IRC channel logs
2023-11-17.log
back to list of logs
<johnwind1>i'm experiencing a slow download speed when 'guix refresh ci.guix.gnu.org <Kabouik>Any idea what the issue could be? https://0x0.st/HvWk.txt I don't see any clear error indicating a cause. The same package installed fine on another machine with Guix system (this one is on a foreign distro). <podiki>Kabouik: it is trying the same derivation hash? generally that shouldn't happen but indeed not every build is reproducible so you could try again (out of memory can cause odd failures or sometimes when multiple threads going) <Kabouik>It does not use a hash podiki (it builds based on git-checkout master) and this is the same commit in both cases. OOM could be the issue but that surrprises me because I changed the compiler to clang to reduce RAM usage to <2GB while the machine has 6. I could try with fewer threads maybe. <podiki>i meant the hash of the store output it wants to make (or of the derivation itself) <podiki>the /gnu/store/<hash>-<package name>-<version> (or with .drv) <Kabouik>Oh, then I need to check tomorrow, the other machine is not here. <dthompson>does anyone occasionally run into a situation where the users/groups that a service creates need to be tweaked? I like that it's possible to tweak them by making a new service type, but if that service type is extended by other services things get messy. <zjabbar>Hello Guix. I would like to attempt GPU passthrough (like others: Issue 55907). I was able to set my kernel arguments in a way that I think sets iommu and vfio-pci in a way that PCI passthrough is supposed to work. However I get an error when looking through `dmesg` for `vfio`: [ 1.118676] VFIO - User Level meta-driver version: 0.3 [ 1.843906] shepherd[1]: could not create '/dev/vfio/vfio': File exists <zjabbar>I do not see a mention of shepherd throwing an error in that issue. I am wondering if the root cause is the same. Forgive the rehashing of the same error if it is the same. <PotentialUser-30>As a side note, I see that someone spammed commits for what appears to be all of texlive. Is that sort of thing acceptable for package ecosystems? <mirai>usually it's one commit per package <mirai>given the size of texlive it's not surprising <podiki>PotentialUser-30: all of texlive should be in guix, there is (was?) the monolithic package with everything but i think recently everythign was individually packaged that existed at the time <PotentialUser-30>I had just noticed that there were hundreds (maybe thousands) of commits for individual texlive packages all on the same day <Andronikos>PotentialUser-30: Probably the merged texlive branch. <Andronikos>Not sure but they changed something which required changing all the packages. <PotentialUser-30>is there any way to know what is currently being packaged? I have some things I want to package but don't want to waste the time if there's already a branch somewhere that's going to get merged <PotentialUser-30>It's a little discouraging to look for packages to add and then just see a wall of commits adding all of them <PotentialUser-30>nevermind, apparently you need to use texlive-scheme-basic to get the binaries <Andronikos>If you install just texlive, you can the whole texlive distribution. <Kabouik>sneek: later tell podiki: Re: building error on one machine and not the other (https://0x0.st/HvWk.txt), I checked what you said and no, it doesn't seem to be the same /gnu/store/hash-package-version on both machines. One is x86_64 and the other (where it fails) is aarch64 too, but I know that the sources compile fine on this machine when not using the Guix package definition. I am nto sure what is wrong, I don't understand the compilation error I <Kabouik>Oh, I guess only more members can use sneek. <futurile>efraim: ok maybe you consider it frozen for now. Figured I would ask. <zamfofex>Hello. I noticed cJSON vendors its test library, which is currently not available in Guix. Is this deemed out of scope for Guix, or would a patch to package the test library and use it instead of the vendored version be appreciated? <cbaines>zamfofex, is there any software apart from cJSON that uses that test library? <zamfofex>I don’t know. It seems to have many stars on GitHub, so it seems somewhat popular. I don’t know any other packages in Guix that use it. <zamfofex>Note that I noticed it because the MinGW build fails for cJSON, which uses ‘-ansi’ to compile it, but it uses ‘long long’ on MinGW. The upstream repository uses ‘-std=c11’ instead. (It’s also seems fairly outdated on cJSON.) <gabber`>i have an OS definition that doesn't mount the https://termbin.com/ouni parition with label GNU-ESP -- and i'm not sure why? it should automatically be mounted, right? what should i look out for in the system log? <ekaitz>efraim: zig packaging is givin me extra headaches... it's not finding