IRC logs for #baserock for Tuesday, 2016-04-12

*** edcragg has quit IRC00:26
*** ctbruce has joined #baserock06:56
*** edcragg has joined #baserock07:49
*** toscalix has joined #baserock07:57
*** rdale has joined #baserock08:16
*** jonathanmaw has joined #baserock08:18
*** ssam2 has joined #baserock08:35
*** ChanServ sets mode: +v ssam208:35
*** CTtpollard has quit IRC08:42
*** CTtpollard has joined #baserock08:43
*** CTtpollard has quit IRC09:03
*** CTtpollard has joined #baserock09:04
*** edcragg has quit IRC09:12
ssam2Aboriginal fails to build the cross GCC on my system because I have systemtap installed09:17
ssam2cross compiling :-(09:17
ssam2seems someone sent a patch to fix this to GCC in 2012 but it never went anywhere09:18
richard_mawwould building a native chroot first make it work?09:19
richard_mawi.e. is it just systemtap headers being installed that's the problem, or does it really complain if there's kernel modules about?09:20
ssam2it's just the headers being installed09:20
ssam2so running in a chroot would work around the problem, but is a bit sad09:21
richard_mawI'm lacking all the context on the aboriginal stuff, but for reliability I would have thought the best approach would have involved making a native chroot first09:22
*** anahuelamo has quit IRC09:22
*** anahuelamo has joined #baserock09:22
ssam2I've always looked at Aboriginal as something that's expected to work anywhere09:23
ssam2but actually perhaps it's not09:23
ssam2i'll ask in #toybox later I guess09:23
richard_mawI think you need something like the google blaze sandboxing thing if you're going to try building without a chroot, since otherwise you're at the mercy of *every* component to behave09:24
richard_mawIIRC blaze's sandbox lets you restrict which paths the build system is allowed to access, so you can at least detect when the build system is misbehaving09:27
ssam2i would love to rewrite GCC's build system, but i'd need infinite time09:29
ssam2Note Aboriginal aims to build the minimum possible number of things09:30
ssam2it's akin to Baserock's build-essential.morph09:30
ssam2the idea being that it builds you a minimal build environment that you can then run clean builds in09:31
richard_mawah, so it's how you produce that chroot09:31
richard_mawI fear it would fall into the same problems we had, namely all the other self-hosting toolchains09:32
*** locallycompact has joined #baserock09:32
ssam2it certainly doesn't have any magic solution to the problem of cycling dependencies09:33
*** edcragg has joined #baserock09:33
ssam2*cyclic09:33
ssam2it doesn't claim to, though09:34
ssam2compiler authors can either provide a way to bootstrap their compiler from C, or they can do their own cross-building for every platform they want to support and provide binaries09:34
ssam2right, it seems like the problem is that my GCC build is: build=x86_64, host=armv6l, target=armv6l09:51
ssam2since host==target, GCC's configure looks at /usr/include for native headers09:52
ssam2I guess Aboriginal can't run this build in a chroot, because my machine can't run armv6l code09:52
ssam2wait, it could run it in an x86_64 chroot. That would mean a whole extra build step... but it might be worth it for the extra reliability09:58
ssam2maybe we could do recommend a 2-stage Aboriginal build, first host->host, second host->$target09:59
ssam2-do09:59
ssam2although i would still like gcc to be less broken09:59
richard_mawI think that's up there with the moon on a stick.10:00
pedroalvarez:q10:02
jmacsWouldn't we all.10:04
pedroalvarezwrong window, but :q!10:06
edcragg:p10:06
ssam2hmm, possibly this is an issue in tristan's build script for new GCC10:10
ssam2which could pass in --with-native-system-header-dir if it wanted to, to point to the armv6 headers10:11
ssam2seems --with-sysroot and friends wouldn't have any effect because host==target10:11
ssam2which seems wrong, but i think it'd be a hard assumption to change10:11
persiassam2: If your machine isn't the architecture you want it to be, linux's binfmt-misc allows you to silently invoke an interpreter for non-native code: while this is often used for Java, you could use it with qemu-static, so that you have a foo chroot on your bar machine that apparently executes.  I know this works for armv5, armv6, and armv7 on ppc or x86_64, and for armv5, armv6, and armv7 on ppc.10:12
persias/on ppc or x86_84/and ppc on x86_64/10:13
