IRC logs for #baserock for Tuesday, 2015-05-05

*** zoli__ has joined #baserock00:56
*** zoli__ has quit IRC01:00
*** inara has quit IRC02:48
*** inara has joined #baserock03:08
*** zoli__ has joined #baserock04:04
*** zoli__ has quit IRC05:18
*** zoli__ has joined #baserock05:49
*** straycat has joined #baserock06:16
*** straycat_ has joined #baserock06:34
*** straycat has quit IRC06:35
*** straycat_ is now known as straycat06:35
*** gary_perkins has joined #baserock06:58
*** sherm_ has joined #baserock07:15
*** Albert_ has joined #baserock07:35
*** a1exhughe5 has joined #baserock07:43
*** straycat has quit IRC07:45
*** straycat has joined #baserock07:45
*** rdale has joined #baserock07:53
*** mariaderidder has joined #baserock07:54
*** bashrc has joined #baserock08:04
*** sambishop has joined #baserock08:04
*** gary_perkins_ has joined #baserock08:10
*** bashrc has quit IRC08:11
*** gary_perkins has quit IRC08:11
*** a1exhughe5 has quit IRC08:11
*** flatmush1 has quit IRC08:12
*** sherm_ has quit IRC08:12
*** a1exhughe5 has joined #baserock08:24
*** sherm_ has joined #baserock08:26
*** jonathanmaw has joined #baserock08:27
*** tiagogomes_ has joined #baserock08:31
*** gary_perkins_ has quit IRC08:34
*** flatmush has joined #baserock08:36
*** bashrc has joined #baserock08:37
*** CTtpollard has joined #baserock08:44
*** ssam2 has joined #baserock08:50
*** ChanServ sets mode: +v ssam208:50
*** jonathanmaw has quit IRC08:51
*** gary_perkins has joined #baserock08:54
*** Krin has joined #baserock09:04
*** jonathanmaw has joined #baserock09:06
*** edcragg has joined #baserock09:10
*** inara has quit IRC09:11
*** gary_perkins has quit IRC09:16
*** gary_perkins has joined #baserock09:17
*** lachlanmackenzie has joined #baserock09:46
*** inara has joined #baserock09:50
pedroalvarezHi everyone09:57
SotKhi pedroalvarez09:57
pedroalvarezI'm having weird issues when trying to clone from gerrit over https :/09:57
* pedroalvarez prepares some logs09:58
pedroalvarezmeh... nevermind, network issues...10:01
*** inara has quit IRC10:03
*** inara has joined #baserock10:14
jjardonssam2: Hi, if you have some free time maybe you can take a look to ? (sorry, I'm still unable to add you as a reviewer)10:21
pedroalvarezjjardon: re - what do you mean by "systemd will create other users...."10:26
jjardonpedroalvarez: systemd support declarative allcation of system users and groups:
jjardonpedroalvarez: check your /etc/paswwd in a baserock system. users with UID > 900 have been created by systemd in boot time10:31
* pedroalvarez checks /usr/lib/sysusers.d/10:32
pedroalvarezjjardon: ni e10:42
pedroalvarezI was expecting systemd-sysusers to create also the home folder, and also change its permisions for the new user10:43
jjardonpedroalvarez: AFAIK, systemd-sysusers is for system users, not normal ones10:44
jjardonit should be possible ro remove the sshd user as well (I do not have any in my distro), but I think we need to change our openssh .service files to make it work10:45
pedroalvarezI would say that system users can also have a home folder10:48
persiaGenerally $HOME for system users should not be under /home10:49
*** inara has quit IRC11:04
*** Krin has quit IRC11:12
*** inara has joined #baserock11:18
*** inara has quit IRC11:36
*** Krin has joined #baserock11:37
*** inara has joined #baserock11:39
*** mariaderidder has quit IRC12:15
*** fay__ has joined #baserock12:24
*** fay_ has quit IRC12:24
*** fay__ has quit IRC12:34
*** inara has quit IRC12:39
*** fay__ has joined #baserock12:40
*** inara has joined #baserock12:40
*** ssam2 has quit IRC12:51
*** ssam2 has joined #baserock12:51
*** ChanServ sets mode: +v ssam212:51
tiagogomes_richard_maw, another fact is that is the system installed that doesn't boot, the installer works fine12:54
richard_mawoh, that *could* be the GCC update12:54
