[Ffmpeg-devel-irc] ffmpeg-devel.log.20190420

[burek]

[00:14:25 CEST] <cone-932> ffmpeg 03Carl Eugen Hoyos 07master:48860df34df6: configure: Add .exe suffix to toolchain calls.
[00:17:57 CEST] <cehoyos> Any comments about -Wno-gnu-variable-sized-type-not-at-end? http://ffmpeg.org/pipermail/ffmpeg-devel/2019-March/241729.html
[00:19:07 CEST] <cehoyos> What are win32 arm/arm64 devices? Compilation works fine but I wonder what to do with the binaries...
[00:20:16 CEST] <nevcairiel> technically you can run Windows 10 IoT on a variety of ARM devices, including a Pi3
[00:20:40 CEST] <cone-932> ffmpeg 03Carl Eugen Hoyos 07master:93209902ede3: lavfi/fspp: Simplify a macro.
[00:20:47 CEST] <nevcairiel> not sure if any are officially sold yet
[00:21:11 CEST] <cehoyos> I see.
[00:21:20 CEST] <cehoyos> Did you look again into the aac vs2019 issue?
[00:21:42 CEST] <nevcairiel> not yet, probably not before next week
[00:22:45 CEST] <cehoyos> In-tree compilation works now without changes, I will look into mslink next week to see if I can fix out-of-tree builds.
[00:22:56 CEST] <cehoyos> mklink
[00:25:55 CEST] <cehoyos> Matebook E 2019 is apparently a real-life arm Windows device
[00:27:00 CEST] <nevcairiel> wbs has been doing a lot of work with windows on arm if there is any problems
[00:28:47 CEST] <cehoyos> I had to realize that gas-preprocessor is needed, then it worked fine
[00:29:12 CEST] <cehoyos> I know now how to test future arm64 patches for compilation issues.
[01:28:17 CEST] <cone-932> ffmpeg 03Jun Zhao 07master:d93e44332f1b: lavf: bump version/add APIchanges entry when cleanup applehttp
[05:21:06 CEST] <hanetzer> question. should I be using https://github.com/FFmpeg/FFmpeg.git or git://source.ffmpeg.org/ffmpeg.git ?
[05:34:32 CEST] <kierank> the latter
[05:34:35 CEST] <kierank> the former is a mirror
[05:34:38 CEST] <kierank> it may not be up to date
[05:34:47 CEST] <kierank> likely it will be but i have seen it be delayed before
[06:46:19 CEST] <hanetzer> kierank: gotcha. looking into the possibility of porting ffmpeg to meson as 'exercise' and it may be useful to you guys as well :)
[07:16:38 CEST] <cone-553> ffmpeg 03Gyan Doshi 07master:3153a6502a28: avcodec: add AV_CODEC_FLAG_DROPCHANGED to flags
[07:19:09 CEST] <cone-553> ffmpeg 03Gyan Doshi 07master:3a07aec82741: doc/APIchanges: update for 3153a6502a
[08:21:15 CEST] <nevcairiel> hanetzer: someone already did that once
[08:21:27 CEST] <hanetzer> orly. 
[08:21:37 CEST] <nevcairiel> i forgot where its at
[11:06:18 CEST] <durandal_1707> ubitux: i did .csp for lut3d/lut1d on ML
[11:16:20 CEST] <ubitux> looks fine at first glance, i'll send an official answer soon
[11:16:53 CEST] <ubitux> can you give me some context on where it comes from and why the sudden interest in it?
[11:20:15 CEST] <durandal_1707> ubitux: i work for secret agency and they work with bunch of cineSpace LUTs, on serios side I just found it exists
[11:20:53 CEST] <ubitux> ok :)
[11:24:25 CEST] <J_Darnley> hanetzer: look in the ML archives if you want to see when it was last discussed
[11:24:46 CEST] <J_Darnley> June and July 2018 http://ffmpeg.org/pipermail/ffmpeg-devel/2018-June/thread.html http://ffmpeg.org/pipermail/ffmpeg-devel/2018-July/thread.html
[11:28:53 CEST] <BtbN> don't use git:// urls, ever. They are unsafe.
[11:29:30 CEST] <BtbN> I also don't think a switch of build-systems would be possible to get accepted.
