IRC channel logs
2025-07-17.log
back to list of logs
<ekaitz>xtls: check the gnu/packages/commencement.scm to make sure <Kolev>I'm going to install Guix on my Fedora Chromebook now. <ekaitz>being those version numbers mes and nyacc <Kolev>csantosb, I can't find the binary for nsncd. <Kolev>ekaitz, I'm installing Guix on a foreign distro (Fedora). <apteryx>anyone knows how to deploy nginx configs to bordeaux? <apteryx>I've only ever taken care of berlin machines; but being able to edit bordeaux's config is necessary now that it hosts the website <Kolev>I'm making notes on how to deploy Guix to Fedora. <apteryx>in this case the machine hosting our website is currently hydra/bayfront.scm, IIUC. <sibl>I recently added my ssh config in my guix home config, but now when I try to connect to my server I get this: <sibl>Jul 17 10:19:41 lexipine auth.info sshd-session[11837]: Authentication refused: bad ownership or modes for directory /gnu/store <sibl>Jul 17 10:19:41 lexipine auth.info sshd-session[11837]: Connection closed by authenticating user azz 192.168.0.175 port 57876 [preauth] <sibl>did anyone get a similar issue ? I could find that kind of issue in NixOS blog but couldnt find about Guix <Rutherther>sibl: so how did you configure it with guix home? <Kolev>Still waiting on `guix pull`. <Kolev>Is something wrong with `guix pull`? My laptop battery might die by the time it finishes. <squid64>Kolev: I guess it just uses a lot of resources? Maybe you haven't done it in a long time? <Kolev>squid64, just installed Guix on {distro}. <squid64>Kolev: maybe plug the laptop in and do a guix pull and next time try without plugging it in. <squid64>As the disto you download are usually older versions <squid64>So could take quite a bit of time on first update <parnikkapore_x>Morni'n, Guix! I'm trying to update python-pygobject; apparently, the gobject-introspection-tests typelibs are being built with the shared-lib path wrongly prefixed with #$output (the shared library is in the build dir). What's the best way of dealing with this situation? <sibl>Rutherther: I used `home-openssh-configuration`, for `authorized_keys` I used a (list) of `plain-file` (easier if I just wanna drop the config file to another server) <sibl>did anyone setup their ssh authorized_keys with guix home and have it working ? <sibl>whats weird is that the same config on a Guix server works, but on this one (Alpine Linux) it says wrong permission <sibl>does it have to do with `StrictModes` in sshd_config ? <sibl>but even that wouldnt make sense, the authorized_keys file in the gnu store is owned by root and read-only for everyone <sibl>oh I see, in my .ssh/authorized_keys the file actually set as rwx for all <sibl>but thats the same on the Guix server and sshd isnt complaining <sibl>and the logs actually complains about /gnu/store not the rest <apteryx>cbaines: hello! could you (or anyone already admin on bayfront) please set my user (maxim) password? I can't reconfigure (can't sudo). <sovrcat7>Was calling stumpwm with `(,stumpwm lib) <apteryx>what's the status of the core-packages branch? it's been blocking the merge queue for a while on QA <csantosb>Kolev: guix pull is slow when you do it for the first time; then, it caches previous pulls, and pulling is faster <sibl>is it expected than when I make a custom guix image with custom channels, those channels arent integrated in the image ? I still need to guix pull to get them <sibl>why arent they already setup in the default guix running config of the image ? <sibl>because now when I do `guix system init /run/current-system/configuration.scm /mnt` it doesnt work because it doesnt know the imported package <sovrcat7>Ohh it looks like it's because I don't have the right localtime, mine says Lwos_Angeles instead of Los_Angeles <sovrcat7>But I think I chose that from the graphical installer, I wonder if there's a bug <sovrcat7>Also my /etc/timezone points to America/Los_Angeles <ekaitz>we don't have that string anywhere in the codebase <sovrcat7>Bruh, it's a typo in my config.scm. Sorry <ekaitz>don't say sorry, Lwos_Angeles sounds better <sibl>I see `guix pull` will soon allow to select from which channel to pull, to avoid pulling