richard_mawsince then it would be the syslinux built with the gcc on the system that gets used12:55
tiagogomes_¬_¬, I'll revert it on my local machine and try it. We really need to stop to update GCC unless there is a valid reason to do so (like adding support for aarch64)12:56
* richard_maw has tested (SHA512 hashing of passwords) and can confirm that it works12:56
KinnisonWe will need to compare it with pedro's work on fixing PAM12:57
richard_mawtiagogomes_: the issue is with a toolchain newer than some other component on the system12:57
Kinnisonpedroalvarez: how is that going btw?12:57
richard_mawtiagogomes_: that, and compiler developers not treating undefined behaviour as ABI12:58
pedroalvarezKinnison: currently testing it12:58
Kinnisoncool beans12:58
tiagogomes_well, they shouldn't use undefined behavior in the first place12:58
persiaI'd prefer to fix all the other things, rather than not updating the compiler, but I'm forward-looking that way12:59
richard_mawtiagogomes_: I'd agree, but the C specification is not easy, so it's very easy to accidentally use undefined behaviour13:00
*** mariaderidder has joined #baserock13:24
mwilliams_cthas anything been built with Baserock on sysV init before?13:27
rjekProbably not.13:28
rjekI think we started with systemd13:28
rjekI think we enable sysV init scripts though?13:28
rjek+support for13:28
richard_mawrjek: we started with busybox init, migrated to systemd fairly early and we have always built systemd with sysv init script support13:29
richard_mawI don't even know if systemd _can_ be built without sysv init script support13:30
rjekmwilliams_ct: Do you have more context for your question you can share?13:30
persiaI thought the minimal system still wasn't systemd13:30
richard_mawpersia: correct, it's busybox init13:31
richard_mawwe migrated back partially for a specific use-case13:31
mwilliams_ctrjek: honestly, you've answered the question as much as I needed to already. but context is that we're trying to port openBMC onto Baserock, and atm it has a variety of lovely sysV init scripts13:31
richard_mawcustom verbage? lack of pidfiles?13:32
richard_mawthere's no issue with using the legacy sysv init scripts13:32
richard_mawprovided you don't need it to start up faster or extra dependency information13:33
* pedroalvarez ends up with a system on which he can't log in13:41
jonathanmawpedroalvarez: pam config doesn't have nullok?13:43
pedroalvarezhm.. to allow null passwords..13:45
jonathanmawpedroalvarez: nullok is required for 'login' at the bare minimum, I think.13:45
jonathanmawI'm not sure if that's the problem you're experiencing, though, just an educated guess13:47
pedroalvarezjonathanmaw: thanks for the info.13:50
pedroalvarezI was going to add nullok and I found that more things were wrong13:50
ssam2 !!!!!!13:53
ssam2probably the most important fix in Baserock ever: fixing the delete key13:53
*** inara has quit IRC13:58
*** Albert_ has quit IRC14:02
*** Albert_ has joined #baserock14:02
*** inara has joined #baserock14:10
jjardonssam2: Any idea when the new release is going to be made? Any outstanding problems with the current rc?14:19
jjardonpedroalvarez: about ; not sure touching systems files in random extensions is in general a good idea (vagrant.configure changes /etc/profile as well, apart from adding a system user)14:25
jjardonif a new system user is needed, maybe we should modify /usr/lib/sysusers.d/ instead?14:26
pedroalvarezyes, that might be the right way to do it14:26
tiagogomes_jjardon yes, images installed with the baserock installer don't boot14:26
jjardontiagogomes_: ah, whats the baserock installer?14:27
pedroalvarezjjardon: I'm not against a better solution, I'm just against breaking the current solution14:27
jjardonpedroalvarez: sure, can you point in the review what extensions/chunks will break with the change?14:29
ssam2jjardon: no idea about that, I'm afraid14:29
