IRC logs for #baserock for Monday, 2014-11-10

*** zoli_ [~zoli_@linaro/zoli] has joined #baserock06:32
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]06:41
*** zoli_ [~zoli_@linaro/zoli] has joined #baserock06:42
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]06:45
*** zoli_ [~zoli_@linaro/zoli] has joined #baserock06:45
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]06:45
*** zoli_ [~zoli_@linaro/zoli] has joined #baserock06:46
*** rdale [] has joined #baserock08:16
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]08:20
*** zoli_ [] has joined #baserock08:20
*** zoli_ [] has quit [Changing host]08:20
*** zoli_ [~zoli_@linaro/zoli] has joined #baserock08:20
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]08:37
*** zoli_ [~zoli_@linaro/zoli] has joined #baserock08:38
*** franred [] has joined #baserock08:39
*** mariaderidder [] has joined #baserock09:00
*** wdutch [] has joined #baserock09:03
*** tiagogomes [] has joined #baserock09:06
*** mauricemoss_ [] has joined #baserock09:13
*** sambishop [] has joined #baserock09:20
*** jonathanmaw [] has joined #baserock09:23
*** Krin [] has joined #baserock09:57
* pedroalvarez spams the mail list10:08
pedroalvarezgood morning!10:08
Kringood moaning10:09
*** ssam2 [] has joined #baserock10:13
Mode #baserock +v ssam2 by ChanServ10:13
pedroalvarezbtw, I think now I know what is broken in our cross-bootstrap process10:16
pedroalvarezwe are generating the same cache key for a chunk when doing a normal build and when doing cross-bootstrap10:17
KinnisonFor stage 1 ?10:18
pedroalvarezwell, I guess that only stage1 is the problem10:18
richard_1aw is now known as richard_maw10:18
pedroalvarezthe thing is that when testing cross-bootstrap, morph was fetching the gcc-stage1, and then morph couldn't use it to build10:19
ssam2pedroalvarez: oh! that shouldn't happen because we include the environment in the cache key10:22
ssam2and set TARGET_STAGE1 in the environment to produce the cross compiler10:23
KinnisonDoes the environment carry the HOST and the BUILD arch?10:23
ssam2but, maybe that's going wrong somehow10:23
KinnisonThen that's very odd10:23
Kinnisonpedroalvarez: if you unpick your gcc-stage1 artifact, and look at the meta, was the environment sane?10:23
pedroalvarezyou can do a quick test10:23
KinnisonIf so, it might simply have failed to produce a proper x-compiler10:23
pedroalvarezKinnison: the problem is that morph *downloads* the one in g.b.o from the latest release10:24
KinnisonOh dear10:25
KinnisonThat is most odd10:25
pedroalvarez` morph cross-bootstrap armv7lhf baserock:baserock/definitions baserock/release/baserock-14.40.1 systems/cross-bootstrap-system-armv7lhf-generic.morph`10:26
*** locallycompact [] has joined #baserock10:29
ssam2pedroalvarez: that downloaded a chunk from, but it looks like the correct chunj10:32
ssam2contains lots of files under 'tools/lib/gcc/armv7lhf-bootstrap-linux-gnueabi/'10:32
*** Krin [] has quit [Remote host closed the connection]10:33
pedroalvarezssam2: but you can't use it to build gcc stage 210:34
ssam2pedroalvarez: right, seems like it's somehow built for ARM!10:35
pedroalvarezyup, because that artifact is part of the latest release10:35
ssam2stage1 gcc is used to native-build stage2 gcc on the host, so it should be built for x86 :(10:36
ssam2hmm ...  I wonder if we actually encode the host architecture anywhere in the cache key for chunks ?10:36
KinnisonIf it's part of the environment then we must10:36
ssam2its env contains "TARGET": "armv7lhf-baserock-linux-gnueabi", "TARGET_STAGE1": "armv7lhf-bootstrap-linux-gnueabi",10:37
ssam2oh, and "MORPH_ARCH": "armv7lhf",10:37
KinnisonBut not hint of the BUILD or HOST arch?10:38
KinnisonThen that's likely the issue10:39
ssam2yeah, I guess the cross stage1 compiler has the same cache key as a native stage 1 compiler10:39
*** Krin [] has joined #baserock10:41
pedroalvarezrichard_maw: should I add your explanations in the commit messages?10:53
richard_mawit would be nice, but you'll probably still get a +1 without it10:54
richard_mawI haven't finished reading the thread yet10:54
*** jonathanmaw [] has quit [Remote host closed the connection]11:10
*** jonathanmaw [] has joined #baserock11:10
jjardonwoot! new glibc patches! nice work pedroalvarez! :)11:25
pedroalvarezjjardon: yup!11:26
pedroalvarezI tested it on x86_64, x86_32, ppc64, and armv7lhf.11:26
pedroalvarez jjardon did you finally tested the previous glibc patches with your work?11:27
jjardonpedroalvarez: no, I stopped when somone give them a -1, but I can test this new ones11:28
pedroalvarezjjardon: note, is going to cause a full rebuild.11:29
jjardonpedroalvarez: yeah. I will test them with new coreutils, systemd, xserver and gnome when they got merged11:30
jjardonI think you already have a +2?11:31
Krinhmm, downloading the baserock-current-delev-system-x86_64-generic, should it come with a baserock_src.img? or do i have to make one manually? and should fallocate create said image or do i just need to touch it in the correct dirrectory?11:33
perrylKrin: afaik baserock_src.img is just a ~30Gb storage image11:33
SotKKrin: you'll need to make it yourself11:34
KinnisonKrin: fallocate it11:34
perrylit is created with fallocate, or vboxmanage createhd depending on what virtual manager you use11:34
perryls/it is created/you create it/11:34
Krini get this result:11:34
Krinmikesmith@Kitsune:~/Downloads$ fallocate -l $((30*1024*1024*1024)) baserock_src.img11:35
