[Ffmpeg-devel-irc] ffmpeg-devel.log.20180107

burek burek021 at gmail.com
Mon Jan 8 03:05:04 EET 2018


[00:31:16 CET] <cone-529> ffmpeg 03Aman Gupta 07master:cc90ee0b7e47: Changelog: add VideoToolbox HEVC support
[00:57:32 CET] <klaxa> atomnuker: >Invalid return value 0 for stream protocol and then i get a double free(), looks like some API change caught up to me
[01:09:12 CET] <atomnuker> I think that's just a warning
[01:23:02 CET] <wm4> you need to return AVERROR_EOF instead of 0, but that shouldn't make it crash, at worst you get an endless loop somewhere
[01:33:56 CET] <klaxa> okay, there must be a bug somewhere else as well then, also do i just check the return value of my *read_packet() function so basically: return buf_size ? buf_size : AVERROR_EOF?
[01:35:56 CET] <klaxa> maybe the examples should be updated as well then
[01:41:45 CET] <wm4> yeah the examples are in a permanent state of brokenness
[02:27:17 CET] <cone-529> ffmpeg 03James Almer 07master:6ec91319204b: fate: remove ffm reference files
[02:27:18 CET] <cone-529> ffmpeg 03James Almer 07master:9e68f472ec5d: configure: remove check for SA_RESTART
[04:10:58 CET] <cone-529> ffmpeg 03KO Myung-Hun 07master:d03c39b46b21: configure: check SDL2 function with a header
[04:15:05 CET] <cone-529> ffmpeg 03KO Myung-Hun 07master:3c0a081a1e9f: compat/os2threads: support static mutexes
[04:48:54 CET] <cone-529> ffmpeg 03Michael Niedermayer 07master:476665d4de98: avcodec/h264_slice: Do not attempt to render into frames already output
[09:02:35 CET] <JEEB> hah, every time I build with clang I get reminded of
[09:02:35 CET] <JEEB> src/libavcodec/hevc_cabac.c:37:21: warning: variable 'num_bins_in_se' is not needed and will not be emitted [-Wunneeded-internal-declaration]
[09:02:39 CET] <JEEB> static const int8_t num_bins_in_se[] = {
[09:02:46 CET] <JEEB> maybe I should finally post a patch to remove that
[09:03:06 CET] <JEEB> unless that is needed for something that's not yet implemented
[12:44:32 CET] <durandal_1707> when will we convert to c99?
[13:27:51 CET] <atomnuker> what do you need from c99?
[13:36:08 CET] <kierank> for( int
[13:37:03 CET] <atomnuker> except that
[13:37:30 CET] <atomnuker> I personally would like c11 generics
[13:40:16 CET] <durandal_1707> atomnuker: complex.h
[13:41:58 CET] <atomnuker> go ahead and use it, I don't think there's an issue, its a c99 requirement
[13:43:40 CET] <atomnuker> though you can't really replace it globally, we do plenty of things which don't exactly map to complex multiplies using fftcomplex
[13:44:42 CET] <nevcairiel> complex.h is not supported on all compilers we support
[13:48:23 CET] <jkqxz> It's technically a C99 requirement, but was never implemented everywhere.  It's optional in C11.
[13:48:59 CET] <atomnuker> what? no
[13:49:14 CET] <atomnuker> complex.h along with tgmath was widely implemented
[13:49:38 CET] <atomnuker> didn't know they made it optional though
[13:49:42 CET] <jkqxz> "Implementations that define the macro _ _STDC_NO_COMPLEX_ _ need not provide this header nor support any of its facilities."
[13:49:43 CET] <atomnuker> well that sucks
[13:50:08 CET] <nevcairiel> things that became optional in c11 might as well not have existed =p
[13:50:22 CET] <jkqxz> Like __STDC_NO_THREADS__...
[13:52:46 CET] <jkqxz> And __STDC_NO_VLA__, of course.
[13:55:04 CET] <atomnuker> so they made 1 useless feature optional, made 1 useful feature optional and added 1 useless feature
[13:55:36 CET] <atomnuker> karma=neutral
[13:56:25 CET] <atomnuker> well and added 1 useful mandatory feature
[14:07:50 CET] <atomnuker> actually 2, forgot about the atomics
[14:09:00 CET] <atomnuker> nevermind, that too is optional
[14:10:35 CET] <jdarnley> What is the 1 useful mandatory feature?