all of them, and I saw it has been reviewed but cant see if it has been merged or when it is planned to be added <cbaines>apteryx, I can look at setting your password on bayfront if that's still needed? <sibl>non-gnu can change the behavior of the guix binary ? <sibl>seems like it will still pull `guix` anyway even when specifying which channel anyway... <apteryx>me wonders what happened to www.dicollecte.org; seems their website has been defaced <apteryx>it's full of adds and the links do not work <cbaines>apteryx, I've set your users password now and put it in your home directory <apteryx>is it OK to reconfigure and restart nginx nowÉ <cbaines>yes, I believe so, I've done that on bayfront recently <cbaines>apteryx, I generally do herd stop nginx, check it's actually stopped, then herd restart nginx, just in case <sovrcat7>What's the suggested way to set the default X11 font in guix? <cbaines>I've had it before when there's been multiple instances of NGinx running at once, and that produces a wierd behaviour <apteryx>seems it was reconfigured on a custom commit <apteryx>guix system: error: aborting reconfiguration because commit 95d88456844fc460fd1708c0fa1e04ab473af0ba of channel 'guix' is not a descendant of 938dbd33e93f7dcc52a1155271c1fb81a1b2fcef <apteryx>ah nevermind, I had to relogin to have the right guix <apteryx>maybe 'hash guix' would have worked too <civodul>i think ‘nginx -s stop’ doesn’t quite work <cbaines>civodul, I've deployed the latest data service commit to hydra-guix-130 now <janneke>ekaitz: headsup, xtls' bug triggered me to try to compile tcc-mes-0.25.3 using newer versions of nyacc <janneke>i found a problem, however it looks to be a different one, compiling libtcc.c instead of tccpp.c <janneke>ACTION pushed a fix to master and rebased wip and wip-gcc4 <janneke>ACTION is at loss how "we" missed that, but yeah <sibl>how are channels handled ? everytime a package get updated, someone rewrite the package.scm and propose a pull request with their commit ? <janneke>a [p]match error in compile.scm, i.e., nyacc >= 1.02.0 uses abs-ptr-decl instead of abs-decl <ekaitz>uh... classic nyacc breaking changes... <janneke>used in the declaration of command-line options, things like: { offsetof(TCCState, warn_unsupported), 0, "unsupported" }, <janneke>using offsetof, so not something you'd find unless you really compile all of tcc <gabber>in a python package running `tox` does not find other dependencies of the package (so tox tries to install them with pip and fails in the guix build process). this is not reproducible in a `guix shell --pure --container` environment. any ideas? <csantosb>I understand you're writing a package definition, which uses tox; and the problem is tox cannot find the dependencies which are already included, right ? <csantosb>Regarding `pyproject-build-system`, I remember that `python-setuptools python-wheel` are mandatory as propagated inputs, is that correct ? <Rutherther>csantosb: setuptools and wheel are used during build, why would you propagate them? they go to native inputs <csantosb>Sure, native, sorry; I just noticed that some packages don't include them, and they should, right ? <gabber>csantosb: yes. well i'm trying to make our mailman package run (the adequate) tests. which can be done by invoking `tox -e py311-nocov`. we currently do not run tests and running this in a `guix shell -D mailman python-tox --pure --container` yields pretty ok results (even though output color is all red) <Rutherther>csantosb: if they build, no, why would they have them? <gabber>why does my bare-bones.tmpl raise "no value specified for service of type 'dhcpd'" when just exchanging dhcp-client-service-type with dhcpd-service-type? does the latter need some arguments? <luca>Hi, does anyone have a working gnome-keyring setup working? (without running the gnome desktop environment) <csantosb>gabber: guix shell -D mailman -C python-tox python-nose2 python-flufl-testing -- tox -e py311-nocov ? <gabber>csantosb: my manual command works just fine. when i try to reproduce that in a package definition (by re-enabling tests and changing the 'check phase to invoke said command) it fails miserably <csantosb>I