Krinfallocate: baserock_src.img: fallocate failed: Operation not supported11:35
Krinis fallocate something that i ned to install?11:35
ssam2fallocate: I think you can answer that question :) if 'fallocate' is not on your system them you'll need to install it. But looks like it is there.11:37
ssam2i don't know why it's giving that error.11:37
richard_mawI do11:37
mauricemoss_Krin: I had the same problem last week. fallocate didn't work for me although it can be used with ext4. the only workaround for me was using truncate (which takes a long time because of writing zeros)11:37
richard_mawfallocate forces the allocation of the disk without writing to it11:37
Krinoo could you enlighten me richard_maw 11:38
richard_mawuse truncate instead11:38
richard_mawtruncate ensures that there's a file of the specified size11:38
richard_mawwhich is sufficient, as you can have lazily allocated files11:38
Krinis truncate it's own command or a subcommand of fallocate?11:38
Kinnisonits own command11:38
richard_mawseparate command11:38
Kinnisonrichard_maw: we recommend fallocate so that you don't get the unpleasant surprise when your VM which has "free space" suddenly can't write anything because the host disk is full11:39
richard_mawso our 3 options are 1. fallocate, which is fast and allocates everything 2. dd which is a complicated command, slow and allocates everything or 3. truncate which is fast but leaves it unallocated11:44
KinnisonBut fallocate is somehow failing for krin :-(11:45
richard_mawthe backing filesystem needs to support it11:45
KinnisonKrin: what FS do you run on?11:45
richard_mawkernel version also has an effect11:45
ssam2pedroalvarez: thinking about the cross-bootstrap problem, and the fact that merging glibc will cause rebuilds of everything, maybe now is a good time to fix morph to include architecture in cache key calculation11:50
ssam2i'll cook up a patch11:50
pedroalvarezssam2: makes sense to me11:51
persiaCan it be a complex inclusion, so we can distinguish between native arm-glibc, x86-built arm-glibc, and power-built arm-glibc?11:51
Kinnisonpersia: done right, it should be MORPH_ARCH for the host, the build, and the target11:51
ssam2persia: what do you mean by 'complex inclusion' ?11:52
ssam2target architecture is already set in the environment11:53
ssam2I was thinking of just including host architecture11:53
ssam2i don't have time to do a complex patch11:53
pedroalvarezI think that's the only thing we need11:54
jjardonHi, all my systemd fails to build with this error: "ERROR: System gnome-system has stratum specs that are not mappings." Any idea why?11:56
richard_mawjjardon: let's see your morphology11:56
richard_mawthe stratum specs are the entries in the list of strata11:57
jjardonrichard_maw: probably is something wrong there, but this happen if I try to build other systems too. Is this intended?11:57
richard_mawwe currently load all the morphologies into memory as part of the temporary build branch creation, so if you've got an invalid morphology in your working tree you can't build11:58
Krini dont know how to know what FS i'm on Kinnison 11:58
ssam2krin: run 'mount' with no args11:58
jjardonmmm, git clean -dfx solved the problem11:59
persiaKinnison: How are you distinguishing "host" and "build"?11:59
KinnisonWell in most things Baserock does, HOST and TARGET are likely to be the same11:59
* Kinnison is using the GNUish definitions11:59
jjardonrichard_maw: but only *~ files were removed11:59
ssam2TARGET only has meaning when building a compiler11:59
KrinKinnison, /dev/mapper/ubuntu--vg-root on / type ext4 ? so ext412:00
persiaAh, yes, we need all of BUILD, HOST, and TARGET12:01
* persia had not previously understood the madness of cross-compilation in sufficient depth12:01
Kinnisonssam2: Or any other code-generaty thing12:01
KinnisonKrin: looks like ext4 yes12:02
persiaErr, no, cross-compilation doesn't need that.  *canadian-compilation* does, but still, people do that.12:02
Kinnisonrichard_maw: does ext4 not support fallocate then? :-(12:02
richard_mawnot all versions, hence why I also asked about kernel version12:02
Kinnisonpersia: I have no intention of strongly supporting canadian cross for baserock TARGETs, but it may be necessary since a system you want to build might contain a cross compiler12:02
Kinnisonpersia: e.g. I'm on x86_64 and I want to build an x86_32 system which contains a compiler targetting ARM uC's12:03
persiaPrecisely, which should not be considered uncommon given the nature of Baserock and the desire to support build clusters.12:03
persiaAnd we can't rely on the results of a canadian compilation being the same for different BUILD architectures.12:04
Krinand now apparently virt-manager cannot allocate ram... yep it's a monday12:04
KinnisonOh dear :-(12:05
radiofreejjardon: configure: error: Package requirements (libdrm_freedreno >= 2.4.55) were not met:12:58
radiofreewhen building mesa, on arm, from master12:58
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]13:01
*** zoli_ [~zoli_@linaro/zoli] has joined #baserock13:02
*** jonathanmaw [] has quit [Ping timeout: 260 seconds]13:03
jjardonradiofree: thats strange, we are building your drm branch and thats 2.4.5613:03
radiofreeProbably not with freedreno support13:04
jjardonoh, yeah, we have to --enable-freedreno-experimental-api in our drm morphology13:06
jjardonradiofree: are you going to send a patch or should I?13:07
jjardonthis should be enough:
persiaThat seems a sane and tiny patch to me: +113:18