pedroalvarezssam2: hit this error using bsaerock-export today https://github.com/ssssam/baserock-export/pull/210:49
ssam2thanks10:52
ssam2adding --with-native-system-header-dir=$(STAGE_DIR) seems to have fixed my Aboriginal woe10:52
ssam2ha, my next issue is the OOM killer killing this 'genautomata' program11:07
ssam2https://gcc.gnu.org/bugzilla/show_bug.cgi?id=7047311:09
*** cosm has quit IRC11:26
ssam2I've got my aboriginal-ybd into gear, but this initial swap file creation takes a really long time!12:39
ssam210 mins and counting12:39
pedroalvarezI fear it is going to depend on lots of ram and ssd to be usable12:50
locallycompactis it not possible to create repositories on a trove manually with gitano?12:53
locallycompactssh@troveprefix create troveprefix/foobar12:53
locallycompact*ssh git@12:54
pedroalvarezit should be possible12:55
locallycompactWhen I do that I get a fatal anyway12:55
ssam2i think you need troveprefix/projectname/foobar12:56
ssam2and you need to be in projectname-admins or projectname-managers group to do that12:56
ssam2and you need to manually create those groups before you can be in them12:56
locallycompactah12:56
ssam2It might be documented/scripted somewhere12:57
locallycompactThanks I get the idea12:57
ssam2http://wiki.baserock.org/Trove/reference/#index2h3 suggests i'm wrong, but I think it's actually the manual that's wrong12:58
ssam2i have a feeling my Aboriginal worker VM is stuck in a busy loop :(13:04
ssam2100% CPU but no IO at all13:04
ssam2the last line of the log file is 'Uncompressing Linux... done, booting the kernel.'13:04
ssam2this is the full log.. also has some warnings which may be relevant http://paste.baserock.org/uhegoxogig13:05
ssam2seems that the dev-environment.sh script from this Aboriginal build isn't working as expected. I'll try the same thing with upstream Aboriginal13:16
ssam2ok, upstream Aboriginal doesn't seem to actually manage to boot either for armv6l13:42
* ssam2 tries a prebuilt Aboriginal armv6l image13:44
ssam2ok, seems that is broken too. but the prebuilt armv5l image has booted13:50
*** gtristan has quit IRC14:01
paulsherwood18:00 <+ssam2> if the goal is for developers to use this, it should be able to run on my laptop, surely?14:25
paulsherwoodssam2: i think developers doing builds can usefully reduce their waiting time using AWS, is all :)14:26
*** gtristan has joined #baserock14:27
paulsherwoodi know gtristan bought a humungous devbox14:28
ssam2it's an option, I agree14:28
ssam2in some cases14:28
ssam2it wouldn't fix whatever has broken armv6l in aboriginal, though :-)14:29
paulsherwooddid you manage to 'build some Baserocks' ?14:29
ssam2not yet, i'm retrying with armv5l14:30
ssam2the prebuilt armv5 Aboriginal image seems to work, unlike the prebuilt armv6 image14:30
paulsherwoodack14:31
paulsherwoodbtw i'm unclear what to say to your simplification of build-steps... it's elegant, but it means build-tools are back to hard-coding pre- and post- iiuc14:32
ssam2is that a problem?14:32
ssam2if you don't like the approach, the original idea is workable14:33
ssam2I would want to put a comment in our DEFAULTS file explaining the purpose of the pre- and post- commands, though14:33
ssam2since if people predefine values for them in DEFAULTS, they become pointless14:33
ssam2or we could remove the pre- and post- variants, but we do actually use them for the intended purpose (prepending or appending the default commands) in various places14:34
paulsherwoodfor purity we should define pre- and post- in the schema too, is all?14:40
ssam2I don't understand what you mean..14:41
ssam2my point is that pre-build-commands and post-build-commands shouldn't be treated the same as build-commands:  users should be allowed to predefine build-commands in DEFAULTS, but should not be allowed to predefine pre-build-commands or post-build-commands in DEFAULTS14:42
paulsherwoodi agree... my concenr is that we shouldn't hard-code 'pre-' and 'post-' in the tools. there should be build-steps and build-step-stages or soemthing14:45
ssam2i see. but what would we gain from that?14:45
paulsherwoodpls hold14:46
gtristanssam2, unable to boot the armv6l arch ? I think I did not test that arch after switching to the virtio 9p mount :-/14:53
gtristanssam2, could it be that you have an older qemu ?14:54
ssam2could be that14:56
ssam2it seems the upstream Aboriginal armv6 image is broken, so this isn't your fault :-)14:56
gtristanis that output from aboriginal-start [...] --interactive ?14:56