understand; previous command tries to download with pip for me, "py311-nocov: install_deps> python -I -m pip install 'flufl.testing>=0.8' nose2" <csantosb>So "ERROR: No matching distribution found for flufl.testing>=0.8". Extrange. <gabber>that's exactly what i am getting <gabber>i guess if tox /knew/ about the dependencies (that *are* there) it wouldn't try to install them separately <csantosb>Exactly: "ERROR: No matching distribution found for nose2". It cannot find the deps installed. Path problem ? <sibl>I get this error when trying to setup podman-rootless on a headless server: guix system: error: service 'cgroups2-limits' requires 'elogind', which is not provided by any service <gabber>sibl: i guess between writing the docs and now elogind was removed from %base-services. <gabber>try adding it manually to go forward <sibl>I just did, I will see how it goes when generating the image, thanks ! <gabber>and if you wouldn't mind, please open an "Issue" on codeberg, describing the issue. this is either a service or (more probably) a documentation bug <futurile-afk>gabber: did you figure out the dhcpd thing, I'm also getting that warning but keep ignoring it <apteryx>off, the GNU ftp currently serves our guix-binary installer at <100 KiB/s <sibl>what issue do you have with dhcpd ? I also got one when trying the image, maybe I can help <gabber>guess i'll write the mailing list (and then maybe some patch for the docs eventually) <gabber>sibl: guix tells us that dhcp-client-service-type is deprecated, and to use dhcpd-service-type. but that unfortunately seems not to be a drop-in replacement <sibl>ah I see, I got the same issue <futurile-afk>yeah, they don't actually seem to be the same thing at all, dhcpd-service-type is running a dhcpd server (I think), It's off <sibl>I went with wpa-supplicant and network-manager which didnt give me issue, because dhcpd required some kind of config which I didnt have <gabber>huh. dhcpd needs a config-file and does not provision 'networking? <jnms>Hi, binaries compiled for other distros (with glibc) use the interpreter at "/lib64/ld-linux-x86-64.so.2". I wanted to symlink the guix ld-linux there, and get it to use /run/current-system/profile/lib as fallback path. To be compatible with nodeflib, only LD_LIBRARY_PATH can be used, but as that's evaluated *before* DT_RUNPATH, setting it globally it breaks most guix programs. <jnms>I made a glibc patch to introduce LD_FALLBACK_LIBRARY_PATH=, which is only used as a very last resort. (I only copy lib64/ld-linux-x86-64.so.2 from the output to /lib64/ld-linux-x86-64.so.2, and export LD_FALLBACK_LIBRARY_PATH=/run/current-system/profile/lib). <noe>Rutherther, does it exist as a guix service? <jnms>Rutherther: I didn't, thanks. But nix doesn't put libraries in the profile /lib so it has to be a lot more involved than my ~50 line glibc patch. <futurile-afk>we're on a run of people asking about running compiled binaries / or libs they've installed through a package manager at the moment <yelninei>janneke: with core-packages-team being close to getting merged how should the 64bit childhurd configuration look like? Just a binary #t/#f or more general for future ports. Also how to require "noide" in the kernel options? <janneke>yelninei: good questions, i was actively trying to postpone them until core-packages-team was merged, so that we could work on that while having substitutes <janneke>it's probably better to use (system x86_ <janneke>...something like (system x86_64) [or (arch/cpu/machine x86_86)] than a 64bit #f/#t <janneke>although we're hopefully going to see an aarch64 and riscv64 hurd some time soon, i don't think upstream is there yet <janneke>because hurd is [still] such a nice, and pretty unuable, i believe it may be overkill to have more than one architecture; possibly we should drop i586 as soon as x86_64 is up and running, wdyt? <yelninei>janneke: I have build almost everything from the hurd manifest for x86_64 rn. Some tests hang the entire process (see issue 1041 for details) , not sure what is going on there, but everything else is fine which is amazing <yelninei>not sure about removing i586 so soon. It feels a lot more stable right now (is anyone