pedroalvarezI was going to give mine to have my place in the credits13:20
persiapedroalvarez: Is there a rule that there can only be two?  This isn't Highlander13:21
pedroalvarezthere isn't actually13:22
persiaWell, if you are quick, perhaps you can show support before the merge happens13:23
pedroalvarezjjardon: +113:23
* jjardon doesnt mind to put as many review-by as possible: more people to blame if something goes wrong ;)13:34
*** jonathanmaw [] has joined #baserock13:41
radiofreejjardon: you do it!13:42
radiofreeoh he has done it13:42
paulsher1oodif we're going to do full rebuilds etc, and change cache-calculation can we also do the genivi updates at this time?14:00
paulsher1oodpedroalvarez: ^^14:00
pedroalvarezpaulsher1ood: to prepare for the genivi release you mean?14:01
pedroalvarezI'm not sure about what changes we need to do for this release14:02
paulsher1oodnor am i. i can check14:02
pedroalvarezradiofree: is your baserock/james/unify-jetson-bsp-14.40.1 branch ready for review?14:07
radiofreei'll rebased against master14:10
radiofreethough it will need jjardon's mesa/libdrm fix14:10
radiofree(and assuming that works)14:10
pedroalvarezpaulsher1ood: please, let me know if you find anything useful wrt GENIVI14:17
* straycat doesn't want to start a bikeshed but...14:18
straycatwhen are we going to rename everything14:18
straycatI just wanted to explain some stuff, and felt quite ridiculous mentioning "chunk"14:18
straycatcan we just call them packages, or modules or something?14:18
straycatsomething similar to whatever everyone else is using perhaps?14:18
pedroalvarezI normally use the word "component"14:19
straycatalso a request from another channel about our site14:19
straycat14:20  Fuuzetsu$ straycat: can you put links to videos on your front page, the embedded stuff is a PITA 14:19
straycat                 if you don't have flash and I can run the links through youtube-dl or something…14:19
richard_mawSo Chunk => Component, Morphology => Definition.14:20
straycatSorry to be so negative, to counter that:14:20
straycat14:19  goibhniu$ ah baserock! sweet!14:20
radiofreepedroalvarez: sent for review14:24
*** jonathanmaw [] has quit [Quit: Leaving]14:24
paulsher1oodradiofree: latest rc is v3.18-rc4 now :)14:26
pedroalvarezradiofree: cool, building14:27
paulsher1oodpedroalvarez: i will14:27
straycatrichard_maw, I guess, idk really14:29
paulsher1oodstraycat: shall i add the video links, or will you?14:30
richard_mawwell, we need a straw-man to work from, I'd appreciate an RFC mail, even if it just helps organise the discussion14:31
*** jonathanmaw [] has joined #baserock14:32
paulsher1oodrichard_maw: for rename?14:32
paulsher1oodok, i'll do that14:32
paulsher1ood(since i have some quite strong opinions, as you know)14:33
richard_mawyep, and we should weigh your opinions as valuable too, since you've done a lot of explaining Baserock to people14:34
radiofreepaulsher1ood: good demo opportunity then ;)14:35
richard_mawI'd like to keep the discussion open until friday at least, since I'm sure petefoth will have valuable input too.14:35
franredrichard_maw, what is the priority on "Allow chunks with multiple source repositories" in your todo list? Could we do some minor patch to fix the submodules nighmare and then use that approach to do with multiple sources in one chunck?14:46
straycatpaulsher1ood, i can later14:48
richard_mawfranred: There's a reason we turned on the Vote button. I'll use the votes to prioritise work.14:49
pedroalvarezmy vote havn't been listened :)14:49
franredrichard_maw, I will vote :)14:50
richard_mawsorry, I must've missed it when I scrolled thorugh the list last wednesday14:50
jjardonmmm, probably Im missing something, but any idea why my interfaces network interfaces doesnt appear in ifconfig? only lo is present15:03
jjardonI run "qemu-system-x86_64 -hda gnome-system.img -netdev user,id=network0 -device e1000,netdev=network0 -enable-kvm" and I see the e1000 driver being loaded15:04
*** zoli_ [~zoli_@linaro/zoli] has quit [Remote host closed the connection]15:07
straycatjjardon, probably because the interface is down, try ifconfig -a  ?15:12
* paulsher1ood sends his rfc to the list, and ducks15:19
straycati've added video links, hopefully they're ok15:20
paulsher1oodlooks good to me. probably need to do the same on the videos page15:21
ssam2there are ducks interested in Baserock now ? (sorry)15:21
straycatssam2, ducks have been interested in Baserock for a while15:22
richard_mawssam2: nah, he sent ducks to the list15:22
franredthese ducks always seem to be ducked15:23
jjardonoh, of course, thanks straycat !15:25
radiofreejjardon: i do -net nic,model=virtio,macaddr=52:54:00:12:34:11 -net tap15:34
radiofreewhich is probably wrong these days... but that's what i'm used to15:35
* jjardon confirms network is working with systemd 217 (only a little configuration file is needed and you have network out of the box!)15:37
richard_mawwith networkd?15:38
jjardonas simple as adding this:
paulsher1oodwow, that looks simple :)15:40
jjardonwill try to send the patches today, can someone take a look to the coreutils patches? They already have a +1 by Sam and they are needed to compile systemd15:40
richard_mawthat's what I like about systemd, the complicated stuff is moved into the daemons15:41
paulsher1oodjjardon: i'm building your coreutils branch now - anything particular i should test?15:43
jjardonpaulsher1ood: not really, but remember to add the coreutils-common stratum to the system you build15:46
