IRC channel logs
2023-01-31.log
back to list of logs
<yewscion>Hello all, I'm having an issue with Cuirass that I've run into once before. I'd thought it was an issue with my previous installation, but I've just noticed it happened again: Cuirass is, all of a sudden, no longer building any new evaluations. <seninha>hi, i found a bug in the docs, in the var `cert` at openvpn-client-configuration, it says the default is `/etc/openvpn/client.crt`. for `openvpn-server-configuration` the default is also `/etc/openvpn/client.crt` (it should be server.crt) <seninha>it seems that it was copy-pasted from the openvpn-client-configuration into openvpn-server-configuration <seninha>wait, in /gnu/services/vpn.scm both configurations are defined as the same thing? <seninha>yeah, the cert and key configurations should be split into the server- and the client- specific configurations for different defaults <dcunit3d>does anyone have experience trying to use doom emacs while loading some packages through guix? <dcunit3d>i know it's a pretty bad idea. at this point, i would love to get away from doom. guix makes things so easy. when i run into issues with doom, it's either super obvious or it's devious. <dcunit3d>i feel like doom would still be useful as a source of configuration ideas when using the emacs-xyz packages, but that if i didn't use it, that it would be much easier to maintain multiple configs and troubleshoot. <shcv>hello! is there a simple way to set a guile function to run on a matching udev event? I'd like to disable my laptop's internal keyboard and change layouts when the external is plugged in / vice versa when removed <jgart[m]>dcunit3d: start small and build your config up <shcv>jgart[m]: looks useful; I guess the next step is figuring out how to write a pure guile shepherd service <shcv>jgart[m]: maybe? I guess I could also use the default "udev" service, though it looks like it's just adding strings to a configuration file, and therefore less guile-centered <bumble[m]>s/dbus/elogind/, s/note_1340995/note\_1340995/ <VesselWave>Hello, could someone help me? I am trying to use TRAMP with ssh. Client and server run guix. I can ssh via cli, but Emacs just stucks when I try to open a file: /ssh:user@192.168.x.x:1.txt <abrenon>the installation part of the documentation seems to be ok with ext4 filesystem for the store <abrenon>it doesn't recommend any fine-tuning in particular, especially regarding inode ratio <abrenon>but my disk is now regularly saturating its inodes when I try and update my user packages <abrenon>has something changed lately which could account for a higher use-rate of inodes ? <Htop170>Hey guix, I'm trying to package morfologik-fsa, wich is located in morfologik-stemming (https://github.com/morfologik/morfologik-stemming). Since I only need the morfologik-fsa directory, is there some particular method or git-fetch strategy I should use? Or is it better to download the whole repository and navigate through it/potentially delete <rlp10>Is asking about comparisons between Guix and Nix considered off-topic? <abrenon>Htop170: all parts can be built separately / have their own building scripts ? <Htop170>abrenon: I think so! Every parts have their own pom.xml file <abrenon>Htop170: I wonder if that's not a case where the several possible outputs of a package would make sense <abrenon>but I have never used it yet sorry ^^' <abrenon>so to answer your question I think that would be download the whole repository and make a separate output by stemmer you are interested in <civodul>rlp10: hi! i'd say it's on-topic :-) <Htop170>Thanks! Are you aware of a package I good look the definition of, doing what you said ? <rlp10>Thank you, civodul. I'm trying to decide whether to put time into learning nix or guix first. I recognise there is going to be a significant effort either way. Any suggestions as to which one is more accessible? <abrenon>hmmmmm I'd say pandoc top of my mind, but it's documentation/static library/the rest, so not as symmetric as what you want to do, but I guess I'd start there if I had to look at a multi-outputs package <abrenon>anyone knows of a better example ? (please ?) <iyzsong>rlp10: that depends on many factor, i'd say if you like lisp/scheme/emacs then guix is a better choice. <rlp10>iyzsong: Thanks, I definitely like s-expressions: I use Emacs as my main editor and have some experience with Racket. <rlp10>I guess I'm worried that Nix might have some innovations that Guix lacks e.g. flakes? (Note, I don't even know what Nix Flakes are, just that I've heard of them!) <iyzsong>nix has bigger community and