IRC logs for #baserock for Friday, 2017-04-28

*** brlogger has joined #baserock01:06
*** brlogger has joined #baserock01:32
*** rdale has quit IRC02:17
*** SotK has quit IRC02:22
*** SotK has joined #baserock02:25
*** gtristan has quit IRC05:33
*** rdale has joined #baserock06:25
*** gtristan has joined #baserock06:29
*** paulw has joined #baserock07:03
*** ctbruce has joined #baserock07:37
*** toscalix has joined #baserock08:03
*** ssam2 has joined #baserock08:25
*** ChanServ sets mode: +v ssam208:25
*** jonathanmaw has joined #baserock08:32
*** jude has joined #baserock09:29
*** ctbruce has quit IRC10:35
*** ctbruce has joined #baserock10:35
*** ssam2 has quit IRC10:42
paulsherwoodjjardon: sorry to bug you... did you manage to figure out what broke the gnome build i was trying?10:53
jjardonpaulsherwood: didn't manage to reproduce, I was planning to try again with several instances: how many are you using?10:57
paulsherwood410:59
jjardonok, thanks, lets see how it goes: https://gitlab.com/baserock/definitions/pipelines/792773911:03
gtristanjjardon, is that CI using YBD master ?11:09
jjardonyes11:10
gtristanstrange that it would pass, and it's building from scratch right ?11:10
gtristannot reusing any artifact server ?11:10
ironfootmy guess is that this problem becomes visible with more than one instance11:11
gtristanI think that is still by pure chance11:12
gtristanthe "fix" though, I've said before, is to make all things in build-essential stage3 depend on fhs-dirs11:12
jjardonrunning this in parallel with default number of instances (1) https://gitlab.com/baserock/definitions/pipelines/792770611:13
paulsherwoodi guess i can just try that on my machine11:13
jjardongtristan: every branch that gets merged in ybd needs to build build-essential before from scratch (without using kbas)11:14
paulsherwoodgtristan: how come builstream does the gnome image successfully, if this is a definitions issue?11:17
gtristanpaulsherwood, build-essential was a manual conversion for several reasons, the conversion script rebases any definitions against an already made "base"11:28
gtristanSo, I handled that myself, wondered if it way my own problem, ended up ensuring that things in stage3 depend on fhs-dirs11:29
gtristanthat was in a second (manual) conversion which took into account the /usr merge stuff11:29
gtristanwas not needed pre /usr merge since the symlinks were not there11:30
ironfootright, then patch should be simple11:30
gtristanjjardon, fwiw building build-essential by itself is not enough to catch this11:30
gtristanjjardon, it's once you start staging the output of build-essential that symlinks try to stomp on non-empty directories at random11:31
gtristanironfoot, yes, it's an easy fix, just build-depend on fhs-dirs for any of the last stage build-essential11:31
jjardonsorry, its building the minimal system, not only build-essential11:32
jjardongtristan: but minimal = build-essential + bsp, so maybe not enough to catch that11:32
gtristanthat should fail honestly, the only reason I can think of it consistently passing, is if with instances = 1, then staging order becomes the order listed in the build-essential stratum (which *happens* to have fhs-dirs first)11:33
gtristanjjardon, well, true... actually it will only trigger an error once you reach a package which has a build script that explicitly says #!/bin/bash11:33
gtristannot all of core does this, but some do11:34
gtristanrunning a shell in PATH tends to work11:34
ironfootsimple fix! anyone? :)11:34
gtristanOh look11:37
gtristanit looks like I *did* push a fix for the gcc issue11:37
gtristan2bf01667c7ef1f79a6b94693bb796a457f837b2b should fix that for the runners with gcc 6 host tool11:37
* jjardon is amazed how many things works by pure coincidence11:37
jjardongtristan: yup, I stole your patch from here and put it in a MR :P11:38
gtristanAh11:40
gtristanlooks like it only needs linux-api-headers to depend on fhs-dirs11:40
* gtristan kicks off a local build11:40
gtristan(because glibc depends on linux-api-headers, and everything else in stage 3 depends on glibc in some way)11:42
gtristanwell, that does reduce at least the random possibilities11:42
paulsherwoodgtristan: like this? https://gitlab.com/baserock/definitions/commit/ef6a124911dc2b821e163a6254992f1bc8be215a11:42
gtristanit's only once out of all stage3 chunks that it could randomly stage fhs-dirs *after*11:42