[14:12:07 CET] <atomnuker> the _Generic keyword
[14:16:26 CET] <BBB> arent there fallback implementations we can use for sucky compilers?
[14:16:34 CET] <BBB> we did that for stdatomic.h also, no?
[14:17:38 CET] <nevcairiel> the stdatomic one is already vastly inadequate, more fancy features would be even worse
[14:21:17 CET] <BBB> yeah, life sucks that way
[14:24:46 CET] <wm4> _Generic is way less useful than typeof or ({ ... }) would have been
[14:24:53 CET] <wm4> seriously fuck the C committee
[14:30:05 CET] <atomnuker> ^^
[14:31:24 CET] <atomnuker> we could use julia I suppose
[14:32:09 CET] <atomnuker> its basically C (abi and api) but with threads, complex, and loads of math functions
[14:32:53 CET] <wm4> and dynamic typing
[14:40:59 CET] <atomnuker> apparently that's optional
[14:41:14 CET] <atomnuker> (see, its just like c)
[15:28:46 CET] <Chloe> atomnuker: if we're rewriting in a different language clearly we should use Rust
[15:37:30 CET] <wm4> is AVX 512 actually useful
[15:42:31 CET] <BBB> wm4: yes, parts of it are
[15:42:45 CET] <BBB> kierank did a blog post on useful aspects of it
[15:43:06 CET] <BBB> whether zmm is useful ATM is questionable, it probably is but possibly only for a few limited use cases
[15:43:15 CET] <BBB> but the new instructions are very useful
[15:43:26 CET] <BBB> and I think zmm can help in some ways
[15:43:43 CET] <wm4> what I don't like is adding overhead of at least 32 bytes per packet on all platforms
[15:44:05 CET] <wm4> even if AVX 512 is useful, it's kind of obscure
[15:44:21 CET] <BBB> itd be nice if that was platform-dependent, yes
[15:44:24 CET] <BBB> but right now it is not
[15:44:25 CET] <BBB> ...
[15:44:28 CET] <BBB> brb
[15:46:00 CET] <kierank> BBB: was jdarnley who wrote that
[15:47:40 CET] <jdarnley> Now is a good a time as any to make all padding and alignement use the avutil function
[17:06:05 CET] <cone-217> ffmpeg 03Paul B Mahol 07master:6c65de3db06c: avfilter/af_aiir: add support for alternative coefficients format
[19:06:12 CET] <durandal_170> atomnuker: for removing ffserver you need to close all ffserver bugs
[19:12:29 CET] <RiCON> there's quite a few issues
[19:20:45 CET] <BBB> kierank: youre right, sorry about that
[19:51:53 CET] <atomnuker> "Last modified 6 years ago" <- someone's in for a blast from the past
[19:55:21 CET] <wm4> do we have some official replacement suggestions for ffserver?
[19:55:58 CET] <atomnuker> klaxa's thing when it works better
[19:57:26 CET] <wm4> atomnuker: what does klaxa's "thing" do?
[19:57:50 CET] <philipl_> Is Carl on holiday this weekend? I have to assume there'll be a show tomorrow.
[19:58:31 CET] <atomnuker> wm4: you feed it matroska and it streams it out to multiple clients over http
[19:59:37 CET] <atomnuker> (I know matroska over http is bad but it ought to be better than ffm)
[20:01:11 CET] <wm4> atomnuker: isn't that a web server
[20:01:15 CET] <durandal_170> atomnuker: would you write fft for doubles for me?
[20:01:59 CET] <atomnuker> why use doubles?
[20:06:04 CET] <wm4> and I think streaming mkv isn't too bad
[20:06:29 CET] <wm4> better than taking an inherently unstreamable format and then hacking it into small pieces to make it streamable
[20:07:47 CET] <wm4> sfan5: git send-email would have been simpler
[20:08:48 CET] <atomnuker>  /end of spam
[20:16:49 CET] <durandal_170> atomnuker: more bits, better quality
[20:17:31 CET] <atomnuker> floats are already overkill for audio
[20:19:43 CET] <durandal_170> atomnuker: nope
[20:20:48 CET] <durandal_170> michaelni: iirc rgb48 ffv1 files are already created by arhivist folks
[20:24:11 CET] <atomnuker> durandal_170: 16 bits are enough to map to every single volume perceivable
[20:24:30 CET] <atomnuker> you have 16 bits of headroom on top of that
[20:24:36 CET] <atomnuker> there's no way that's not enough
[20:24:43 CET] <atomnuker> if its not enough you're doing something wrong
[20:24:58 CET] <durandal_170> atomnuker: think about math and 24bits
[20:25:44 CET] <atomnuker> if you're running out of precision at the low range consider multiplying
[20:30:03 CET] <FishPencil> For some reason 'git clone --depth 1 https://git.ffmpeg.org/ffmpeg.git' fails but 'git clone --depth 1 https://github.com/FFmpeg/FFmpeg' works
[20:31:00 CET] <durandal_170> why you use that?
