durandal_1707> ok, that one is resolved, but I'm scared about flying
[17:48:47 CEST] <atomnuker> you've never flown before? issue solved, you get to fly twice and see what its like!
[17:50:10 CEST] <durandal_1707> no, i can not go there without parachute
[17:52:00 CEST] <atomnuker> there are parachutes, they're in the cargo hold, I've seen it in action movies
[17:52:41 CEST] <Gramner> just pull a D. B. Cooper
[17:56:31 CEST] <atomnuker> he never gets on a plane in either series though
[17:57:00 CEST] <atomnuker> oh or do you mean teleportation through electrical sockets?
[17:57:18 CEST] <Gramner> https://en.wikipedia.org/wiki/D._B._Cooper
[17:58:26 CEST] <atomnuker> I thought you meant Dale Cooper, whose middle initial is also B.
[17:59:23 CEST] <durandal_1707> will I sleep alone in room? Paris is veeeeerrry big
[18:03:09 CEST] <atomnuker> usually yes, with someone from videolan, unless you request someone specific
[18:04:01 CEST] <atomnuker> inner paris isn't that big, and you won't be going outside that unless you want to go to disneyland on friday
[18:05:34 CEST] <durandal_1707> will there be live stream?
[18:06:44 CEST] <Gramner> the community event is a visit paris competition, not disneyland, according to the videolan vdd page
[18:13:26 CEST] <atomnuker> oic, I don't go to them as I always have to work my slides
[18:14:16 CEST] <atomnuker> durandal_1707: don't count on it, don't count on the videos being uploaded for around 8 months either
[18:14:26 CEST] <durandal_1707> atomnuker: you are little lazy, work on slides days before and then socialize a litle
[18:15:10 CEST] <atomnuker> no way, all pros do their slides from scratch the night before
[18:15:25 CEST] <durandal_1707> atomnuker: what kind of unprofesionalism is that? amateurs, give me live stream!
[18:16:53 CEST] <durandal_1707> i will go to NTTW X when it being held in Slovenia or Italy or somewhere even closer, they have live streaming too
[18:17:01 CEST] <atomnuker> its not a big marketing event, its a technical unconference
[18:17:05 CEST] <Gramner> the pro way is to create the next slide while presenting the current slide
[18:17:30 CEST] <durandal_1707> that is called the improvisations
[19:25:53 CEST] <JEEB> btw, what's a nice thing to snapshot memory usage from a running process?
[19:26:32 CEST] <JEEB> if I have a long-running process and I'd like to take snapshots of its memory usage every hour or so
[19:27:02 CEST] <JEEB> because I think there's a memory leak somewhere, but if the process exits it might be all released properly
[19:27:09 CEST] <kurosu_> cat /proc/$PID/stats or something ?
[19:27:25 CEST] <JEEB> I think that doesn't show whereabouts the memory usage is
[19:27:39 CEST] <JEEB> I think someone at some point noted gprof
[19:27:50 CEST] <kurosu_> *status
[19:27:51 CEST] <JEEB> but I think that might be a full all-the-way dumping
[19:28:08 CEST] <JEEB> instead of "let's see where memory is used at time points X,Y,Z"
[19:28:11 CEST] <kurosu_> how you mean like profiling inside of the process ?