richard_mawwouldn't need to remember that if we had run-depends15:47
* richard_maw ducks15:47
* paulsher1ood is just building devel as jjardon intended it15:47
jjardonmmm, how baserock fill /etc/resolv.conf rigth now?15:47
richard_mawit's called by ifup, based on the contents of /etc/network/interfaces15:48
richard_mawifup is called by ifup@.service, which is run for every network device15:48
richard_mawbased on some udev rules, which I think are called 100-baserock.rules15:48
jjardonwhere is all that configured? it can be removed now15:48
richard_mawit pre-dates deployment and deployment-time configuration, so I think it's part of our delta against systemd15:49
rjekWhich web servers are ready-to-go in baserock's stock definitions?15:49
rjekISTR there was lighty at some point15:49
richard_mawlighttpd, but it's currently intertwined with the trove definition15:49
richard_mawhang on, I've got a branch to split that out15:50
ssam2jjardon: there's some stuff in the baserock branches of delta:busybox that installs the networking units15:50
ssam2would be good if you could remove them as part of your 'update systemd' branch15:51
jjardonrichard_maw: ssam2 thanks, I will try to clean that stuff15:51
straycati've also updated the ones on the video page15:52
jjardonwhat way do you use to know the delta in a specific module? seems we merge the upstream tag instead rebase against it in some modules15:57
richard_mawthe web system is a bit of a dumping ground for anything web related, and I don't think we've integrated any configuration for starting lighttpd independently of a trove for it to serve though15:58
richard_mawbut I did get cobbler mostly working with that web server, before discovering that cobbler is not flexible enough for baserock deployments, as you'd need to patch cobbler for every variation of baserock system15:59
* persia is generally opposed to dumping grounds, and would rather see dangling component definitions to strata with a usage model involving "delete the things you don't need from here. No, there isn't documentation on which bits interdepend."16:00
rjekrichard_maw: Thanks.16:00
richard_mawpersia: this is part of why I did the run-depends patch, you're more likely to know the functionality you want, so you can work out which chunks you aren't interested in, and can delete16:01
persiarichard_maw: Right (although I'm still trying to understand that completely).  That said, I just don't like the "delete what you don't want" model at all.  I prefer the "Add what you want" model, where w strive to make "what you want" to be sensible blobs of stuff, rather than 50,000 tiny packages.16:02
richard_mawWe'll need another way for discovery of what's available to include, browsing systems for something relevant and deleting what's not needed is an easy way, but well named strata and `ls strata` could be another way to do it.16:04
pedroalvarezjjardon: I guess you could do `git diff` against the latest version we merged16:08
richard_mawrjek: comedy option for web servers: busybox httpd16:10
rjekDoes that support gatewaying to HTTP-based application servers?  If so, it's probably fine! :)16:10
rjekApache would be nice, but I bet a ballache16:10
*** CTtpollard [] has joined #baserock16:10
richard_mawif by "HTTP-based application servers" you mean CGIs, then sure!16:11
rjekI don't :)16:11
richard_mawI've just posted the patch I did to split lighttpd out to the mailing list16:12
jjardonI think we can remove the ntpd stuff as well16:12
persiaApache shouldn't be that bad: most of the runtimes required for most of the modules are already present16:12
richard_mawconfiguring apache is an exercise in terror in my experience16:13
ssam2jjardon: I think we still want/need ntpd to run at boot if possible16:16
ssam2jjardon: but the units we have in delta:busybox are bad16:16
jjardonssam2: there is systemd-timesyncd for that since systemd 21316:16
richard_mawsystemd has built-in support for ntp now16:16
jjardonwill rebuild and restest after removing all these stuff16:17
richard_mawI'm not 100% sure about the details, I got the impression it will call out to chrony or isc-ntpd in some circumstances16:17
pedroalvarezI've just remembered I've found a bug some days ago16:21
persiaConfiguring apache is not so bad with automation.  Debian provides some decent single-host automation support for Apache configuration.16:21
jjardonpedroalvarez: tyeah, the inconvenience is that I have to clone the package for that16:22
straycatpedroalvarez, what bug?16:22
pedroalvarezwe are not checking that the name of a stratum matches with the stratum filename, or that the name of the stratum doesn't match with the name we put to the stratum in a system16:22
jjardonpedroalvarez: and counting that we branch from a tag, not a random commit16:22
straycatpedroalvarez, true but the name is irrelevant anyway16:23
pedroalvarezthis causes some random behaviour16:23
pedroalvarez(not random)16:23
persiarandom is unfortunate16:23
pedroalvarezbut the stratum with a different name is not built, and morph doesn't fail16:24
radiofreehmm i just did an upgrade to a system using my master branch and the fstab looks a bit buggered16:29
richard_mawit doesn't handle conflicting fstabs well16:30
richard_mawit should leave a copy of your old one around, but it's unfortunate that you have to fix it with the recovery console16:30
radiofreepedroalvarez: where's the pastebin script, do you have a link handy?16:31