[20:32:07 CET] <FishPencil> durandal_170: because sometimes you don't need the entire history?
[20:32:08 CET] <wm4> FishPencil: I had problems with it too, --ipv4 works better
[20:32:47 CET] <klaxa> i've been looking into the double free i get (with valgrind it doesn't happen) and it seems the avio api changed since i implemented it? still investigating, this didn't happen a few months ago :P
[20:33:01 CET] <klaxa> since i implemented my server i mean
[20:34:41 CET] <FishPencil> Wow, Ubuntu ships with 2.7 git so I can't even use --ipv4
[21:02:33 CET] <jdarnley> Don't you love Debian, its children, and the software they provide?
[21:03:55 CET] <RiCON> jamrial: d03c39b46b2 broke --enable-sdl2 in mingw at least: https://i.fsbn.eu/ohwq.txt
[21:04:08 CET] <RiCON> the -Dmain=SDL_main probably needs to be removed
[21:06:35 CET] <jamrial> RiCON: did this start after KO Myung-Hun's commit?
[21:08:45 CET] <RiCON> i'll try reverting it myself, someone said that would fix it
[21:12:05 CET] <atomnuker> jdarnley: I do, I run debian unstable and I have git 2.16-rc0
[21:14:31 CET] <RiCON> jamrial: yeah, reverting it fixes the check
[21:23:06 CET] <FishPencil> Is there a command to quickly see where something is defined in a source tree? Specifically looking for AVFilterPad, and 'find . -name "*.h" -exec grep -Hi AVFilterPad {} \;' sort of works
[21:23:47 CET] <cone-217> ffmpeg 03Paul B Mahol 07master:2d3df8e2e9e6: avfilter/af_aiir: rename options, provide gains in separate option
[21:27:49 CET] <durandal_170> now to figure how to convert random iir to cascaded biquads
[21:30:48 CET] <RiCON> jamrial: i think filtering out -Dmain=SDL_main and adding -DSDL_MAIN_HANDLED fixes it too
[21:31:14 CET] <RiCON> it stops SDL_main.h from defining main as SDL_main again
[21:31:38 CET] <jamrial> RiCON: -Dmain=SDL_main is added by both the pkg-config file and sdl2-config.sh
[21:31:45 CET] <RiCON> yeah
[21:31:50 CET] <RiCON> but it breaks configure's checks
[21:32:00 CET] <RiCON> since they use main(void)
[21:32:11 CET] <jamrial> i filter it at the end of configure so it doesn't pollute cflags, but it didn't do anything before that commit with these checks
[21:32:20 CET] <jamrial> I'm looking for a cleaner alternative
[21:32:25 CET] <RiCON> because we didn't include SDL.h before
[21:34:25 CET] <jamrial> RiCON: how about this? https://pastebin.com/XZ9nuwSY
[21:35:47 CET] <jamrial> i'm not sure why the SDL_Init check was added after the SDL_PollEvent one in the pkg-config path
[21:36:02 CET] <jamrial> maybe it was needed back when we supported sdl1, for some weird environment
[21:44:20 CET] <RiCON> jamrial: yeah, that worked
[23:10:49 CET] <kierank> durandal_170: when did you learn so much about filters
[23:14:59 CET] <durandal_170> kierank: which filters?
[23:15:08 CET] <kierank> iir and fir
[23:15:58 CET] <durandal_170> self taught, besides its just math
[23:22:11 CET] <FishPencil> What's the simplest well written video filter to read and use as a reference?
[23:23:50 CET] <durandal_170> FishPencil: null filter, crop/pad
[23:31:29 CET] <FishPencil> durandal_170: thanks
[23:45:06 CET] <Compn> kierank : you could pay durandal_170 to write filters for you :P
[23:45:19 CET] Action: Compn ducks
[00:00:00 CET] --- Mon Jan  8 2018


More information about the Ffmpeg-devel-irc mailing list