ssam2hopefully the project I'm on will fund a day for me to do another one, don't know when14:29
richard_mawtiagogomes_: likely culprits are mkfs.btrfs, the rawdisk write extension, the installer script itself or gcc 5.114:29
* ssam2 is happy to explain the release process to anyone who fancies doing them in future14:30
jjardonssam2: if you write it down in the wiki anyone can do it :)14:31
tiagogomes_I want to lorry cdrecord, but I can't find a suitable download link14:31
jjardonssam2: nice!14:32
jjardonso I guess a baserock release is a morph release + some supported baserock systems, rigth? What are the release cycles based on? feature-wise or time-wise?14:33
ssam2there's no cycle. anyone can do one any time14:33
ssam2in practice, we do one whenever one is required for GENIVI, which is every six weeks (but not always needed)14:34
tiagogomes_what do we use to download tarballs in lorry? This link will work with wget but not with curl, as the latter doesn't follow links14:34
ssam2jjardon: the systems released are
rjektiagogomes_: curl -L is what you want?14:36
ssam2jjardon: 'morph' doesn't really have a release process at all, we just roll up a given version into a Baserock release. It would be good to separate Morph out, but it'd be more work, and nobody seems interested in doing it14:36
tiagogomes_rjek yep, now I wonder if lorry does the same14:36
ssam2wait, it's that does the work14:36
ssam2so, we use Python urllib2 ... which we can be fairly sure will do the wrong thing14:37
ssam2worth testing though in case it doesn't14:37
ssam2updating 'lorry' to use 'requests' instead would be easy and might improve matters14:37
paulsherwoodif we're going to fiddle with lorry, can we make it do shallow clones, too?14:38
tiagogomes_urllib2 it is14:38
ssam2paulsherwood: i'm not sure what you mean by that14:39
paulsherwoodwe don't lorry gcc from git, because it's too large (and possibly other stuff too). but we could use shallow clones to have gcc from a reasonable previous version and ongoing14:40
rjekISTR that there are quite serious reproducability issues with doing that, but I don't recall the details and thus I might be making it up.14:41
paulsherwoodrjek: my 'tests' showed that the SHA1s are the same14:41
tiagogomes_urllib2 did the right thing this time :)14:41
paulsherwood(iiuc git has improved since lorry was originally developed)14:42
rjekThat is likely14:42
ssam2paulsherwood: shallow cloning could be 'opt in'. I wouldn't want it there by default (history is useful)14:42
rjekI find having history for loads of projects in one place one of the best things about having a Trove.14:42
paulsherwoodi agree.14:42
paulsherwoodmy problem is we don't have any history for *-tarball things14:43
paulsherwoodespecially gcc14:43
pedroalvarezthat is true, some years of history would be better than no history at all14:45
paulsherwoodwe're also missing the ongoing *collection* of history from those things14:46
jjardonagree, IMHO lorrying (and use) a tarball is a bug that should be fixed when possible14:47
ssam2if Trove automatically picked up the latest tarballs of things every time a new one appeared, it'd be good14:48
ssam2for gcc we should just fix gcc.git, but not everything has a git repo14:48
straycatpedroalvarez, do we have any guide for deploying distbuild to openstack, out of interest?14:50
pedroalvarezstraycat: I don't think so14:50
straycatokay, i think all i'm missing is the template and that should just be the entries i have in my distbuild.conf, right?14:52
rdaleat the end of last week, i created a new development image based on the master of definitions with gcc 5.1. that image runs fine, however when i build an image and copy it to my laptop to run under kvn, the linux kernel doesn't start14:55
rdale^i build an image^i build an image in the new development vm14:56
tiagogomes_rdale, yep it is broken at the moment. Most likely due the gcc update14:57
