IRC Logs for #crux-devel Tuesday, 2018-10-02

Worksterjaeger, you'll need to add xorg-libxrandr to mesa3d on iso/setup.dependencies and it needs to be added on the wiki https://crux.nu/Wiki/TODO3501:37
Worksterhttps://crux.nu/gitweb/?p=ports/xorg.git;a=commitdiff;h=89fdc08d922a20b0461dfe43ad6d43c34d144a2501:37
jaegerSo I think it's pointless to have the LLVM JOBS option there01:48
Worksterit's set in the cmake configuring part not on the ninja command. which is how it said to do it.03:01
Worksterso are you saying to remove -DLLVM_PARALLEL_COMPILE_JOBS="${JOBS:-1}" ?03:01
jaegerwell, you can leave it there if you want... but it doesn't WORK03:02
Worksterit's mentioned in the cmake file. bah i need to spend time checkng it again03:02
Worksteri'm using distcc and it seems to be working over my -j2403:04
Worksterbut what you're saying is it wont work less than the number of cores on your cpu03:05
Worksterhttps://groups.google.com/forum/#!msg/llvm-dev/P-kQf_8uhbg/nHXezQTGEwAJ03:06
jaegeryes. the 'ninja' call uses all cores unless you add -jN03:06
jaegertry it and see03:06
jaegerset your MAKEFLAGS to -j1 and build llvm, watch your threads03:07
Worksteri did when i built llvm03:07
Worksterokay i'll look at that later. so i got llvm and clang 7.0.0 do i push or wait on teK_ it's been 12 days now.03:07
Workstersince release03:07
Worksteri couldn't get ninja to use more than my my 4 cores with the ninja -j option so i set it on LLVM_PARALLEL_COMPILE_JOBS note that i didn't set LLVM_PARALLEL_LINK_JOBS as uping that with distcc would just make a mass amount of ram used for no benefit.03:09
Worksterso ninja also needs the JOBS set as well as in the cmake part.03:09
jaegerIf you want it to change. I'm not fussed about it, just wanted to point it out for consistency03:14
jaegerIt was breaking my time plot tests but I worked around it with a local copy for that, all good03:18
Workstermay have changed since ninja update or something, i couldn't use more than 4 cores without LLVM_PARALLEL_COMPILE_JOBS and i did try -j then and it didn't do more than 4. might of changed now. i'll retest03:27
jaegerI haven't tested with that flag removed because I didn't notice it before... will do that when my tests finish03:27
Worksteryou built 6.0.1 or 7.0.0 ?03:29
Worksteri'm upto llvm-32 to 7.0.0 now03:29
jaeger6.0.103:30
Worksterjaeger, did you make a note for the new dependency for mesa3d to inject it on the iso for crux 3.5 ?03:31
jaegerindeed03:32
jaegerthanks03:32
Workstercool03:40
jaegerprobably best to wait for 7.0.0 until opt/llvm is updated, might be weird if there's a mismatch03:57
Worksterwell i was thinkingof pushing all 3 but waiting on teK_'s input03:58
jaegerwoot, on build 148 of 160, will be done soonish04:08
Worksterso i've just checked llvm with JOBS= 1 2 and now 3, not touched the llvm port from what is currently in opt and i see 1 2 and now 3 g++ instances when i build.04:20
Worksterbut that's with building llvm 7.0.004:23
jaegerPerhaps it's fixed in 7.0.0, though I dunno, thought the problem was ninja04:23
jaegerMaybe I'm nuts04:23
WorksterACTION shrugs04:23
Workstermy be fixed in 7.0.004:23
Workstermay*04:23
jaegerBefore I modified the port, all my builds were taking ~532s, for what that's worth04:24
jaegerSo they were clearly NOT honoring MAKEFLAGS/JOBS there04:24
Worksterbecause i can remember testing this ages ago. and i wondered why it was so slow to build on my cluster until i found that option to cmake.04:24
Worksteryeah ninja is a pain it hasn't even got a environment variable for for jobs, i had looked.04:25
jaegerwas that before the switch to ninja?04:25
Worksterafter, i don't recall what i did before.04:25
Worksterwell llvm/clang didn't have to use ninja but it greatly makes the build faster. you could test without ninja if you wanna have a benchmark from before ninja was used. https://crux.nu/gitweb/?p=ports/opt.git;a=commitdiff;h=6d0a17f046fee9e2995e17fbe5241c3b0302055004:29
jaegernah, not THAT interested, hehe04:29
jaegersince I think ninja is the issue04:29
Worksteri didn't think so :)04:29
Workstermight be a bug in ninja but i can't see it on llvm 7.0.004:30
Worksteri can test on llvm 6.0.1 if you really want me to.04:30
jaegerIf you want to, sure. But don't feel obligated04:31
jaegerIt takes a lot of time to build llvm 160 times :P04:36
Workstertrue04:46
*** prologic_ has quit IRC09:16
*** prologic has joined #crux-devel09:20
*** Workster has joined #crux-devel12:33
*** prologic has quit IRC12:51
*** prologic has joined #crux-devel12:51
jaegerRomster: ooooh, I've just realized I'm stupid. Somehow I failed to set JOBS and MAKEFLAGS, just MAKEFLAGS. Seems like that was the problem. Sorry for the noise. :(14:57
jaegerpedja: building my kernel config takes 0m55.541s15:25
pedjadamn, that's quick :)15:30
jaeger:D15:30
pedjawhat GPU ended in that machine, 1080ti?15:42
jaegercurrently a 108015:48
pedjanice15:52
teK_Romster: I notice of bug instead of changing my ports would be preferred22:25
teK_s/I/A/22:25
teK_(referring to gegl here)22:29
teK_also, I can confirm LLVM_PARALLEL_COMPILE_JOB referencing $JOBS to work as suggested by Romster. Will stick to that, thanks.22:37
jaegeryeah, looks like it works when you're not stupid (me)22:38
teK_I still think it's odd to have cmake and ninja interact in that way22:39
teK_ACTION shrugs22:39
WorksterteK_, it would be nice if you built your ports in a clean environment, also your vlc port is broken but i never bothered to fix it again. and i was waiting on you if i can push the llvm and clang updates or you want to do them yourself.23:31
Worksterin the pst 've opened bug reports for updating llvm and i got told to push those when you were MIA but since you're back. lst time i tried to email your email was not working. i've just got frustrated and anything trivial i've just went ahead and fixed, i guess i should stop doing that and bug report it all again.23:33

Generated by irclog2html.py 2.14.0 by Marius Gedminas - find it at mg.pov.lt!