more packages/hardware support, while i think guix has more well-designed core layers. <iyzsong>guix time-machine with channels (see 7.3 Replicating Guix in manual) is equal to nix flakes. <abrenon>(oh, so this is what those flakes everyone mentions are about ? thanks !) <abrenon>I'd like to mention how the nix configuration language seems inelegant and painful to write compared to guile <Htop170>abrenon: thanks for your answer! I will look into that <civodul>rlp10: most of us here are going to be biased :-) maybe you can browse the manuals and tutorials of both to get an idea of what using it is like <abrenon>there's also openjdk which has the same structure <abrenon>retrying now that the chan is a bit less empty: any opinion on the inode ratio for the store ? <rlp10>iyzsong: Thanks, that's very helpful. On that basis, I probably would prefer Guix, because Nix seems very difficult to understand. <rlp10>rekado: Thank you for that, I'll check it out <seninha>hi, in `/gnu/services/vpn.scm`, the default for the `cert` variable in `openvpn-server-configuration` is `/etc/openvpn/client.crt` (it should be `server.crt`). the same goes for the `key` configuration. it happens because of the define-split-configuration defining both the client and the server configuration at the same time. <rlp10>In Guix is there any support for maintaining a profile of software for several people across different machines? I have a small office and would like everyone to have the same basic software available <abrenon>rlp10: I suppose you could version and share a common manifest file (~list of packages) <rlp10>I see, and perhaps I could keep that in a git repository such that users could pull and update it - or something like that <janneke>rlp10: we've been doing something like that <rlp10>Oh OK sounds good. How have you found it? Has it been difficult? <janneke>although now, we have a different manifest for every project <janneke>moving to guix for dependencies has solved a lot of problems, but some people who use guix on a foreign distro still manage to break their guix setup sometimes :) <janneke>civodul: people can be very creative <janneke>but mostly changing .bashrc to change environent variables and break guix environent/shell <efraim>The ones I've seen are LD_PRELOAD, GUI packages (including env vars like janneke said) and installing packages like sudo <efraim>quick! who wants to put together a presentation for Guix Days "How to break your <foreign distro> with Guix in 5 easy steps!" <janneke>yes, running `sudo' on every command that gives an error gives interesting results too <seninha>civodul: i mentioned you because of that bug i found: client and server configurations defaulting to the same certificate and private key files in openvpn <efraim>mismatched kde dependencies made it hard for me to test kmail on aarch64 with guix on arch <futurile>Q: I have a package (rust-heapless-0.5.6) that inherits from Guix's rust-heapless-0.5. I want to add the argument #:tests? #f - how do I do that? I tried substitute-keyword-arguments (package-arguments rust-heapless-0.5) ((#:tests? #f)) - but it says 'source expression failed to match any pattern'. I guess the issue is Im 'adding' not 'substituting'? <efraim>I'm guessing xf86-video-vmware should be limited to x86_64 and i686 only, but the build failure on aarch64 doesn't suggest there's an architecture problem <janneke>efraim: might be an interesting section in the manual/cookbook, short recipes to break your guix <civodul>futurile: you can try: (arguments `(#:tests? #f ,@(package-arguments rust-heapless-0.5))) <civodul>this is less robust than using 'substitute-keyword-arguments', because you could end up with #:tests? twice <civodul>the "failed to match" error probably means there was a syntax error in your use of 'substitute-keyword-arguments' <futurile>civodul: OK, thanks - that works - appreciated. I'll play around with `subtitute-keyword-arguments`. Need to learn some/more Guile heh <gabber`>my relatively lean system config fails to build due to python-matplotlib (which seems to be a dependency of font-abattis-cantarell). i don't use gnome or mate (which seem to depend on that font). <gabber`>it's /gnu/store/yrclrqxr5jpndsqakjjqyqlqyl1kbik2-python-matplotlib-3.5.2.drv which fails to build. can others reproduce that failure? <gabber`>how can i find out what caused the latest derivation of that package? the package itself seems to be unaltered since may 2022 <abrenon>gabber`: not the same guix version (because different hash) but here I've just built 3.5.2 without any hassle <abrenon>(/gnu/store/gk9s8b30vdq7mm4h52jgvc052sxqvrif-python-matplotlib-3.5.2) <abrenon>I updated yesterday, I'm on guix 595b53b <gabber`>abrenon: thanks! that seems to work! the CI has not gotten to rebuild the font but i guess something between yesterday and now broke python-matplotlib <abrenon>oh I wonder what that is, thanks for reporting it <abrenon>wow a failed test ? I have no idea what that means, sorry <seninha>hi, how can i pass a custom option to a daemon? <nckx>abrenon: I'm not sure what you're looking for, but a 123G store here is 7.6M inodes... I think. <efraim>I personally recommend btrfs for the store partition, hasn't given me problems yet <nckx>janneke: "Changing .bashrc" to get Guix things to work? <nckx>Same, I haven't used an fs with 'inodes' in a very long time. <nckx>This server is an exception: <nckx> /dev/mapper/root 26214400 6233304 19981096 24% / <janneke>nckx: for any sort of "reason", add ~/bin or something to PATH, setting GUIX_ENVIRONMENT; seen it all <nckx>Aha. I was mostly fishing for that 'set XXX' Guix message that some people interpret as 'add XXX to all your shell configuration files forever'. <nckx>No, I'm glad there are legion ways users get creative in breaking their system. <gabber`>seninha: what do you mean, exactly? what are you trying to do? <gabber`>(where) should i report issues with the IRC-logger? <gabber`>ACTION will come back here after a (hopefully) quick and successful reboot <nckx>ACTION guesses that the issue is ‘it's dead’. <seninha>gabber`: nvm, `extra-options` was what i'm looking for. just have to check whether it works for all daemon services <nckx>gabber`: Is this about the search? <nckx>seninha: It's not generic. Some services provide a field (the name may or may not be ‘extra-options’) like that, some don't. <seninha>nckx: i need to pass the `--engine=myengine` to openvpn server service <seninha>there's no engine configuration, tho <nckx>The OpenVPN service isn't currently one of the best-written ones, probably just because it's relatively old. <nckx>There's no way to pass a regular configuration file, for example. <seninha>i'm really new to guile and guix, so i think i cannot send a patch for it and for that other bug on defaults <nckx>I wonder if some kind of generic ‘service-command’ record, that also handles arguments, would be a sensible replacement for the ad hoc ‘package’ fields that many services have. Instead of (package cups) you'd write (service-command-or-whatever [(package cups)] [(binary "sbin/cupsd")] [(arguments '()]). <nckx>In better news, Guix IRC search now finally searches beyond 2021. \o/ <abrenon>nckx: thanks for your answer ! I gotta go now but I'd love to discuss that <abrenon>I have ~6M inodes for 90G, so I guess it's quite similar <abrenon>don't you ever hit any errors when reconfiguring ? <abrenon>how, sorry, discuss has a different meaning in english <nckx>No, that's correct in English as well, I just assumed an underlying ‘discuss an issue’ meaning that's common here. <abrenon>right now my 90G is only 60% full, so quite ok, but that is 82% of inodes ! <nckx>Like, you don't really discuss facts like ‘files need inodes to live’ :) <abrenon>no, I wanted to discuss the issue of my running out of inodes : ) <abrenon>(when using guix package to update my user profile from a manifest, I witness a sharp increase in inodes, which I then compensate by deleting the older generation and garbage collecting) <nckx>I think my earlier count was inaccurate (and too high). I excluded .links, but not actual hard links between store items. I'll do a better. <abrenon>anyway, thank you so much for your catching my question in the limbo of the logs <nckx>Sure, but that's not just your store. <abrenon>and sorry for running away once someone is there to answer it <nckx>The total/used/free for my entire / is: /dev/mapper/root 26214400 6233269 19981131 24% / <nckx>It certainly exposes my ‘7.6M’ answer above as a lie. <nckx>I'm currently running ‘find /gnu/store -printf '%i\n' | sort -u | wc -l’ which is probably adorably naive but should be more correct (…right)? I'll post the result here if it looks plausible. <nckx>Which sounds about right compared to the total. <seninha>openvpn service is failing because of an unset command-line option. how can i change its configurations to use a few more options from my /etc/config.scm? <gabber`>soooo, seems my little reboot attempt wasn't as successful as i had hoped. i ran into a nasty OOM/kernel-panic (on 6.1.8 libre-kernel with an Alder-Lake architecture). this is the wrong place to ask, but should i report this to.. the linux community? the linux-libre