rdaleah ok14:57
tiagogomes_I reverted the gcc update locally and I will test it when it finishes building14:58
rdaleok, i'll be interested in how you get on14:59
persiaI'm a big fan of long histories (I've often traced bugs back 10-15 years), but +1 for having 5 years history over none.15:00
*** zoli__ has quit IRC15:02
pedroalvarezstraycat: sorry, I was finishing the patch15:09
pedroalvarezI have around an example for the workers15:09
richard_mawpedroalvarez: is there no way of specifying natively in the format that modules are optional?15:10
pedroalvarezrichard_maw: oh, yes, I think we can do that as well15:10
richard_mawI'd rather we had static config that worked if the module was missing, rather than removing the config that uses the module at build time15:11
pedroalvarezstraycat:, if you change the IPs of the controller and the cache server, you can use the same one to spawn as many workers as you want15:12
ratmice__hmm, wonder if you can do multiple shallow clones, or skip commits inbetween lorrying, not sure exactly how much something like the gcc repo would grow between lorrying15:12
pedroalvarezrichard_maw: so, instead of removing the lines at system-integration time, modify them at post-install time to make selinux support optional15:14
richard_mawif we can't upstream a change to make selinux optional, yes15:15
richard_mawthough I'll accept unconditionally making selinux optional, as we don't have the effort or a large enough stick to ask for proof that upstream won't take the change before accepting it as patched downstream15:16
radiofreerdale: the kernel isn't booting?15:17
radiofreei wouldn't have thought that was down to the gcc 5.1 change15:17
radiofreewhat panic do you get?15:17
rdaleradiofree: it doesn't get as far as panicking15:17
radiofreehow far does it get?15:18
richard_mawAIUI we can prepend the line with a `-` and change it from reqiured or requisite to sufficient or optional15:18
radiofreeso old gcc -> built new image with gcc515:18
radiofreethat's ok? the resulting image works?15:18
richard_mawbut this isn't the same as having an optional requirement, as selinux failing to auth would be ignored15:19
radiofreebut then gcc5 image -> build gcc5 image, that one doesn't work?15:19
rdaleit hangs on the line 'SYSLINUX 4.06..'15:19
radiofreesounds like syslinux is buggered then15:19
richard_mawcould be from gcc515:19
radiofreeso it's probably not the kernel15:19
richard_mawit can't be the kernel15:20
radiofreerdale: try adding a few menu entries to extlinux.conf and see if that shows up15:20
rdalehow do i do that?15:20
radiofreemount baserock.img /mnt15:20
radiofreevim /etc/extlinux.conf, copy and paste15:21
radiofreegive them new labels though15:21
radiofreehmm hold on a think you might need a bit more in there15:21
*** zoli__ has joined #baserock15:21
radiofreerdale: this is what system-version-manager generates after a few updates15:22
pedroalvarezif it's hanging on selinux, I don't think that would be the kernel15:22
ssam2syslinyx :)15:23
ssam2syslinux even15:23
radiofreetime to upgrade!15:23
pedroalvarezwe had a similar problem when upgrading btrfs[-utils]15:23
pedroalvarezyeah, syslinux :)15:23
jonathanmawhave the permissions on just changed? I'm wondering why I can't push to the branch baserock/jonathanmaw/genivi-demo-platform15:28
jonathanmawI'm suddenly getting
* pedroalvarez pushed once to master using `git push origin HEAD`15:29
Kinnisonjonathanmaw: Permission denied (publickey,keyboard-interactive).15:30
Kinnisonjonathanmaw: have you got ssh credentials on the system you're pushing from?15:30
jonathanmawah, it's because This was from a screen session, and I was using a previous tab15:31
jonathanmawso ssh -A got cut out15:31
KinnisonI keep meaning to try and write ssh agent support for mosh15:31
Kinnisonand an agent aggregator for m15:31
Kinnisonme even15:31