the shared objects! <ekaitz>same package, only changes the version <dthompson>has anyone seen process-compose from nix land? it's like docker-compose but without docker. <dthompson>I've wanted something like this for years but haven't built it. I think we could integrate it directly into guix by using shepherd to manage all the services. <dthompson>I see it as complementary to 'guix shell'. your dev environment needs not only some tools to build the project, but often it requires some daemons. lately I've been working on guile-hoot and I often keep a little guile web server running to serve the wasm files I build. <dthompson>it would be great to have a 'guix ____' command that spawned a shepherd instance to support the additional daemons needed for development. <dthompson>another use-case: when I work on my blog I'd like 'haunt serve --watch' running in the background at all times. <dthompson>web apps usually need a local db server, etc. <gabber`>on my development machine (a Guix System) this works without issues <attila_lendvai>civodul, random: (spawn-shell-command) is not compatible with (system): "If system is called without arguments, return a boolean indicating whether the command processor is available." <attila_lendvai>shepherd replaces system with s-s-c, and it caused me some headaches until i realized <gabber`>it is a "/gnu/store/<hash>-guile-3.0.9/bin/guile': No such file or directory" error - but the file exists? <attila_lendvai>gabber`, could it be for a different architecture than yours? i.e. some cross-compiling is involved? <gabber`>as stated above it is -- i do a `--system=aarch64-linux` on a debian machine <gabber`>so not cross compiling but qemu-binfmt native compilation <redacted>I'm using guix deploy. My nginx config uses (local-file) to add an htpasswd file. Changing that htpasswd file locally and re-running guix deploy appears to succeed, but the remote htpasswd file (and the resulting nginx config) are left unchanged. <redacted>Oh wait... I'm using a gexp, maybe that's the issue <dthompson>redacted: have you restarted the nginx service? you need to do that <redacted>dthompson: I thought so, but I'll try that again <Guest33>guix shell --container --preserve="^DISPLAY$" --expose=$HOME/.Xauthority xeyes -- xeyes <Guest33>this fails with Error: Can't open display: :0.0 <Guest33>however, adding --network to it makes it work <Guest33>is there a way to allow X connections without the network? <civodul>Guest33: i think you need --expose=/tmp/.X11-unix <redacted>I am deeply suspicious of this thing that I have done: #~(format #f "auth_basic_user_file ~s;" #$(local-file "htpasswd")) <jlicht>dthompson: having such a `docker compose' analogue /w guix would be pretty cool <civodul>actually one could abuse ‘guix home container’ for that purpose, but it’s not made with that in mind <redacted>... I think my problem was that guix deploy <file> -x doesn't actually do the deployment. I misunderstood the docs <civodul>redacted: ah yes, ‘-x’ just lets you execute a command on all the nodes <redacted>when I do stuff with guix, I get a "man, computers are fun" feeling <isaneran>would be sick to be able to use guix shell and get a shell that not just has the package dependencies but also starts services needed for the development (like databases web server etc) in proper isolation <dthompson>civodul: yeah I thought about 'guix home' but it's not the right fit. <dthompson>there are some UI things to figure out... does it complement 'guix shell' or is it a separate tool? <dthompson>seems easiest to make a separate tool... but what's a good name? <dthompson>'guix compose' isn't a good name since everything in guix composes, more or less ;) <dthompson>it could be a tool that just handles these dev shepherd services... or it could be a tool that wraps 'guix shell' as well <dthompson>as far as prototyping goes, the easiest thing is to make something completely separate from 'guix shell' and think about the UI later. <civodul>dthompson: i guess like for Docker Compose it should let you specify not just services but also bind mounts, open ports, and all that <dthompson>civodul: yeah, I see that as step 2. first, get unisolated processes running. then, add optional containerization. <dthompson>many simple cases require no isolation at all. I certainly don't want it when running 'haunt serve' or my hoot dev web server. <dthompson>let's go classic lisp and give it an unreasonably long name like 'guix local-development-services' ;) <isaneran>ACTION should use haunt instead of forcing myself to do a DIY site generator that I don't really wanna make <dthompson>I had the name 'guix develop' in mind for a tool that combined 'guix shell' with service management, but it's not a good fit for a more standalone tool. <isaneran>it seems to be working well for both guix and spritely so <isaneran>I'm