[11:29:44 CEST] <BtbN> Unless you somehow manage to make it 100% regression free, which seems impossible.
[11:29:45 CEST] <hanetzer> BtbN: yeah. I was more pointing at the domain/etc; I believe some portion of the documentation mentions the git:// url
[11:33:52 CEST] <hanetzer> BtbN: elaborate the regression free bit?
[11:34:13 CEST] <BtbN> Someone on the ML will find an edge case the new build system breaks, and veto it on that basis.
[11:34:48 CEST] <hanetzer> gotcha. so until its 1:1 feature parity that's a no until it is
[11:35:04 CEST] <BtbN> Even then I'd suspect people will give it a hard time
[11:35:19 CEST] <BtbN> Cause, the bunch of script we have right now do work surprisingly fine
[11:36:32 CEST] <J_Darnley> I too think it is unlikely to get accepted wih anything less then full feature parity
[11:36:41 CEST] <J_Darnley> *than
[11:36:53 CEST] <BtbN> I also still don't feel like meson is mature enough
[11:37:13 CEST] <BtbN> Every time I use it I end up in a situation where I have to manually install a sometimes project-specific version of it
[11:37:54 CEST] <hanetzer> expected. I did the mesonbuild port for mate-desktop and there is currently some grumbling about ninja uninstall 'erroring' because it tries to remove icon files twice, but honestly irl make install and the equivalent to the root filesystem is rather uncommon
[11:39:25 CEST] <J_Darnley> I think meson looks like one of the configure & make relacements with promise
[11:40:03 CEST] <BtbN> But why does ffmpeg configure need replaced?
[11:40:27 CEST] <hanetzer> BtbN: do me a favor; got a clone? assume you do. time ./configure
[11:41:04 CEST] <J_Darnley> I don't really think it does, especially not since some genius fixed the massive slow-down that crept in.
[11:42:16 CEST] <J_Darnley> But as I'd like to say to people who say "why don't you write it in X?": please go ahead and do that
[11:42:46 CEST] <hanetzer> J_Darnley: found the stuff you were talking about; https://gitlab.freedesktop.org/gstreamer/meson-ports/ffmpeg << gonna give this a go
[11:44:12 CEST] <nevcairiel> configure time complaints are such an overrated problem, especially since the big slowdown from recent years has been fixed
[11:44:26 CEST] <nevcairiel> so it takes 30 second to configure, big deal
[11:44:51 CEST] <JEEB> yea, I stopped building on windows when it was 10min, but since that has dropped dramatically
[11:46:07 CEST] <J_Darnley> Ah look at that repo.  Good to see gaps being papered over with python scripts.
[11:46:37 CEST] <nevcairiel> that sort of defeats the purpose doesnt it
[11:46:37 CEST] <hanetzer> nevcairiel: that's fair. but then again, the fact that 'some genius' had to fix that issue and not just 'some dude' kinda sucks :)
[11:47:32 CEST] <hanetzer> J_Darnley: directory of python scripts? curious, I am :)
[11:47:53 CEST] <hanetzer> theoretically said gaps could be included into meson proper if needed.
[11:47:54 CEST] <J_Darnley> The root directory!
[11:48:04 CEST] <nevcairiel> there is at least 7 python scripts that fix up a lacking build system
[11:48:39 CEST] <BtbN> And then we end up with the same situation other projects have. A build system that only works with a very specific and/or super bleeding edge meson version
[11:48:40 CEST] <J_Darnley> And "some dude" won't know how to profile a shell script, let alone python.
[11:50:16 CEST] <hanetzer> J_Darnley: ah, that's a place I personally wouldn't put my extra glue so was looking in ./tools :P
[11:50:29 CEST] <nevcairiel> if anything it should've gone into ffbuild
[11:50:42 CEST] <J_Darnley> Sure, a new build system will have some gaps but why would you chose python, aka the worst language
[11:51:01 CEST] <hanetzer> that too, but ffbuild is the ./configure guts right? thought they may want to avoid mixing it up :)
[11:51:06 CEST] <durandal_1707> ubitux: the cineSpace luts supports XxYxZ luts (instead of just YxYxY) is such thing easily added to lut3d?