KinnisonI need new fingers and it's only 16:3015:32
jonathanmawhrm, browser-poc uses a submodule at, purely for documentation purposes15:43
jonathanmawshould that be lorried, or should we find a way to disable gitmodules for browser-poc?15:43
ssam2i'd lean towards disabling it, as I don't think people use built-in documentation in Baserock much right now15:51
ssam2but I have no strong opinion either way15:51
ssam2unless the repo is massive15:51
jonathanmawrepo is tiny, documentation is only built as PDF, so I'd say probably isn't what we intend to install in a system.15:52
jonathanmawIs there a nice way to disable gitmodules that doesn't involve carrying patches?15:52
richard_mawjonathanmaw: no15:54
tiagogomes_ok the update to gcc 5.1 is the culprit of not having bootable systems. I'll test build syslinux with gcc 5.115:57
persiajonathanmaw: Chunk-splitting15:58
persiaYou build the PDF, and then you split it out as a separate chunk, and then you fail to install it.15:58
richard_mawpersia: won't prevent it attempting to build it, splitting occurs post-build15:58
jonathanmawpersia: it doesn't try to install it anywhere, the problem is that it's trying to initialize a submodule from the wider internet at build time.15:59
persiaOh, ugh.  Right.  The ugly way is to just delete the relevant tree from the source as the first step in the build commands.15:59
richard_mawpersia: unfortunately no, as the submodule checkout happens before any build commands16:00
persiaNow I'm intrigued.  I no longer believe I have any answer, but I'm vaguely curious how to construct a repo that would cause morph to try to download something from the internet at build-time that cannot be overridden by the build-commands.16:01
richard_mawdo it as a submodule, morph will cache the submodule for you and check it out in the repository16:01
richard_mawit will only check out the commit mentioned in the .gitmodules file16:01
richard_mawso you ought to not lose reproducibility from this, unless the repository stops existing, or the target repository has a sha1 collision in it16:02
richard_mawmorph augments the set of repositories it needs to fetch and extract into the staging area from the git submodules16:02
richard_maweven if a build theoretically doesn't need to clone all its submodules to be functional, morph will ensure they are all available, as if you were building from a working tree prepared by running `git clone --recursive`16:03
persiaIs there any sane way to work around this at lorry-time?16:05
richard_mawyou'd have to do a filter branch of the entire history, and you'd end up with something different from upstream16:06
persiaHrm.  Annoying.16:06
richard_mawwe could fix up the submodules without modifying history by having morph translate the repository urls at parse time16:08
richard_mawthe issue becomes ensuring the translation's target contains everything necessary16:08
* Kinnison has suggested a submodule URL rewrite behaviour in definitions16:09
KinnisonBut I don't want to try and specify such until we know how definitions goes together16:09
richard_mawand that would require a few days of someone who knows definitions well, and someone who doesn't's time16:10
KinnisonJust to get started, yes16:10
Kinnisonand an ongoing effort to maintain16:10
Kinnisonwhich is why, I fear, noone has committed to it yet16:10
*** Krin has quit IRC16:29
*** a1exhughe5 has quit IRC16:34
*** Krin has joined #baserock16:40
*** Krin has quit IRC16:43
*** sherm_ has quit IRC16:46
*** mariaderidder has quit IRC16:49
*** bashrc has quit IRC17:00
ssam2it's not causing me problems, but seems 'yarn --snapshot' is broken in Baserock17:05
ssam2it expects to be able to '-x' to 'cp', but Busybox cp doesn't work17:05
ssam2s/work/support that/17:05
ssam2but the only symptom is there are no snapshots, so it's not really a problem17:05
ssam2(for me)17:05