project? the machine boots just fine with 6.0.8 <gabber`>nckx: yes, it was about the search -- searching for some (?) nicknames only results in relatively old matches <mirai>seninha: inherit/make a copy of the original service-type and change it <apteryx>gabber`: likely to the linux project <gabber`>i tried the vanilla kernel and have booted successfully -- and reported the issue to the linux-libre project <rekado>nckx: re IRC search: what was the reason for the broken search? <rekado>I assumed but never verified that it’s due to separate mount namespaces <rekado>(search index file written outside of the mount namespace where goggles reads it) <gabber``>does anybody here have a Sway configuration handy? do i need to change more to my system configuration than passing (wayland? #t) to gdm and adding sway to the system-wide packages? <bjc>setting ‘wayland’ to #t was enough for me <bjc>you will need to install sway in your system profile, not your user profile <nckx>gabber``: The search index was extremely out of date (Dec 2021). rekado: Wish I knew, actually trying to debug it ‘properly’ was unsuccessful; ended just deleting the database & rebuilding it from scratch with ‘goggles index’. I just used ‘sudo -u goggles’ for that, so I'm hesitant to blame namespaces. <pjals>gabber``: i use a setup with greetd + seatd instead of gdm but yeah, it should work fine <nckx>The mtime of the Xapian files was up-to-date, so the job was running. <nckx>Running ‘goggles index’ completed without error but didn't change anything until I deleted the old index. <gabber``>i tried that (adding sway, setting wayland?) but logging in instantly brought me back to the gdm login screen (so sway crashed?) <nckx>rekado: It appears that building the index from scratch works fine, but subsequent (and identical) ‘goggles index’ invocations don't update it once it exists. <nckx>gabber``: Just producing content for the content god. :) <nckx>OK, the files do change, but the search results don't. <Htop170>Hey guix, I'm trying again since it's the afternoon :). I'm trying to package morfologik-fsa, wich is located in morfologik-stemming (https://github.com/morfologik/morfologik-stemming). Since I only need the morfologik-fsa directory, is there some particular method or git-fetch strategy I should use? Or is it better to download the whole repository <Htop170>and navigate through it/potentially delete the other directories ? Someone suggested me earlier I could look at the packages that produces multiple outputs but so far I only saw that the function was used to produce docs and stuff not multiple sub-packages/software (outout '("out" "static" "doc")) <nckx>Htop170: If you're currently interested only in m-fsa, I would do what you suggest but not bother with deleting anything. Simply navigate to the m-fsa directory after 'unpack. If people later want to add separate packages, they can use (source (package-source morfologic-fsa)) or refactor harder, e.g., make a morfologic-stemming ‘primary’ package name as Debian seems to do. This answer assumes that you can build only -fsa without building the whole thi <nckx>ng; otherwise outputs are indeed the way to go (:fsa, :tools, etc.). <nckx>‘guix install xapian’ on bayfront currently downloading mesa, wayland, and libcloudproviders 👍 (but the latter is -minimal so it's OK). <Htop170>nckx: Thanks a lot! I will try to see if it's not too big of a bother to produce all the outputs. Do you know how I should reference them later when I want to include the produced results in an other package ? <nckx>(inputs (list foo `(,fiz "output-name") fuz)) <nckx>Then use search-input-file to address them—probably. Depends on your exact use case. <nckx>Htop170: If you can get away with building one m-fsa, I would try that first. <nckx>The multiple-outputs suggestion was more for if you're forced to build everything at once. <Htop170>So the good practice here would be to write some parent package that could download the git repository and then sub packages that would inherits the parent package and package each software independently rather than doing a package that would build everything at once ? <nckx>Maybe. I can't answer that without digging into how this package builds. In your case I'd start simple: if you want (only) m-fsa, write an m-fsa package. Really, it's fine that its ‘source’ will contain more than strictly needed. Don't worry about that yet. <nckx>In writing that, you'll probably find out what the best ideal fit would be. <cbaines>the disk usage on bayfront has shot up by 15% since midday today, and I'm not sure why <cbaines>nginx also