pedroalvarezradiofree: yeah, you should have /etc/fstab.rej, and /etc/fstab.orig to see what haas happened16:32
radiofreein good news at least you now get a recovery console over both the serial and framebuffer console :D16:33
ssam2oh, that's good!16:33
radiofreei don't have a fstab.rej :\16:33
persiapedroalvarez: Could you add that link to the /topic?16:33
pedroalvarezpersia: the paste url, or the paste client?16:34
persiaThe URL that allows folk to store text which then becomes http-accessible and for which the link is provided such that people can put that in channel in lieu of pasting directly.16:35
richard_mawwe might want to instead have a link to our infrastructure on the wiki, so we can include the trello and stuff16:37
pedroalvarezLooking at other channels I think we can remove "Please note: This channel is logged"16:37
pedroalvarezrichard_maw: we have to create that page first16:37
franredpedroalvarez, I was going to ask you for that16:37
pedroalvarezfranred: remove that bit?16:37
persiaI'm not convinced that we want to rely on Trello, although I'm not opposed to folk who find it useful using it temporarily16:38
radiofreecouldn't get the paste script to work in recovery mode16:38
franredpedroalvarez, yes, it is contained on IRC logs:
radiofreethat's the fstab it created16:38
radiofreeshould be fixable, but there's no fstab.rej around16:38
pedroalvarezradiofree: and fstab.orig?16:39
radiofreejust that16:40
pedroalvarezradiofree: can you mount /dev/sda (or /dev/mmcblkp1) and check the versions of /etc/fstab?16:40
radiofreeback in after deleting the offending lines16:40
franredpedroalvarez, could we don't use:  for the IRC links adding a new Title "IRC tools" for them?16:40
richard_mawradiofree: is it missing the subvolumes then?16:40
radiofreein recovery mode i couldn't16:40
radiofreeit says it was already mounted16:40
richard_mawthen that's a kernel regression16:40
radiofreei'm back in now (i could switch to a previous upgrade etc...)16:41
pedroalvarezbaserock upgrades ftw16:41
radiofreeoh, it was me richard_maw16:41
richard_mawyou're supposed to be able to mount a btrfs subvolume multiple times IIRC16:41
radiofreecrappy options to it16:41
radiofreeso i'm in after fixing the fstab, anything you want me to check?16:42
radiofreealso, i just got a load of "[   37.013464] systemd-journald[245]: Failed to write entry, ignoring: Invalid argument" messages16:42
radiofreethere's no or .rej, just the fstab16:44
radiofreemaybe it thought it did a good job?16:44
radiofreehmm now i'm getting "no more space left on device" :\16:45
radiofreethere's 1.2G according to df16:46
radiofreeCloning into 'nouveau'...16:47
radiofreefatal: cannot copy '/usr/share/git-core/templates/hooks/post-update.sample' to '/root/newdrm/nouveau/.git/hooks/post-update.sample': No space left on device16:47
radiofreeif i manually copy that, then i get some btrfs errors16:47
richard_mawbtrfs filesystem df16:47
radiofreesorry that's the btrfs errors16:47
radiofree = df16:48
radiofreehave i fun out of space?16:50
jjardonwhat is the best morph stage to add a file? post-install-commands?16:51
richard_mawjjardon: if you need it at build-time, pre-configure-commands, otherwise I tend to use post-install-commands, yes16:52
richard_mawradiofree: it looks like it, though btrfs isn't great at reporting it16:52
radiofreewell i'm currently 1GB into scping something onto the device, so i'm guessing i haven't run out?16:53
radiofreeoh actually16:53
radiofreeit failed16:53
radiofreei'll delete a previous upgrade then16:54
radiofreeOSError: [Errno 28] No space left on device: '/tmp/tmpsMcLG4/tmpepXJls'16:54
paulsher1oodwhat's the problem with specifying build-system: for every chunk (in the stratum definition of the chunk)? istm this would be trivial, and allow removal of some unnecessary magic?16:55
radiofreeok that seemed to have worked, i think, even though it failed16:56
paulsher1oodjjardon: i have built and booted your coreutils devel system16:56
radiofreeat least it's not showing up in system-version-manager anymore16:56
pedroalvarezradiofree: do you have more free space now?16:56
radiofreeyes, after i git cleaned some things16:57
radiofreethe annoying thing is there's 6.6G free at the end of the drive16:57
radiofreecan i use that for / ?16:57
radiofree(this is my fault, i set the ROOTFS_SIZE var to 8G in the flashing script)16:57
jjardonpaulsher1ood: that would means we have to create a morph file for every chunk16:58
richard_mawradiofree: if it's not partitioned, you should be able to run `btrfs filesystem resize max /`16:58
ssam2paulsher1ood: possibly no problem16:58
paulsher1oodjjardon: no - we could add buildsystem: to the field in stratum16:58
radiofreerichard_maw: there's quite a few other partitions in the way16:59
radiofreei'm not sure we need them anymore though16:59
* radiofree will throw caution to the wind and delete them all16:59
paulsher1oodjjardon: repo:, ref:, build-system: etc ?16:59
jjardonpaulsher1ood: ah, yeah, Id be ok with that17:00
paulsher1oodrichard_maw: ^^ ?17:00
richard_mawradiofree: I can't think of a way to allow you to use that space for /, but that's only because we don't run btrfs device scan in the initramfs before mounting /17:01
jjardonthe "magic" to detect the build system was to avoid to fork every repository in the world when we used to have the morph file in the chunk repo: dont think is needed anymore with your suggestion17:01
radiofreerichard_maw: deleting all the partitions worked :D17:02