*** jonathanmaw has quit IRC17:06
*** sambishop has quit IRC17:07
ssam2the yarns have caught some really obscure errors in the patch series I'm working on, incidentally, they are coming in very handy17:17
*** edcragg has quit IRC17:19
*** bfletcher has left #baserock17:24
jjardonradiofree: definitely the new mesa version doesn't put the dri drivers in /usr/lib/xorg/modules/dri , at least for x86 (they are built though)17:36
*** Albert_ has quit IRC17:39
*** Albert__ has joined #baserock17:39
*** sambishop has joined #baserock17:46
*** gary_perkins has quit IRC17:52
*** Albert__ has quit IRC17:54
*** zoli__ has quit IRC17:57
*** ssam2 has quit IRC18:00
*** lachlanmackenzie has quit IRC18:03
*** zoli__ has joined #baserock18:46
* paulsherwood hates deploy extensions20:06
SotKpaulsherwood: why? :)20:06
paulsherwoodbecause they are a mess20:07
paulsherwoodhere i am trying to work towards tidying up definitions.... and i suddenly discover that build is picture of tidiness by comparision with deployment20:08
SotKhm, you mean the configure extensions or the write extensions?20:09
paulsherwoodboth, i think. we have extensions in various places, and they are coupled to morph, and to cliapp (even the ones in definitions).20:11
paulsherwoodi think definitions should be buildable and deployable with minimum tooling, no 'unusual' dependencies or coupling. but perhaps i am fueled by alcohol20:12
paulsherwoodif they are just 'plugins to morph' then fine, but they should not be in definitions. if they are in definitions, they should not be coupled to morph, istm20:14
paulsherwoodand in any case, we should not be collecting jrandom files in the root of definitions20:16
robtaylorI seem to remember there was some discussion about write/configure extensions in definitions at the summit20:16
robtaylorand maybe something about making them more declarative?20:16
SotKI think the configure extensions should all be in definitions20:16
paulsherwoodi'd *prefer* to see an extensions directory, with them tidily in there.20:17
SotKpaulsherwood: +120:17
paulsherwoodbut they should not require morph or cliapp.20:17
paulsherwood(or any other non-standard library)20:17
* paulsherwood has tried moving them... but morph doesn't like that20:18
SotKI'd imagine it would be a pretty easy fix20:18
paulsherwoodso in effect we have stuff hard-coded to be in root of definitions and requiring morph20:18
paulsherwoodSotK: you're probably right20:18
paulsherwoodbut note that as i said on there are three things here - morph, systems, definitions.20:19
paulsherwoodrobtaylor: thanks :)20:20
paulsherwoodso actually, fixing it properly won't be 'pretty easy'20:22
SotKoh yes, making them nicer will be much more involved than just putting them in a subdirectory20:24
robtaylori suspect there's a step-wise approach that could be taken20:31
paulsherwoodrobtaylor: every time we take a step, we risk forgetting that there are three parts here, not just morph, but also definitions (which users may not upgrade) and existing systems (which users may not upgrade)20:38
*** persia_ has quit IRC20:39
*** persia has quit IRC20:40
*** persia has joined #baserock20:41
*** persia has joined #baserock20:41
* paulsherwood notes, in passing, that there are 2470 loc in write extensions...20:42
paulsherwood3044 in configure extensions20:42
paulsherwoodand 1753 in ybd20:42
persiaThat last one needs to be smaller.20:43
persiaThere's no good reason to have the sandboxing logic built into the build tool.20:43
persiaMind you, there's functionality I want that ybd doesn't have, which pushes it in the other direction, but still.20:43
*** persia_ has joined #baserock20:45
*** persia_ has joined #baserock20:45
* paulsherwood sees midnight approaching, and heads to bed20:46
*** zoli__ has quit IRC21:58
*** zoli__ has joined #baserock22:58
*** zoli__ has quit IRC23:03

Generated by 2.15.3 by Marius Gedminas - find it at!