seems to be receiving an unusual amount of traffic <cbaines>I can see corresponding traffic on bishan (which bayfront reverse proxies nar requests to), so it's probable that lots of nars are being downloaded <cbaines>if that's the case, the disk space usage increase is just the NGinx cache being filled up, which shouldn't cause an issue <nckx>Heh. That is very coincidentally when I started playing with Goggles' Xapian logs, but that can't be related. <nckx>But the increase is in /gnu/store, no? <cbaines>nckx, I don't know, my working theory is that it's the NGinx cache that's filling up <nckx>I'm just looking at the mago graph. <nckx>Did you manually free space? <nckx>It's basically back to where it was. <nckx>ACTION did nothing. Nothing! <cbaines>I did just restart NGinx on bishan (a different machine), so maybe that's connected <cbaines>bayfront and bishan do send requests back and forth between each other, so I wonder if it can loop somehow... but that's quite a wacky theory <nckx>It decreased linearly for 2.5h, using >600G, then promptly deleted that *exact* amount again. <cbaines>I think something odd may still be going on as well, since the traffic out from bishan is still unusually high <nckx>nginx buffers big requests to disc, which would look like this (write, write, write, delete), but that's… a lot. <nckx>No intermediate spikes so it would be one big buffer. <abrenon>the same calculation (which seems to make sense to me) yields 4639878 inodes, which indeed accounts for a vast majority of the 4953931 used in total <abrenon>so it is not a huge exageration to say that it is my store which is eating up my inodes <nckx>abrenon: This server is many years old, but I don't think I would have manually tweaked the default ratio. <nckx>So I wonder why I'm fine. <abrenon>I guess it must have something to do with the packages I've installed ? <nckx>The disc is 400G by the way, so the inode usage of 24% is roughly proportional to the storage used. <abrenon>what strikes me is how it peaks once I run a guix package -m <nckx>How much space did it use? <abrenon>(the whole filesystem is 52G if I believe df) <nckx>Right. So that's 1/3rd of my store (123G) but over half the inodes. <nckx>That would be influenced by the type of store item you have. More small files. I can't do much more than point out the obvious. <nckx>You could write a fancy script to ‘blame’ the worst ones. <abrenon>by the way, can I trust your find command applied to /run/current-system/profile to return a good estimate of the number of inode for one generation of my profile ? <abrenon>I wonder why my packages make more crumbs <nckx>abrenon: If you make sure to dereference (but also count!) symlinks. <nckx>ACTION has to go AFK for a few hours. <apteryx>abrenon: have you considered stopping worrying about inodes and switching to btrfs (which dynamically allocates inodes as it needs) ? for such a small file system, the zstd compression would give you a nice 2x capacity as well <abrenon>apteryx: fun fact: I had considered using btrfs for no particular reason a while ago, and then I discovered the horror of fixing an unbalanced system "full" with so much space left and then I thought I wouldn't use it <abrenon>but, granted, it would solve my inode issues <abrenon>then again, I'm simply a liveUSB reboot + tune2fs -T news away from fixing the actual symptoms <abrenon>it's just I thought I'd check with the people-who-know-stuff if that could be a sign I was doing something terribly wrong without noticing, and if there was a good reason why I was using so many inodes <apteryx>abrenon: oh yes, recovering from a full btrfs file system can be a pain <apteryx>I thought the inode count was fixed after creating an ext4 file system (and that changing it would require recreating it anew); is this not the case? <abrenon>hmmm I may be too optimistic ^^ I thought I remember tuning an existing one, but in any case the worse that could happen is to have to format the partition and reinstall from my guix configuration, home is safely elsewhere and doesn't lack inodes, so I wasn't too afraid <apteryx>right, I think inodes is fixed at file system creation time <abrenon>ok so… apparently icons are the ones eating up inodes <abrenon>/run/current-system/profile/share/icons: 1514452 <abrenon>the problem appears to be with the papirus theme, which I indeed installed a couple days ago <abrenon>each version of it (ePapirus, Papirus-Light, Papirus-Dark, etc.) has over 250k inodes ?! <apteryx>here: 16088 /run/current-system/profile/share/icons; I have GNOME and Xfce installed <abrenon>I think people can live