guessing there is no good org-mode generator thing for it <dthompson>I think it exports org-mode files at html with a haunt metadata prelude. <dthompson>I don't love it since the haunt post is no longer the source but it seems to work for at least someone out there <dthompson>guile-present has a parser for a subset of org-mode I think <isaneran>actually the thing I started forcing myself to write a little was some very minimal html gen from org <dthompson>I don't want to do it, but if someone wanted to use that as a basis for a dedicated guile-org library or something that would be great :) <isaneran>Maybe I don't have to be so purist and should just use markdown :P <isaneran>it would be great to have though, but I don't feel _that_ motivated to do it <dthompson>guile-commonmark really needs to be updated to support embedded html tho <lilyp>(how) can we query i.g.o by QA tag? <lilyp>not as user-friendly but did the job, thx <vagrantc>just realized guix hash can be useful to compare two directories for bit-for-bit identicalness <h4>\/etc/config.scm:11:0: error : #{\x7d;}# : invalid field specifier <attila_lendvai>shepherd: i do a guix shell, then ./configure leads to "configure: error: pkg-config is missing, please install it", but pkg-config --help runs happily from that shell. any hints? <podiki>sneek: later ask efraim are there some build jobs that need to be (re)started for mesa-updates? Last I saw on QA coverage looked stalled (but I can't load the page today so...) <ekaitz>hi i'm playing with tuned packages and I need to know how can I make a build-system choose the default -march <ekaitz>i think i managed to add the correct one when tuning, but when it doesn't tune I need to add a baseline cpu <lilyp>do you really need to? gcc ought to produce non-cpu-specific code if invoked without -march <lilyp>just delete any bs code that adds it and you're good to go <ekaitz>lilyp: well, the thing is i'm not with gcc but adding support for other compiler <ekaitz>by default this compiler (zig) does "native" but that's not what we want, we need "baseline" <ekaitz>efraim suggested to use the tuning system to adjust that, but if we don't have any way to remove the "baseline" flag from the compiler flags, I don't know what we can do <apteryx>ACTION is not impressed about startpage: slow to load, and gives me a page full of sponsored link before the actual results for each search <apteryx>does someone know how to configure GNU IceCat to use Google as the search engine? <Kolev>Ironic, heh. But I'm sure there's a way. <Kolev>Hm. Google isn't in Settings > Search. <apteryx>I made it to some page that offered search add-ons once, but the page had no download link or anything, so I was just confused <Kolev>I dread running `guix pull`. <cbaines>jackhill, yeah, there's an issue around the derivations, since the guix-daemon capabilities detection doesn't work in the situation used by the build coordinator <ieure>apteryx, Hmm, I guess Google dropped support for OpenSearch. Used to be that you could right-click the search bar and add it as a search engine. <ieure>Noting also that I have ethical objections to using Google products. And their search just doesn't work very well these days, gives horrible results polluted with "AI" lies. Recommend that you find a better SE, frankly. <ieure>I run a SearxNG instance and that's been working pretty good for me. Public instance list at https://searx.space if you want to try it out. <cbaines>jackhill, there's an issue now #67250 <jackhill>cbaines: I see, thanks!! Thanks for writing it up. From the outside it was pretty confusing. <jackhill>so as it is not it will eventually get lucky? <cbaines>no, that derivation needs building on a machine with a newer guix-daemon <cbaines>I think there's a spread of machines with newer daemons now though, so maybe you'll get lucky <jackhill>🤞 I'm happy to waid as long as it isn't anything I messed up. Until I clicked through the see the build log, I worried I'd gotten the hash wrong or something like that. <lilyp>ekaitz: that's… unfortunate to say the least. You'll have to dig deep into zig to make the baseline (i386 or x86_64 sans extensions) the default <ekaitz>lilyp: it shouldn't be thaaaat hard <lilyp>okay, just dig and not deep then :) <ekaitz>lilyp: it should be some kind of easy patch in argument parsing <lechner>Hi, can we use srfi-69 hash tables in Guix proper or is it just alists? <lilyp>lechner: probably maybe; don't try to shove them over compute boundaries tho <lechner>lilyp / a gexp is a compute boundary? <lechner>i don't need the hash tables but it's currently a plain list and an integer index, which is error prone <lilyp>what do you need integer indices for tho? <lechner>lilyp / i don't. the list of boot parameters is currently a plain list. i think it should be a mapping from generation number to boot parameters