[Ffmpeg-devel-irc] ffmpeg-devel.log.20160809

burek burek021 at gmail.com
Wed Aug 10 03:05:02 EEST 2016


[00:08:43 CEST] <iive> i don't see patch for that commit in the maillist
[00:09:32 CEST] <iive> so this cannot be old patch that have been committed by somebody else.
[00:11:11 CEST] <durandal_1707> your point being?
[00:11:24 CEST] <llogan> he simply wants to know where the patch came from
[00:15:24 CEST] <durandal_1707> live mplayer fork irc channel
[00:18:05 CEST] <iive> the point is, that this commit seems to be directly from Derek and he had left ffmpeg.
[00:18:54 CEST] <nevcairiel> git records who pushed a commit separate from who authored it, you should check it out
[00:19:34 CEST] <durandal_1707> If Elvis can why can't Derek?
[00:20:05 CEST] <iive> durandal_1707: never heard of fork named live mplayer before.
[00:20:31 CEST] <durandal_1707> its #mpv
[00:21:38 CEST] <iive> so you are the committer. fine.
[00:23:39 CEST] <iive> nevcairiel: has that always been the case. I remember having issue with finding out the committer, some years ago.
[00:24:47 CEST] <nevcairiel> ever since I used git, so at least a couple years now, cant speak for "always"
[00:25:01 CEST] <iive> :)
[00:25:14 CEST] <durandal_1707> svn days?
[00:38:45 CEST] <iive> svn doesn't have an author, only committer. And it seems that both fields are present in initial git versions.
[00:38:56 CEST] <iive> well, my bad.
[01:44:49 CEST] <jya> is there a way with ffmpeg to remux a mp4 and either add a proper edit list or rework the pts so that the first sample starts at 0 ?
[01:55:55 CEST] <llogan> that's a question for #ffmpeg
[02:03:34 CEST] <jya> llogan: thought that one day someone will interject that :)
[02:18:42 CEST] <cone-371> ffmpeg 03Derek Buitenhuis 07n3.1.2:HEAD: docs/filters: Fix parameter names for colorspace filter
[04:19:15 CEST] <hpj_> hi, guys! since it doesn't seem to have been reported in the irc channels and the mailing list, i'm here to break the bad news: one of the ffmpeg contributors died last week
[04:22:11 CEST] <hpj_> LEI Xiaohua, doctoral candidate at the Communications University of China in beijing, age 25, died on 4 august on the university campus
[04:25:51 CEST] <hpj_> he had worked through the previous night and was found dead in the morning near the elevator in the lab building
[04:26:58 CEST] <hpj_> cause of death was presumably exhaustion, no suspicion of foul play, he had been generally healthy and even a half-marathon runner
[04:29:27 CEST] <hpj_> despite his youth, lei had established himself as a highly respected expert in the chinese language world of a/v developers, writing numerous tutorials and always eager to help other developers with their problems
[04:32:48 CEST] <hpj_> his award-winning blog on a/v development has 4.5 million visits and is considered a reference guide in the chinese a/v devel world
[04:33:31 CEST] <hpj_> here is a newspaper report on this death (in chinese): http://edu.qq.com/a/20160804/003711.htm
[09:29:46 CEST] <Timothy_Gu> hpj_: Not to be insensitive, but I don't see his name mentioned in the log of any FFmpeg commits. Nevertheless, I just read some of the blog posts on FFmpeg and those were helpful indeed, especially for people whose Chinese is better than English. As a compatriot and another contributor to FFmpeg, I'd like to think he is now in a better place.
[09:45:11 CEST] <hpj_> yes, i guess his work was more around ffmpeg than within the project itself: http://leixiaohua1020.github.io/#ffmpeg-development-examples
[10:52:27 CEST] <Compn> hpj_ : wow , thats crazy. although unexpected for a 25 year old to die from exhaustion doing computer work
[10:52:36 CEST] <Compn> thanks for sharing
[11:39:08 CEST] <hpj_> Compn: in fact, there have been a series of high-profile sudden deaths (heart attack and alike) in the chinese it industry in recent times
[11:42:37 CEST] <hpj_> not to mention cases of sudden deaths of chinese gamers every other day, often having sat in internet cafes for days on end (chinese gamers often prefer to play in internet cafes instead of at home because of the low latency internet connection there)
[12:19:59 CEST] <Compn> hpj_ : i'm aware that this happens in china. but i dont think its common occurance here in usa.
[12:20:41 CEST] <Compn> cultural differences in dealing with long working hours and exhaustion
[12:36:04 CEST] <cone-118> ffmpeg 03Paul B Mahol 07master:011837285ffc: doc/filters: add one more lutyuv example
[12:46:35 CEST] <cone-118> ffmpeg 03Michael Niedermayer 07master:369ed11e3c8a: avfilter/drawutils: Fix single plane with alpha