paulsher1oodjjardon: ah, yes. so we could do it then.17:02
radiofreeoh dear i did something bad17:03
paulsher1oodwe'd need a script to fix definitions.git again though17:03
jjardonpaulsher1ood: why you want to remove the autodetection? is that "magic" causing problems?17:04
* radiofree declares his rootfs now totally buggered and goes to run the flashing tool again17:04
* jjardon wants to type as less as possible ;)17:04
ssam2jjardon: the magic does incur a cost at build time17:04
ssam2well, build graph calculation time17:05
richard_mawjjardon: it slows down the pre-build step, as it has to inspect the working tree to see how it needs to build it17:05
paulsher1oodmauricemoss_: i see you now :-)17:06
* paulsher1ood sees a mail about windows on baserock-dev :)17:09
paulsher1ooddoes anyone here use windows vbox?17:09
pedroalvarezI'm asking here. Have we ever supported that?17:10
paulsher1oodnot explicitly. but vbox is usable on windows, and baserock as vbox guest should work17:10
rjekIs morph perhaps making assumptions about paths that don't hold on Windows?17:11
persiaI recall a discussion in January that implied that Baserock could run in virtualbox on Windows, but that one could not use any of the deployment instrumentation to deploy to windows-hosted virtualbox, and that one could not use a shared filesystem for /src17:11
radiofreeincredibly i managed to fix it17:11
persiaI don't believe those issues were ever fixed, but my memory may be faulty17:11
persiaradiofree: cool!17:11
ssam2the deployment location field is a URL for most extension types17:11
radiofreehowever i have 6GB of free space (after using fdisk to delete the partitions after it) however btrfs doesn't seem to want to use that17:11
ssam2and '/c:/foo/ is not a valid URL path17:12
ssam2I wonder if using URL escape codes will work. If not, I think it should be fixed17:12
persiaThat should start with "file:///" anyway, and there are semantics for that.17:12
ssam2persia: would do you mean by 'there are semantics' ?17:13
pedroalvarezmain problem: C: != /17:13
persiaWindows also has a POSIX compat library that can be optionally installed, which provides a sensible rooted global filesystem.17:13
radiofreeDisk /dev/mmcblk0: 14.7 GiB17:13
radiofreeone partition reported /dev/mmcblk0p1        49152     16826367   8G Microsoft basic data17:13
rjekI wonder what happens if he scps test.txt to GMMCALL@
radiofreebtrfs resize did actually work previously to get it up to that 8G17:13
rjek(ie, with the leading /)17:13
persiassam2: That there exists a set of rules to indicate that specific nouns are used to specify the device mapping in windows file:/// URIs.17:14
* paulsher1ood responds on the list in the meantime, suggesting use to deploy to self :)17:16
ssam2oh, good thinking17:17
* paulsher1ood thinks everyone should use, for most things, most of the time :-)17:18
* persia was happier using morph to do that directly, rather than having it abstracted, but agrees with the underlying concept17:19
*** sambishop [] has quit [Ping timeout: 245 seconds]17:20
paulsher1oodpersia: you mean a new 'morph build-and-deploy-in-an-idempotent-way' command?17:20
paulsher1oodurg... has anyone seen anything like this before?
ssam2that seems like a system-integration command trying to remove a file and being prevented from doing it17:24
paulsher1oodi'm using latest morph, on jetson building james' unified branch... with 'scripts/  systems/devel-system-armv7lhf-jetson.morph clusters/jetson-upgrade.morph'17:25
richard_mawthat's odd, it's not supposed to make anything read-only17:25
ssam2that linux-user-chroot commandline is enourmous !17:25
radiofreei'm not entirely sure the devel system u-boot will boot that kernel, but it's a good test :)17:25
paulsher1oodonly if i can build it, radiofree :)17:26
* paulsher1ood fears morph naughtiness has crept in17:26
Kinnisonlast time I saw a file URI for windows it was of the form file:///c|/stuff/thing17:26
KinnisonAnd some which still use the : instead of |17:27
radiofreepaulsher1ood: it works for me, using morph17:27
radiofreehold on17:27
paulsher1oodlatest morph?17:27
ssam2Kinnison: tells me that both are valid17:28
radiofreelatest morph is db759870b43782bf86801a6b55d14af24445b12117:28
Kinnisonssam2: cool17:28
ssam2I think the morphlib.util.containerised_cmdline() change is at fault17:29
paulsher1oodaha :-)17:29
ssam2but I don't understand how, exactly17:29
richard_mawyep, I'm just trying to work out why17:29
radiofreeare system integration commands run during the deploy or the build? i could try morph master and see if it breaks17:30
ssam2radiofree: during the build17:31
ssam2at system artifact construction time17:31
paulsher1oodradiofree: i think it is broken with morph master17:31
radiofreeguess so, that sha for morph worked17:31
richard_mawssam2: :-| it's the trailing / in the root path that's messing it up17:31
*** mauricemoss_ [] has quit [Quit: Leaving]17:32
richard_mawI changed it to fix up the root path earlier after I tested that bit, and I wasn't testing writability in the test suite17:32
ssam2oh, is this os.path.join()'s doing ?17:33
richard_mawdoes fix it?17:33
* paulsher1ood will check17:33
richard_mawssam2: no, it does string matching to determine whether an entry is in or is a writable directory, and it matches without the trailing /17:34
*** sambishop [] has joined #baserock17:34
radiofreenb all users of baserock-jetson-flash17:35
paulsher1oodin the meantime, i confirm that it worked with 1d6451363c92ec5466b01b5ba2fd327066343ab417:35