gtristanhmmm, so did you boot armv5l without issue ?14:56
ssam2what I pasted was the ybd worker log file14:56
ssam2I've booted the upstream Aboriginal armv5 image OK, i'm yet to try yours (still building)14:56
gtristanright, yeah the output of that is a bit slim, you dont get any of the kernel console output cause, it's detached from any console14:57
ssam2right14:57
ssam2but using the run-environment.sh script it stops at the same point14:57
ssam2so I think this is actually the kernel getting stuck very early on14:57
gtristanwith --interactive you boot the way ybd would start it up, but you get a shell14:57
gtristanI see14:57
ssam2i'll report it to the Aboriginal mailing list I guess, but might try building qemu master first14:58
gtristanyeah some of the qemu-system-arm errors in there seem to point to qemu not supporting the virtio stuff14:58
gtristanI have debian testing so it's quite bleeding edge I guess :-/14:58
ssam2I have 2.3.114:59
ssam2oh, which is quite old -- I see 2.6.0-rc1 in Git14:59
gtristanI asked landley about building qemu as a part of aboriginal, he said he used to do that but gave up once it started depending on glib and such14:59
ssam2i think it's OK to require the user to provide that15:00
ssam2it several orders of magnitude easier to build than the GNU toolchain :-)15:00
gtristanyeah, it would just be interesting to control the exact version of everything you're gonna be using15:04
ssam2that was the plan with Baserock, once upon a time :-)15:05
*** fay_ has quit IRC15:33
*** ctbruce has quit IRC15:38
ssam2ok, it works!15:40
ssam2i have an armv5 native build running on my laptop via ybd, which is pretty cool15:40
ssam2it does seem pretty slow though. Gawk's ./configure script took nearly 5 minutes15:40
gtristanssam2, configure is slow... compiling regular objects is pretty quick; for those low level packages I think they all compiler faster than they configure (except glibc/gcc of course)15:43
gtristanthen compiling libtool objects is hellishly slow again15:43
ssam2yes, build of gawk took 2 minutes15:43
ssam2which is pretty good15:43
ssam2so we just need to kill autoconf!15:43
gtristanhaving like 4 instances at a time is nice, cause qemu doesnt do multi-core so they take a constant amount of cpu resources15:43
gtristanbut then...15:44
ssam2although autoconf is part of the reason we have to do this trickery in the first place... a decent build system would be able to crosscompile in most cases15:44
gtristanthe IPC is shaky, I almost always lose one instance when I set instances to 415:44
gtristanSo one question15:45
gtristanDo we maintain any baserock build of... poky ?15:45
pedroalvarezhahah wat?15:46
*** richard_maw has left #baserock15:46
pedroalvarezlook what you have done ^15:46
CTtpollardhaha15:46
ssam2we don't15:46
gtristanIf someone can come right now and say... I have this definitions branch which builds exactly the content of poky... then there is some opportunity15:47
pedroalvarezactually we looked into exporting yocto recipes into baserock definitions..15:48
pedroalvarezhttp://wiki.baserock.org/projects/importing-from-bitbake/15:48
edcraggthat would be a handy thing to have15:49
*** richard_maw has joined #baserock15:50
*** CTtpollard has quit IRC16:16
*** gtristan has quit IRC16:19
*** gtristan has joined #baserock16:30
ssam2same problem with the Aboriginal armv6 image using git master of QEMU16:38
ssam2i'll report it to the mailing list I guess16:39
*** fay_ has joined #baserock16:39
ssam2oops, now I've killed the wrong qemu, and I guess my aboriginal/ybd build is going to hang forever16:40
*** ctbruce has joined #baserock16:45
gtristanssam2, yup, have I put enough emphasis on how horrible my IPC solution is ? ;-)16:46
ssam2ctrl+c worked at least :-)16:46
*** gtristan has quit IRC16:54
*** jonathanmaw has quit IRC17:00
*** bashrc has quit IRC17:00
*** toscalix has quit IRC17:09
*** locallycompact has quit IRC17:09
*** ssam2 has quit IRC17:27
*** anahuelamo has quit IRC18:07
*** rdale has quit IRC18:07
*** edcragg has quit IRC18:14
*** fay_ has quit IRC18:15
*** ctbruce has quit IRC18:25
*** locallycompact has joined #baserock19:07
*** gtristan has joined #baserock19:37
*** gtristan has quit IRC21:41
*** gtristan has joined #baserock21:44
*** locallycompact has quit IRC21:58
*** gtristan has quit IRC23:47
*** gtristan has joined #baserock23:50

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