IRC Logs for #crux-devel Thursday, 2018-02-22

*** j_v has quit IRC00:00
*** j_v has joined #crux-devel00:00
jaegerso weird... I wish I knew what the affecting factor is01:50
*** chinarulezzz_ has quit IRC01:50
jaegerso far in my tests fakeroot is a clear problem but you're not having that issue01:51
j_vwell, no so far... i just recently uncovered my problem with my container builds in kvm, so was a bit preoccupied with that for a while01:55
j_vi have some stuff to do with my wife, but when i done with that, was going to start container builds inside kvm, and once that was going work on getting virtualbox running and try that01:58
jaegerSounds good, will be interesting to see01:58
j_vi'll start the virtualbox build now, because i don't have qt5 installed, and that will take some time unfortunatly01:59
*** pitillo has joined #crux-devel03:13
*** ___18VADBHIX has quit IRC03:34
*** ___18VADBHIX has joined #crux-devel03:35
jaegerok, filled out the rest of the sheet. fakeroot seems to be the problem06:09
jaegerodd that it's not on the physical hardware but is in VMs06:10
j_vodd indeed06:12
jaegeralso odd that I had no problems in the qemu/kvm VM with your kernel config06:12
jaegerso it still could be a kernel config option06:12
j_vhard to say. i have a series going now, building in containers on a vm. when virtualbox is done building, i'll mess with that. but given how much i have on this box, and how long it takes to build qt5, i may not get to it until morning.06:15
j_vi meant, how much i have going on this box06:15
j_valot running, krikey, maybe some day i will be able to say what i mean06:16
jaegerall good, no worries06:21
frinnstgrep "Building '/home/ports/pkg/go#1.10-1.pkg.tar.gz' succeeded." builds.log |wc -l07:08
jaegerWas that in the esxi VM? With or without fakeroot?07:09
frinnst2 threads, fakeroot07:09
frinnstyes, esxi 5.507:09
jaegerok, thanks. So odd07:10
jaegerI'll have to do some comparison of kernel configs, I guess, if I want to figure this out07:10
jaegerunless j_v beats me to it07:10
frinnstLinux femputer 4.9.79 #30 SMP Thu Feb 1 12:37:30 CET 2018 x86_64 Intel(R) Xeon(R) CPU E5-2620 0 @ 2.00GHz GenuineIntel GNU/Linux07:10
frinnstvm version 807:11
j_vi've looked at the configs some, though not as much as i had intended. one thing i noted was the CONFIG_NO_HZ* options and CONFIG_HIGH_RES_TIMERS option. was wondering if those might be related to what fun_ was mentioning earlier about how clocksource might be a factor.07:13
jaegerFor what it's worth, on VM1 I tested hpet, tsc, and acpi_pm clocksources07:14
jaegerall showed problems07:14
jaegerThat doesn't necessarily rule out some clocksource issue on the host, though07:14
j_vit wouldn't affect fakeroot, but the namespace and cgroup options in that config should rule out effective container building on there07:15
frinnsthave you contacted upstream? They must be (or risk) running in to the same problem with their buildservers07:15
j_vgood point07:16
jaegerI assume they're not, because they always say build in a container07:19
jaegerMaybe I'm mixing up docker and go, though07:19
jaegerwhichever the case, I'm tired, going to sleep07:19
*** j_v has quit IRC09:51
*** j_v has joined #crux-devel09:56
*** chinarulezzz_ has joined #crux-devel10:20
*** pitillo_ has joined #crux-devel10:57
*** chinarulezzz_ has quit IRC10:58
*** nthwyatt has joined #crux-devel10:59
*** pitillo has quit IRC10:59
*** chinarulezzz_ has joined #crux-devel10:59
*** fun_ has quit IRC11:24
*** fun_ has joined #crux-devel11:24
*** dlcusa has quit IRC15:56
*** dlcusa has joined #crux-devel16:02
darfojaeger: i get missing errors from go when built on my desktop
darfobut not when I build in a depinst in a container with nothing but the core ports installed16:08
darfoi don't use fakeroot16:08
darfoargh, that paste is truncated.16:10
j_vdarfo: there is a known issue that hasn't be completedly resolved16:15
darfofull log is at
darfoj_v: yes I know but I thought it was only seen in fakeroot, containers, or KVMs16:17
darfoor is it just that the .footprint in the port REPO is wonky and it is unclear what should go there?16:17
j_vthis patch gets rid of files that are causing that, but the footprint will need to be updated after building with it16:17
darfook, thanks16:18
j_vi think he's holding off on pushing anything until he is sure whole fix, but i'm not sure16:18
j_vi know that once i apply that change and update the footprint, i have been getting consist builds, but jaeger has seen some issues that i haven't managed to reproduce so far16:20
*** john_cephalopoda has joined #crux-devel16:21
darfoThe README in fakeroot has an interesting note about edge cases where IPC can fail and recommends using TCP.17:09
darfoThe default build of fakeroot uses IPC17:10
j_vnice catch... so can that be adjust on the fakeroot commandline or environment, or does fakeroot need to be rebuilt to adjust that?17:12
j_vman faked17:16
darfohehe. The README only mentions the configure option --with-ipc[=IPCTYPE17:32
j_vi doubt it will catch on, but i've done a little bit of messing around with yocto's pseudo, a fakeroot replacement, though not enough to see if it is better/safer than fakeroot17:36
j_vtruthfully, i've only built a handful of things with it, but i should try building texlive with it. i know that fakeroot plays havoc with texlive build.17:38
darfoAm I reading the log right? jaeger is using a fakeroot in a container under VMWare and that is where he has the inconsistent results?17:40