with 16k icons <abrenon>in one version, which has an incredible number of resolutions folders, each contains ~20k icons ?! <abrenon>it's all SVGs ?! so I guess it's just a nightmare of symlinks <abrenon>but why so many ? I wonder what they thought of when they made it <abrenon>but anyway, problem solved ! thanks !! <seninha>hi, i'm trying to set up a openvpn service (for the default one at /gnu/services/vpn.scm cannot be configured as i want), but an exception is raised <seninha>ice-9/boot-9.scm:1685:16: In procedure raise-exception: In procedure for-each: Wrong type argument: #<unspecified> <nckx>On phone, so impossible to see the big picture of what you're trying to do, but: you can't use 'case' like this. Hence neither clause matches, so it's returning unspecified (nothing) for documentation and the rest. <nckx>Use match (ice-9 match). <abrenon>nckx: thanks again for your suggestion to look for the inode-eater, it was the papirus-theme apparently full of a huge number of symlinks <abrenon>at some point I recovered ~1G and almost half my total inodes <nckx>sneek: later tell abrenon: A hack (but totally legitimate fix) would be to replace the thousand of symlinks with hard links. <nckx>sneek: later tell abrenon: I don't actually see any disadvantage to that. <nckx>There's an unofficial one at matrix.org. It's just called ‘guix’ AFAIK. <ieugen[m]>hi, how can I have a set of guix pacakges available to all users ? (globally available) <ieugen[m]>interested in having them to ubuntu and root initially but in a broader sense, I would like to have some apps available to all users on a system <vagrantc>available packages (other than with guix system) are generally per-user ... <vagrantc>though ... you could probably make a profile and then add all the relevent variables in /etc/profile ... <nckx>Guix installs packages into a ‘profile’. Each user has a default profile (~/.guix-profile) that, by default (if you used guix-install.sh), is sourced by /etc/profile.d/zzz-guix.sh. But you can install packages to any profile (guix install|remove|… --profile=/home/foo/.some-profile) and source any [extra] profiles you want. So you can, for example, add an /etc/profile.d snippet that unconditionally sources root's profile for *any* user. Or man <nckx>age a ‘global profile’ that's sourced for all users, and use --profile to manage it. Etc. <vagrantc>though if any user used some of the guix features, it might start having unusual results <nckx>Or be like ‘vagrantc’ and use ‘fewer words’ pff. <vagrantc>so much fewer words that it leaves a bit too much as an excercise for the reader ... given i was entirely speculating :) <nckx>I've never used Guix on a different distribution, let alone tried the above, but it should be a supported use case. <nckx>If you do encounter problems, report them. They might be bugs or otherwise fixable moments. <nckx>This is not as common a question as you'd think because most people interested in this use Guix System. <vagrantc>expected problems might be with things like propegated inputs and whatnot, no? or is that isolated within a given profile? <ieugen[m]>thanks. Planning to start using guix to distribute some binary apps to a bunch of servers I manage. Will see how it goes. <ieugen[m]>RIght now I use ansible and would like to do a progresive enhancement until I get more experience with guix. <nckx>vagrantc: This is similar to the system and user profiles in Guix System. Yes, they do not ‘interact’, but that's something you can deal with and shouldn't be a deal breaker. <vagrantc>nckx: yeah, that is basically what i was thinking <florhizome[m]>Now i am at a Point where Just Log into gnome and it freezes immediately <ChillPC>Hello, everyone! I have a tiny problem with guix. I installed neovim (version 0.8.1) with it. However, some treesitter parser does not work. The error printed when opening an html file (for example) is `Failed to load parser: uv_dlopen: /gnu/store/094bbaq6glba86h1d4cj16xhdi6fk2jl-gcc-10.3.0-lib/lib/libstdc++.so.6: version `GLIBCXX_3.4.29'`. <ChillPC>Is there a way to specify which version of gcc should be used to build neovim? Or is the answer elsewhere? <florhizome[m]>When i try to Install a package, i get "manifest Format Not supported" - in German <VesselWave>Hello, could someone help me? I am trying to use TRAMP with ssh. Client and server run guix. I can ssh via cli, but Emacs just stucks when I try to open a file: /ssh:user@192.168.x.x:1.txt <broken_fontconfi>so I'm trying to build fontconfig-minimal but it fails at a test, couldn't find any information online and the test failure is quite obscure