[11:51:11 CEST] <nevcairiel> they choose meson, meson is in python, so one might assume python is already installed
[11:51:23 CEST] <nevcairiel> ... unless you're on windows and use meson.exe which has python baked in =p
[11:51:39 CEST] <J_Darnley> Yeah but I don't have to see any python code with meson (unlike scons)
[11:51:56 CEST] <BtbN> Python is probably the best choice for build systems right now
[11:52:00 CEST] <hanetzer> nevcairiel: yep. a simple port I'm working on (icon theme, only one source file for the theme engine but its mostly install_data and install_subdir ca
[11:52:03 CEST] <hanetzer> lls)
[11:52:08 CEST] <BtbN> it's relatively fast and powerful, super portable and widely available
[11:52:13 CEST] <nevcairiel> i rather like how i can build eg.  dav1d with meson without needing msys at all
[11:52:39 CEST] <nevcairiel> or python for that matter
[11:52:51 CEST] <hanetzer> the current Makefile solutions I've seen in icon themes are terrible lol, so I'm writing that bit for index.theme generation in python because I know meson users will have python :)
[11:52:55 CEST] <nevcairiel> but its a far simpler project then ffmpeg
[11:53:17 CEST] <J_Darnley> I'd probably like meson when its "finished"
[11:53:26 CEST] <hanetzer> J_Darnley: or at least a 1.x release eh?
[11:53:50 CEST] <BtbN> It would also be entirely possible to just clone what configure does, but writing it in Python instead. Would be a serious performance boost, and if done right, provide the exact same features.
[11:54:16 CEST] <nevcairiel> but outside of possibly performance enhancements, also doesnt offer anything else
[11:54:29 CEST] <nevcairiel> one point of using something like meson is to claim that in the long  run its simpler to maintain
[11:54:46 CEST] <nevcairiel> because you dont have to work on the plumbing all the time
[11:54:49 CEST] <hanetzer> aight guys, later. $dayjob
[11:54:55 CEST] <hanetzer> J_Darnley: thanks for the links
[11:55:00 CEST] <J_Darnley> no problem
[11:55:15 CEST] <BtbN> My experience with build systems is that you have to work on plumbing a lot, because the build systems break thing all the time.
[11:55:18 CEST] <J_Darnley> or you're welcome
[11:55:40 CEST] <BtbN> That primarily a problem of all the fancy new "better" build systems though
[11:55:50 CEST] <BtbN> autotools and cmake always are rock stable in that regard
[11:59:13 CEST] <nevcairiel> writing autotools files is even more arcane magic then writing custom shell scripts tho
[11:59:43 CEST] <BtbN> Yep, autotools is horrible to work with.
[11:59:43 CEST] <J_Darnley> Except for when a dev puts checks in the wrong order letting it detect signed c99 stdint but not unsigned stdint
[11:59:53 CEST] <BtbN> But stuff written 20 years ago pretty much still works unchanged
[12:07:22 CEST] <cone-429> ffmpeg 03Paul B Mahol 07master:fee7c15d8754: avfilter/af_surround: allow user to change overlap and win_func
[12:30:31 CEST] <hanetzer> I don't much mind autotools were it not for the m4sh syntax and long ./bootstrap-alike and ./configure times :)
[13:38:46 CEST] <cone-429> ffmpeg 03Gyan Doshi 07master:bf4245e9521b: doc/filters: list values for af_surround window function
[14:15:14 CEST] <cone-429> ffmpeg 03Paul B Mahol 07master:b9d25b1a6ef0: avfilter/vf_lut3d: add cineSpace 1D lut parsing
[14:15:15 CEST] <cone-429> ffmpeg 03Paul B Mahol 07master:e20ad3bd59cc: avfilter/vf_lut3d: add cineSpace 3D lut support
[14:23:36 CEST] <cone-429> ffmpeg 03Paul B Mahol 07master:782ae68a117f: avfilter: add lagfun filter
[17:46:19 CEST] <cone-100> ffmpeg 03Jarek Samic 07master:1c46ab4815f8: lavfi: add colorkey_opencl filter
[23:56:43 CEST] <cone-146> ffmpeg 03Lou Logan 07master:d8245cff167f: doc/mailing-list-faq: auto unsubscribe due to DMARC
[00:00:00 CEST] --- Sun Apr 21 2019