radiofreeplease change ROOTFS_SIZE=8something to  ROOTFS_SIZE=1503238553617:35
radiofreeon line 2817:35
radiofreethen you can run the btrfs resize command to get a 14G /17:35
paulsher1oodradiofree: probably better to submit a patch to the list?17:36
radiofreei don't think it lives anywhere? where should i submit it to, definitions?17:36
paulsher1oodactually, where is the source to baserock-jetson-flash?17:36
*** tiagogomes [] has quit [Ping timeout: 265 seconds]17:36
*** CTtpollard [] has quit [Ping timeout: 255 seconds]17:36
paulsher1oodrichard_maw: where should we put that? definitions/scripts?17:36
radiofreei fear it's not up to definitions/scripts quality17:37
richard_mawput what now?17:37
* richard_maw has lost track of what's going on17:37
paulsher1oodradiofree's script for flashing jetsons.17:37
ssam2definitions/scripts is not high quality :)17:38
richard_mawI was missing context because I was busy responsing to e-mails.17:38
*** ssam2 [] has quit [Quit: Leaving]17:38
* paulsher1ood is hurt by ssam2's remark :-)17:46
richard_mawI'm afraid I don't have time to work out why that isn't working today, I can promise to investigate tomorrow though.17:51
persiapaulsher1ood: I used to use `morph deploy --upgrade ...`, and I didn't want it to be idempotent, because I routinely didn't like the results and rolled my environment back.17:51
* persia hasn't been using baserock that actively recently, so cannot speak from immediate experience17:52
franredalso there are workflows which does not match with the idempotent behavior that does17:55
*** Krin [] has quit [Remote host closed the connection]17:55
*** mariaderidder [] has quit [Quit: Ex-Chat]17:57
*** jonathanmaw [] has quit [Quit: Leaving]18:04
*** wdutch [] has quit [Quit: Quit]18:05
richard_maw:-( it looks like the read-only system integrations slipped past the test suite because of a bug in the test command18:05
*** locallycompact [] has quit [Ping timeout: 244 seconds]18:10
*** petefoth_ [] has joined #baserock18:21
*** petefoth [] has quit [Ping timeout: 258 seconds]18:21
petefoth_ is now known as petefoth18:21
*** genii [~quassel@ubuntu/member/genii] has joined #baserock18:37
paulsher1oodpersia: allows rollback20:10
paulsher1oodfranred: for example?20:10
persiapaulsher1ood: From , it looks to me like it only supports rollbacks if I supply $newversion on the command line.20:12
franredfor example any deployment which modify /var and other directories which AFAIK are shared/copied between "factory" and Test and the deployment need a clean rootfs20:12
franredpaulsher1ood, ^^20:12
persiaBut if I'm doing that, it's just syntactic candy because I don't happen to want to bother remembering the upgrade syntax20:13
paulsher1oodfranred: so use a different cluster as your target?20:45
paulsher1oodpersia: it doesn't delete the system you're starting from in any case, so you can always rollback.. .unless i'm misunderstanding your point?20:46
paulsher1oodif you don't supply $newversion, $newversion=TEST20:49
franredpaulsher1ood, the problem is that update/upgrade/install an image on the same VM as Im developing, AFAIK, this images will share/update /var and other directories, so my rootfs will be updated - my devel and my test image would be not valid for me20:52
franredI discussed this with pedroalvarez to improve my workflow with my current work20:54
paulsher1oodfranred: not if you create or use another cluster21:05
franredpaulsher1ood, are you telling me that I can update a different VM using cycle? i.e #using a KVM+ssh cluster?21:06
paulsher1oodi think so, at least. 21:07
paulsher1oodmight need a manual step to delete your target vm each time21:07
persiapaulsher1ood: Maybe I am misunderstanding.  I thought that it defaulted to "TEST", so if you ran it twice, it would overwrite itself.21:08
paulsher1oodpersia: it defaults to targeting TEST. if you run it again, it will overwrite TEST (unless you are IN the TEST system, in which case it will advise you to boot into another system firs)21:09
paulsher1ood(can't overwrite the running system)21:10
persiaAh, OK.  That is less worrisome.21:10
franredhehe, paulsher1ood, if I need to remove my VM, it wouldn't be the same as a normal deploy via KVM+ssh?21:10
paulsher1oodfranred: yes. but cycle means you can build and deploy to it in one step21:11
persiaI still prefer the old way, because I use `git parse-rev HEAD` for version, but that only works if one commits more often than is typical.21:11
paulsher1oodyou could (for example) edit to add a line to remove your target vm if you're going to do it over and over again21:11
franredpaulsher1ood, well it is just add deploy at the end of build if Im not wrong. "morph build && morph deploy"21:12
paulsher1oodfranred: it runs morph gc as well21:13
franredI run it by default when I do any build21:13
paulsher1oodright :)21:13
persiaSomething is wrong if we have to manually run `morph gc` with any regularity21:14
paulsher1oodwe do21:15
persiaI know that it is a hard balance between having infinite disk space and supporting infinite cacheable artifacts, but there ought be some way to deal with the problem.21:15
persiae.g. a LRU disposal algorithm, or similar.21:15
paulsher1oodi think morph gc has somerthing like that already. but my previous attempts to get it included in morph build by default failed to convince folks21:16
persiaMy memory of discussion about `morph gc` was that it cleaned up rather more than I would have preferred.21:17