gtristanpaulsherwood, nope11:43
gtristanwell, I wonder if stage2 needs any fixing11:43
gtristanit might11:43
paulsherwoodmy build agrees with you ... failed11:43
gtristanwhat I am trying is, only add fhs-dirs as a dependency of linux-api-headers11:44
gtristan(not stage1 or stage2, only in the third stage)11:44
gtristanstage2-gcc and ticking... will be another 20min I guess before I know11:49
gtristanI think there may still be a bug12:04
gtristanbuilding in stage 3 and I get some messages I wouldnt expect:12:04
gtristan3 17-04-28 00:01:03 [1/57/72] [UTILS] WARNING: Ignoring symlink "/store/BASEROCK/build/tmp/tmpiVikBB/lib" which purges non-empty directory12:04
gtristanAlthough, it _only_ happens with /lib12:05
* gtristan sees build-essential stage3 fly by at lightning speed with instances: 412:05
gtristannice12:05
gtristanOk... so which chunks are known to fail ?12:14
* gtristan is now building core.morph chunks12:14
* gtristan did get more suspicious symlink warnings about /sbin and the like, though12:15
gtristanso it's a bit worrying, not sure if this fixed it12:15
gtristanoh but, now that I'm in core, they are not happening anymore12:16
* gtristan is building file, cmake, libexpat...12:17
gtristanand gettext-tarball, still no errors12:17
gtristanThere you go: https://gitlab.com/baserock/definitions/merge_requests/3512:20
gtristanOn a silver platter12:20
gtristaneven more shiny one than the platter I handed over yesterday, with MR and all :)12:20
ironfootshiny12:21
gtristanusually breaks in core.morph right ?12:22
* gtristan is still building core.morph with no problem, and I expect the CI will do it too12:22
gtristannope12:22
gtristanthe CI only did 'validate-all-definitions' and called it a day12:23
ironfootheh12:23
ironfooterrors reported by paulsherwood were from flac12:23
gtristancore completed12:24
* gtristan launches foundation12:24
ironfootalso linux-x86-64-generic failed for him12:24
gtristanyeah, that certainly fixed it12:24
gtristanno more symlink warnings at all12:24
gtristanjust some overlapping paths12:24
ironfootmultiple instances?12:25
gtristan412:25
ironfootgolden platter12:25
* gtristan has diff, sed and tar currently building12:25
gtristanand even there, have processing to spare12:26
gtristaneven 4 instances does not use enough of my resources :-/12:26
gtristanwell, I guess there is a lot of I/O, I need faster SSD !12:27
gtristanheh12:27
gtristanok that's all for today12:27
gtristanSeeee ya !12:27
ironfootthanks gtristan!!12:27
* paulsherwood fails again, with http://paste.baserock.org/ugowesaduh12:27
paulsherwoodmake: /bin/sh: Command not found12:28
ironfootpaulsherwood: this is the fix that is working for gtristan - https://gitlab.com/baserock/definitions/commit/913944e4c8730d6454751b94f98ef7c4ff35678412:29
bjdookswi/win 1512:29
ironfootpaulsherwood: so what you did, but without removing stage2-fhs-dirs dep12:29
*** gtristan has quit IRC12:31
paulsherwoodironfoot: tvm, trying now12:42
ironfootpaulsherwood: please, if that doesn't work, note it here: https://gitlab.com/baserock/definitions/merge_requests/3512:43
paulsherwoodack12:43
*** ssam2 has joined #baserock14:02
*** ChanServ sets mode: +v ssam214:02
jjardonpaulsherwood: mmm, still not able to reproduce with 4 instances: https://gitlab.com/baserock/definitions/builds/15242422 in what component it normally fails?14:05
paulsherwoodjjardon: i'm tempted to say, let's not worry about it...14:05
paulsherwoodmy build with the fhs-dirs commit seems ot be proceeding fine14:05
paulsherwoodwell, everything built except the system itself... http://paste.baserock.org/usenurusor15:31
paulsherwoodbut i notice that ybd 4 instances *took longer* than buildstream (147 minutes vs 140 minutes)15:33
paulsherwoodsorry, 167 vs 16015:37
paulsherwoodhowever it may be that this is not apples vs apples... maybe bst hasn't built completely from scratch?15:38
*** toscalix has quit IRC15:49
*** paulw has joined #baserock16:09
*** jonathanmaw has quit IRC16:40
*** ssam2 has quit IRC17:02
*** jude is now known as Guest1484417:29
*** persia has quit IRC17:30
jjardonpaulsherwood: reproduced with several instances: https://gitlab.com/baserock/definitions/builds/1524242217:58
*** CTtpollard has quit IRC18:42
*** gtristan has joined #baserock19:01
*** rdale has quit IRC21:44

Generated by irclog2html.py 2.15.3 by Marius Gedminas - find it at mg.pov.lt!