expecting stability?). Maybe with the next glibc update. What are the plans for moving to 64bit only in debian? <jnms>How are you testing hurd? The instructions in devel-hurd64.tmpl on master don't build. <yelninei>i am building from core-packages-team branch , what does not build on master? <efraim>how do I run the tests from just one scm file in tests/ ? <janneke>yelninei: "not sure about removing i586 so soon", sure! let's try to avoid that, we need a good and stable hurd <janneke>ACTION hasn't heard of debian's plans to retire i586 <yelninei>jnms: also try the bare-hurd64.tmpl which has fewer dependencies <Kolev>How do I recover disk space from Guix? I'm limited disk space. <yelninei>ah that also does not build because something pulls in slirp4netns and libcap <yelninei>janneke: For the guix-packages (in the manifest, devel-hurd.tmpl) ) , we'd need to delay the evaluation of (package-direct-inputs guix) s.t. it only gets evaluated when (target-hurd?) is #t to avoid linux-specific inputs (slipr4netns) <yelninei>jnms: or add slirp4netns to the things to remove in guix-packages in devel-hurd.tmpl <yelninei>jnms: Also there is a typo in devel-hurd64. The image type is hurd64-qcow2. with hurd-qcow2 youd get a 32bit image <bdju>Is our Noto Color Emoji font out of date? I've noticed certain emoji like 🥳 (partying face) only show up as a thin black and white outline in my terminal. Although in bemenu I see it in color so I must have some other font that can show it that foot isn't using. <bdju>Kolev: You can run a guix gc to get rid of old versions of packages in the store. <identity>bdju: some programs might not support colour emoji properly, i think <bdju>That is not my issue, I see other emoji fine in color in my terminal. <bdju>I'd say the vast majority show up fine, so I figure it's just newer emoji that got added to unicode later that are likely the issue, perhaps due to an out-of-date font. <bdju>Also in bemenu where partying face does show in color, it's clearly a different "style" <identity>then it may be a font priority issue; noto-emoji seems reasonably up-to-date (latest tag from the github repo) <janneke>sneek: later tell yelnieni, yes that or -- and that's what i pushed -- we also filter out slirp4netns <janneke>jnms: should be fixed on master, thanks for letting us know! <orahcio_>Hello, someone can help me with a question? Why iwd-service-type is not a networking service? <somename>Hello, is it possible to work with device drivers on guix? I am trying to build a "hello world" kernel module but /lib/modules doesn't exist on guix. So I can't find the directory to build the module in. Any pointers are appreciated :) <janneke>somename: don't really know how that would work; i do know that modules live in /run/current-system/kernel/lib/modules <janneke>where kernel is a symlink to a profile in the store <somename>In the tutorials they refer to the folder /lib/modules/$(uname -r)/build. It doesn't seem to exist on Guix. Is device driver development supported/possible on Guix? <janneke>it's certainly possible, not sure if it's supported <janneke>you'd have to add your package with your driver to the kernel profile that is used in the system and i don't know how to do that <jnms>somename: You can find /lib/modules in the LINUX_MODULE_DIRECTORY env var. But that's read only of course. <vagrantc>somename: there are a couple DKMS packages for building out-of-tree modules <jnms>The /run/booted-system/kernel/lib/modules dir I mean <gabber>futurile-afk, sibl: i guess the dhcp thing can be resolved when using *the correct* service as a replacement. which is called dhcpcd (note the c before the final d) <janneke>somename: ah, there is simply kernel-loadable-modules field in the operating system definition, and there is also a linux-module-build-system <janneke>so you look at the gnu/packages/linux.scm module for an example and add your package to your operating system definition <somename>janneke: Ah, thanks! I'll have a look :) <janneke>and then, of course, reconfigure your system -- good luck! <bdju>Kolev: That should work. <jnms>somename: Try copying from something without many arguments, like `guix edit hid-wiimote-plus` ^^