persiaAnd spent too much time trying to determine what was upstream, rather than what I happened to be using.21:17
paulsher1oodsoftware is never perfect, sadly :)21:18
paulsher1oodanyway, i'll stop trying to sell to you guys - you're experts. for novices, it's much simpler than trying to explain the intricacies of morph deploy21:19
straycat is now known as vletrmx2121:20
* paulsher1ood also notes, in passing, that he has been able to identify bugs in morph by using only cycle.sh21:20
franredpersia, morph gc does not only clean cache, it cleans, failed builds, and staging areas21:22
*** vletrmx21 [] has quit []21:22
persiapaulsher1ood: BY moving between versions of morph, or by exercising morph?21:23
persiafranred: Yes.  It didn't used to do so.  Some of that isn't actually very useful, but that is because of other issues with morph's UI.21:23
* persia remains deeply unhappy with gc, but cares more about other bits at the current time21:23
paulsher1oodpersia: just by exercising morph. today's bug in readonly system-integrations for example21:24
persiaMore folk should be using the upgrade feature to install stuff regularly, and using systems != reference systems.21:24
paulsher1oodi wonder if mason should do this by default, and upgrade itself each time :)21:25
persiaMason is a complicated concept.21:25
persiaBut I don't actually believe we want a long-lived Mason instance, as it becomes special.21:26
persiaRather, I think we want to spawn test harnesses built from master when there are candidates against master, and then use that to test the candidates (including testing the ability to build systems, and deploy test harnesses).21:27
paulsher1oodmaybe i'm thinking of something simpler :)21:27
* paulsher1ood has an early start... so heads to the land of nod21:27
persiaI don't see any reason not to test upgrades, but I also don't know how to model this elastically without dropping the idea of there being any system for "Mason",21:27
persiaIt ought all be git hooks21:27
persiaHeh.  Sleep well.21:28
persiaIs the thing that makes systemd do things called a "unit"?21:29
franredpersia, yes, it is21:30
franredbut they will finish in .service21:30
persiaThat's delightfully inconsistent :)21:32
* persia loves consensus nomenclature21:32
* franred goes home after checking that his rabbitmq is configured but he does not have any clue why systemd decides to kill it by timeout :S21:33
franredthere are something missing here but Im out of brain now21:33
*** franred [] has quit [Quit: Leaving]21:44
*** rdale_ [] has joined #baserock21:48
*** rdale [] has quit [Read error: Connection reset by peer]21:50
* persia gets annoyed at network connection drops, and decides to write more email later, rather than now21:59
* jjardon has a system with network and ntpd out of the box using only systemd stuff22:20
jjardonstill having problems compiling resolved (for /etc/resolv.conf), though22:20
persiaLack of dependencies, or something strange in the build environment?22:25
jjardoncompilation error:22:25
juergbijjardon: what compilation error? i've built it fine on x86-64 and arm (outside baserock)22:38
jjardonlet me get there, it takes like 30 min to compile systemd here ;)22:38
jjardon(it also fails if you compile it in more than one thread)22:39
juergbiah, i might not be here anymore in 30 min22:39
juergbibut could take a quick look tomorrow if you paste a build log22:40
jjardonjuergbi: will do22:40
juergbii also have a full log of a successful build here if that helps22:40
jjardonjuergbi: dont worry, I built systemd outside baserock several times, also its building fine in ;)22:41
*** straycat [] has joined #baserock22:42
juergbiwell, i meant for arm. but i take it you have the same issue on x86, then22:42
jjardonjuergbi: persia
jjardonjuergbi: oh, yeah, this is x86-6422:49
juergbioh, looking at that, it's almost strange that it works outside baserock22:50
juergbijjardon: ah, i suspect static_assert is not defined in the old gcc version baserock uses22:52
juergbiso it falls back to systemd's imitation of it and that is buggy22:52
juergbishould be easy to workaround by simply making those static asserts a no-op22:53
juergbinot sure how easy is it to properly fix it in systemd for these older gcc versions22:53
jjardonhopefully pedroalvarez will upgrade gcc after his glibc patches 22:55
juergbiah, maybe the new UNIQ macros could be used instead of __LINE__22:55
juergbiyes, that would be great22:55
jjardonwill see if I can patch systemd until that22:55
juergbibtw: the issue is in src/shared/macro.h22:56
juergbi#define assert_cc(expr)                                                 \22:56
persiafreeciv ran into this issue, and fixed it with some preprocessor code, but I don't understand the details.  It may also be possible to force some older gcc versions to use C11, even if this is not the default.22:57
pedroalvarezjjardon: it will take some time ;) 22:58
*** genii [~quassel@ubuntu/member/genii] has quit [Read error: Connection reset by peer]22:59
pedroalvarezAlthough seems that glibc is nearly there in this version 22:59
jjardon mmm, seems gcc has static assertions since 4.3:
jjardonpedroalvarez: only want you to feel the pressure ;)23:02
juergbimaybe only in c++23:02
juergbigcc 4.623:03
juergbithe static_assert define is in glibc, though23:05
juergbiit might actually just work with glibc 2.2023:05
* jjardon will rebase his branch againt the new glibc one and try again23:18

Generated by 2.14.0 by Marius Gedminas - find it at!