[13:41:50 CEST] <ubitux> we are doing something bad in drawtext
[13:42:07 CEST] <ubitux> fenv.h ’ use of feclearexcept()/fetestexcept()
[13:42:17 CEST] <ubitux> unless i'm mistaken, these are not thread safe
[13:42:33 CEST] <ubitux> this was introduced in dec87454
[13:43:04 CEST] <ubitux> it's the only place where we use such include
[13:43:18 CEST] <ubitux> andrey_utkin ^
[13:46:44 CEST] <rcombs> it's thread-safe as of C11, but we're not that
[13:46:45 CEST] <ubitux> or maybe they actually are
[13:47:01 CEST] <rcombs> probably thread-safe on real implementations but& still yucky
[13:47:25 CEST] <ubitux> does "thread-safe" implies "not global state"?
[13:47:34 CEST] <andrey_utkin> ubitux, i wonder what we could do with that. seems we can only drop this code - just to convert silently and hope for better
[13:48:07 CEST] <ubitux> that's what we do everywhere, isn't it?
[13:48:16 CEST] <rcombs> ubitux: as of C11/C++11 each thread must maintain its own floating-point exception mask
[13:48:31 CEST] <andrey_utkin> i don't remember clearly why i added it. either my own error-handling paranoia, or i was pointed to this during patch review
[13:48:46 CEST] <rcombs> previously it's unspecified, and probably worked the same way since iirc that's what x86 and ARM actually do
[13:49:25 CEST] <andrey_utkin> ubitux, i think since you've found this you want to send the patch yourself?
[13:49:38 CEST] <rcombs> why not just check if it's > INT_MAX or < INT_MIN though?
[13:49:45 CEST] <rcombs> (or whatever the macros for the relevant type are)
[13:50:03 CEST] <andrey_utkin> what "it" (in "it's")?
[13:50:07 CEST] <rcombs> plus isnan I guess
[13:50:21 CEST] <rcombs> these checks are around an assignment from a float to an integer
[13:50:40 CEST] <rcombs> double->int, now that I check
[13:50:53 CEST] <rcombs> should be fine to just check if it's out of range or NaN, yeah?
[13:51:04 CEST] <ubitux> andrey_utkin: i wasn't planing to send a patch :p
[13:51:17 CEST] <ubitux> just reading stuff about fp and came accross this code in ffmpeg
[13:51:42 CEST] <andrey_utkin> i think checks on range of INT_MAX/MIN are pointless because they apparently involve same conversion as the present one
[13:52:12 CEST] <andrey_utkin> check on nan sounds reasnable
[13:55:23 CEST] <rcombs> andrey_utkin: the comparison would be done between doubles
[13:56:22 CEST] <andrey_utkin> rcombs, in "if (res < INT_MAX)" and such?
[13:56:39 CEST] <rcombs> yeah, INT_MAX would be promoted to double there, no?
[13:57:24 CEST] <andrey_utkin> i thought otherwise for no reason, need to check
[13:58:58 CEST] <andrey_utkin> according to this, you rcombs are right https://www.eskimo.com/~scs/cclass/int/sx4cb.html
[14:01:01 CEST] <rcombs> also that check has more problems
[14:01:38 CEST] Action: rcombs double-checks something
[14:04:05 CEST] <rcombs> OK so that'd be CVTTSD2SI
[14:04:21 CEST] <rcombs> the check looks for FE_INVALID|FE_OVERFLOW|FE_UNDERFLOW
[14:04:42 CEST] <rcombs> OVERFLOW and UNDERFLOW are impossible here
[14:04:48 CEST] <rcombs> but FE_INEXACT is possible and not checked for
[14:05:55 CEST] <rcombs> (maybe it's different on x87? I dunno)
[14:11:47 CEST] <rcombs> and even if this was all technically correct and thread-safe, it's pretty rude to change the floating-point exception mask in library code unless you're on a dedicated thread
[14:28:39 CEST] <andrey_utkin> thanks for pointing this all out, i've put this on my schedule and hopefully i'll sanitize this soon. ubitux, rcombs - please give your emails so that i CC you when i send a patcg
[14:29:00 CEST] <rcombs> we're on the list anyway
[14:29:06 CEST] <ubitux> no need to cc :)
[16:57:15 CEST] <durandal_1707> can I get comments for fillgaps filter?
[19:54:33 CEST] <coralisk> rekt
[22:00:40 CEST] <jamrial_> for whoever wants to try converting ffmpeg to codecpar, http://pastebin.com/YKcEFcQ4 combines all five relevant commits that were noop'd during merges
[22:02:12 CEST] <jamrial_> should hopefully be more helpful than looking only at the first one, which still kept some avstream->codec usage in place
[22:03:51 CEST] <durandal_1707> whats so hard about it?
[22:06:18 CEST] <jamrial_> last i checked there were some format specific hacks that used avctx fields not present in codecpar
[22:52:02 CEST] <rcombs> I'd still like some way to export arbitrary extra fields from decoders or parsers
[22:52:37 CEST] <rcombs> that gets exposed on the AVStream (so you can get it out of find_stream_info instead of running the decoder yourself)
[23:05:12 CEST] <j-b> 1m
[00:00:00 CEST] --- Wed Aug 10 2016


More information about the Ffmpeg-devel-irc mailing list