j_vi think the container builds were good. if i'm looking at it right, it's mainly fakeroot on some VM's (no namespace container) that cause failures17:43
darfook. I thought VMware might be having trouble with container namespace IPC memory.17:45
darfoLots of indirection for that memory :)17:46
darfoI don't bother with fakeroot. Too many moving parts. I just build in a throw-away container and then check the .footprint closely.17:47
darfobuild as root, that is.17:48
j_vi've building with containers for a while now... only issue so far is the mknod in core/filesystem, and that is only place i use fakeroot17:49
*** onodera has joined #crux-devel18:43
jaegerI hadn't pushed the go-build fix just because I was using all the hosts for testing builds... I'll get that done soon18:59
jaeger <-- final results are here, though18:59
darfoso the only thing common to all failures is virtualization of one kind or another?19:17
jaegervirtualization and fakeroot19:18
darfoI mean given that the only failures are when fakeroot is used19:18
darfojust for giggles can you run one of the failures with a fakeroot built with --with-ipc=tcp19:20
jaeger--with-ipc shows up in neither fakeroot's nor faked's man page here19:20
darfoNo, it is a configure option while building fakeroot19:21
jaegerOh, I see you said 'fakeroot built with', sorry19:21
jaegerI'll give it a try19:21
darfoit's a guess, maybe there is a problem with IPC virtualization19:22
darfoi used an LD_PRELOAD object to track files installed for years before I came on to CRUX19:24
darfobut I wrote the files installed to a file and just kept that for uninstalls19:24
jaegerfirst I'm going to push the go update because that's not the problem :)19:27
jaegerfakeroot with --with-ipc=tcp seems to break builds entirely19:35
jaegerat least on this VM19:35
jaegerIt's just been hung here:19:35
jaeger=======> Building '/usr/ports/pkg/go#1.10-2.pkg.tar.xz'.19:35
darfoit probably isn't able to connect19:36
jaegeryeah... will have to look into that more later19:36
darfoi am looking at the code for the preloaded .so19:36
darfoman fakeroot: -b fd  Specify fd base (TCP mode only). fd is the minimum file descriptor number to use for TCP connections; this may be important to avoid conflicts  with  the  file  descriptors used by the programs being run under fakeroot.19:40
darfobut the code never seems to use the exported value for -b19:45
jaegeroops, heh19:46
darfoI think the TCP method is/should be deprecated19:46
darfomaybe not. it may be just the -b option in the script and man page is old.20:08
darfoit is interesting that the go-build mismatch is a mix of missing and new20:56
darfothis might imply that the faked daemon is dropping some messages from fakeroot20:57
darfoor rather the from the fakeroot library20:57
jaegercould be, yeah20:58
darfohow many threads do you run on the VM(s)20:59
jaegerthey're all listed in the google doc21:02
darfooh, sorry21:05
j_vit's interesting to note that the fakeroot manual page discourages using fakeroot for building binaries, that it's better suited to handling the packaging stage. or that's how i read it21:23
darfome too21:31
*** onodera has quit IRC21:33
john_cephalopodaNever had any issues with it though.21:45
j_vI used to say that, and I suppose it's the mostly true for building a lot of software with fakeroot. When I found my texlive installation broken is strange ways, found that building without fakeroot fixed it, it was the straw that broke the camel's back. I worked hard to get my build system weened off from using fakeroot. It has presented many challenges, but in the long run I don't regret any of it.21:52
darfolooking at the code for faked I can see why it might fail on texlive21:56
darfoit doesn't check the return value of calloc() and malloc() calls21:56
darfobut it should cause a segfault21:56
darfonot sure what the foreground library thinks about that yet21:58
darfoshould get a broken pipe error21:58
j_vit's been a while since i built texlive with fakeroot. i suppose i could again, but for me it seems better to find ways that work without it than to fix it. just my one and a half cents.22:00
j_vbut i've done so much work to get building with containers to work that i suppose i've developed a bit of disdain for fakeroot22:05
jaegerj_v: regarding your earlier virtualbox question, I use it quite a lot for various tasks... I just use the binary release, though, I don't compile the whole thing22:20
j_vjaeger: thanks. I was looking over the current state of the contrib port. Looks like it could use some work. I'm trying my hand at it now.22:25
j_vWhat I am not too sure about it is how much I could disrupt anyone if I change it too much, so I guess I'll play with a local fork for a time to see how it goes.22:27
fun_Won't be simpler to just move the root stuff into post-install script for those 30something ports that needs to create special files/rights, so that we can build any port with any user without fakeroot or any other trick.23:10
pedjathe majority of upstream projects I dealt with consider failure to build with fakeroot a bug in their software23:26
pedjalibvirt devs coded, tested and pushed the patch that fixed an issue I had with it in a couple of days, iirc23:27
pedjait saved my sorry ass a couple of times when I fucked up (yes, that happens :) )23:32

Generated by 2.14.0 by Marius Gedminas - find it at!