[19:28:20 CEST] <kurosu_> *oh
[19:28:35 CEST] <JEEB> yea, I'd like to figure out where lavf/lavc/lavfi is buffering more and more stuff with time
[19:28:44 CEST] <kurosu_> I think I used valgrind's massif, but that's valgrind, and I forgot how since then
[19:29:09 CEST] <JEEB> ok, I'll have to see if I can get valgrind be fast enough
[19:29:41 CEST] <JEEB> unless I dump gigabytes worth of the source
[19:30:00 CEST] <JEEB> but that'd be rather long to keep it rolling for enough time
[19:30:06 CEST] <JEEB> s/long/large/
[19:30:16 CEST] <JEEB> so being able to run it fast enough with the realtime input sounds like a better idea :D
[19:30:27 CEST] <JEEB> since it seems to constantly replicate
[19:30:40 CEST] <JEEB> and it's specific to certain inputs
[19:30:48 CEST] <JEEB> not all of them lead to such memory usage gains
[19:30:58 CEST] <kurosu_> Or you move the av_ alloc functions to macros that log big enough ones using __line__ and stuff
[19:31:24 CEST] <kurosu_> though you don't know when that gets deallocated
[19:31:32 CEST] <atomnuker> ^^that's a good idea
[19:32:09 CEST] <atomnuker> you can keep track of frees by having a global array store the pointers, and making sure to have a global lock too
[19:32:13 CEST] <kurosu_> maybe alloc eg a bigger area to store interesting info there
[19:32:32 CEST] <kurosu_> at the start of the area, like old unaligned allocs were doing
[19:35:01 CEST] <kurosu_> I remember using massif because of a C++ app that didn't get its dangling references, local copies and such
[19:35:37 CEST] <kurosu_> like "how let's put that image in that container" - ah I just duplicated it
[19:43:13 CEST] <JEEB> I'll have to try massif
[22:22:45 CEST] <rcombs> hmmmmm
[22:23:04 CEST] <rcombs> I think if an error occurs during an avio_read, but there's some data buffered, that data will get returned rather than the error code, right?
[22:23:14 CEST] <rcombs> potentially resulting in a short read
[22:23:26 CEST] <rcombs> mpegts.c seems to assume that all short reads mean EOF
[22:23:28 CEST] <rcombs> who's wrong
[22:25:00 CEST] <JEEB> hmm
[22:26:08 CEST] <rcombs> (got a log + pcap indicating a read timeout in a chunked HTTP stream was treated as an EOF and resulted in ffmpeg.c exiting with status code 0)
[22:34:09 CEST] <durandal_1707> michaelni: let's ban Compn he is ignoring me
[22:35:55 CEST] <Compn> i posted patch durandal_1707
[22:35:58 CEST] <Compn> to -dev-eng
[22:36:01 CEST] <Compn> what out fmt is it
[22:36:49 CEST] <durandal_1707> Compn: same as for y41p
[22:36:59 CEST] <Compn> out Y41P ?
[22:37:17 CEST] <durandal_1707> Compn: oh, noo, its yuv420p
[22:37:21 CEST] <Compn> lol
[22:37:37 CEST] <Compn> k
[22:37:38 CEST] <durandal_1707> i thinked about BT20
[22:37:55 CEST] <Compn> durandal_1707 : why imm4 not imm5 and imm6 at same time? :)
[22:37:59 CEST] <durandal_1707> IMM4 is yuv420p
[22:38:11 CEST] <Compn> working on imm5 / imm6 ?
[22:38:17 CEST] <durandal_1707> no
[22:38:33 CEST] <Compn> binary decoder still needed.
[22:39:26 CEST] <Compn> durandal_1707 : write guide for reverse engineering codecs? 
[22:39:46 CEST] <cone-785> ffmpeg 03James Almer 07master:35c84ae834de: Revert "avcodec/opus_parser: Handle complete frames flag."
[22:40:48 CEST] <durandal_1707> Compn: guide? more like book of 3k pages
[22:42:34 CEST] <Compn> yes write it
[22:43:38 CEST] <durandal_1707> once mplayer lose all users
[22:45:36 CEST] <durandal_1707> also there are no imm6 samples iirc
[23:28:08 CEST] <durandal_1707> who gave Carl so many different computers?
[23:31:24 CEST] <atomnuker> the results are good so its alright
[23:31:52 CEST] <atomnuker> I'd rather have the new reader used everywhere, but I'd be happy with it being x86_64 only
[23:32:11 CEST] <atomnuker> I think the cache size is adjustable, not sure if that would help 32 bits
[23:36:55 CEST] <durandal_1707> atomnuker: what you use to measure ffv2 visual performance?
[23:37:59 CEST] <j-b> ffv2!
[23:40:58 CEST] <rcombs> alright sent a patch (and also a couple others for HEVC bsf stuff)
[23:43:17 CEST] <atomnuker> durandal_1707: erm illustrations and pictures of birds over sky/water, because artifacts around sharp edges are easy to see
[00:00:00 CEST] --- Fri Aug 24 2018

More information about the Ffmpeg-devel-irc mailing list