IRC channel logs


back to list of logs

<RavenJoad>You literally have coq right there...
<Guest69>nckx: Okay I fully upgraded my system.  Both commands still return "$1 = #f" for me.  I am on guix (GNU Guix) bd7bb4794400ee0cbffda52846f6464a5390ea5b
<Guest69>My system also needed to build linux-6.4.14-guix.drv.  Took 2 or 3 hours.  Could we add some flag like --only-substitutes-check or something that it doesn't even start if it needs to build since it is not yet available?
<Guest69> there is 59G left on this device.  How can I fix that?
<civodul>Guest69: most likely, either /gnu/store or /tmp is full
<Guest69>civodul: /dev/sda2       932G  873G   59G  94% / Does that mean there is some protection because my disk is 95% full?  I mean 59G sounds more than enough for me
<lechner->Guest69 / i just saw your lease message, but 'guix gc' may be helpful
<lechner->Hi, where does Guix define the quotation operators #~ and #$ for G-expressions, please?
<mwette>lechner-: I'm seeing it in guix/gexp.scm near line 2297, `read-hash-extend'
<ulfvonbelow>pop quiz: what is the difference between (package (inherit p) (arguments (package-arguments p))) and (package (inherit p))?
<Guest69>lechner-: Was the first thing I tried.  I cleaned up and the command worked.  Seems like 59G was actually not enough
<ulfvonbelow>answer: because the arguments field is thunked, the former will evaluate that thunk with 'this-package' bound to the original package, while the latter will evaluate it with 'this-package' bound to the new package
<ulfvonbelow>this raises an interesting question: if you wanted to define a package with an arguments field which, when evaluated, evaluates another package's arguments thunk with 'this-package' bound to its own package, *and then does additional transformations on it*?
<ulfvonbelow>how would one do that?
<nckx>Guest69: I'm back and at a computer now. I haven't looked at how the ‘did you forget?’ suggestion magic is implemented (I assume civodul does :) but it seems that importing (gnu) is what makes it work for me.
<nckx>Guest69: 5% is a very suspicious number.
<nckx>Which file system is this?
<nckx>OK… not what I was expecting. Many file systems reserve 5% (or in that ballpark) of space for root and file-system-level housekeeping. But… I don't think btrfs does that!
<nckx>However, it's rather sensitive to running at practically-full levels. You can easily run out of usable space even if the numbers say otherwise.
<nckx>95% is very full indeed.
<lechner->Guest69 / are you sure the store is the issue? how about du -sh /*
<lechner->i agree with nckx
<lechner->mwette / thanks!
<Guest69>nckx: That 5% was my theory as well but was not sure if Guix itself had some protection as well.
<nckx>(And if you're running Guix with /tmp on tmpfs, I also suggest the ‘don't’ alternative, unless you have oodles of RAM+swap)
<Guest69>Yes, I do not have much space left.  But I am working on it.  Disks are expensive
<nckx>Guest69: No, only a warning. It will not kill builds for that reason.
<nckx>Guest69: Understood.
<nckx>I'm glad to hear someone resist the stupid ‘storage/RAM is cheap’ meme.
<nckx>If perhaps not entirely voluntary.
<Guest69>Samsung announced a new 256TB SSD, maybe this fixes my storage needs (as I would ever be able to buy one of these)
<Guest69>nckx: Well, in the USA I would say yeah sure.  But every country outside of the USA, definitely no.
<Guest69>USA 5$ per TB, DE 15€ per TB if you are lucky
<ulfvonbelow>Storage getting so cheap actually makes me a little uneasy
<nckx>My X230T's second SSD is a shucked 2TB Samsung T5 I found on sale for stock clearing prices. Maybe something like that is an option.
<lechner->my issue is $0.42 per kWh it takes to power them on
<lechner->sorry, I guess that's not SSD
<Guest69>ulfvonbelow: You mean as of reliability?
<ulfvonbelow>sort of that, yes, but also just thinking about how usage expands to meet availability, and it's doubtless going to mean ever-increasing surveillance
<Guest69>Somewhere those huge AI models need to be stored
<lechner->Guest69 / how about this?
<nckx>You might want to use the ‘usage=’ filter here.
<nckx>‘might’ in its common nckx meaning of ‘mos def’.
<Guest69>Ah interesting thanks.  In my case I use a SSD.  I guess it will not take that long/wear down my drive?
<nckx>It's pointless wear at best.
<nckx>Unless you're taking this opportunity to add/increase btrfs compression.
<lechner->okay, i'm not good with btrfs
<lechner->running out of space is a common complaint, however
<nckx>To be fair to btrfs, there are many common complaints about btrfs.
<Guest69>Probably because it doesn't do anything by itself.  That checksum calculation for bit rotting (scrub is it called I think) isn't done automatically.  In other words, it isn't working magically out of the box and requires to read the manual
<nckx>unmatched-paren: If that example fails, it probably does matter, because that's weird. I don't use Geiser.
<Guest69>though ZFS is the same if I recall it correctly (checksum part)
<nckx>Well, checksums are validated on every read, but indeed: the damaged blocks aren't reallocated automatically.
<nckx>You'll just get csum errors in dmesg ad infinitum [until the second block fails and you finally lose the data].
<Guest69>Oh yea forgot to actually write the keyword: bit rotting
<nckx>No, but I missed it.
<Guest69>yea you are right
<nckx>Actually, no, I don't see how that changes the meaning.
<Guest69>well, the fs doesn't protect you from bit rotting if you don't run the scrub command
<nckx>I'm not aware of any file system that reads your entire array on its own accord, without a scheduled user-space job.
<Guest69>and I think in ZFS it is the same, thinks like TrueNAS do that for you.  They add a cron job to schedule it
<nckx>ZFS doesn't do this *by design*.
<Guest69>What you mean with that? I don't understand it.
<nckx>About whether ZFS autoreallocs on reads: AFAIR it's complicated, e.g., <>, but my knowledge might also be stale.
<Guest69>To be fair, I don't have much knowledge either.  There was just some YouTuber that didn't use TrueNAS or something else and did setup ZFS manually.  After some years they actually lost some data to bit rotting.  They said that they forgot to setup the job but things like TrueNAS do it automatically.  Which sounds to me like this
<Guest69>  does not automatically run it.
<nckx>The ‘by design’ part? ZFS was designed to minimise stress on (potentially failing) drives without explicitly being told what to do by a competent admin. So that would include things like scrubbing the entire array in the background or whatever.
<nckx>If you want to schedule a nightly scrub, you can, but it's not more automagical than btrfs.
<lechner->Hi, I wrote a Scheme-y make engine, if anyone is interested
<Guest69>Ah now I understand it.  BTRFS doesn't do that automatically because they can't know you setup and therefore can't make the best decision => just do it yourself  (read it somewhere, don't know if it was in the manual or some forum)
<nckx>(I don't know why TrueNAS doesn't ship a cron job by default, since they are the de facto admin of many of these consumer boxes.)
<Guest69>they do
<Guest69>that was my point
<nckx>It's ironic.
<nckx>I made a read error.
<Guest69>SW vs HW RAID?  Does someone actually work in enterprise and use HW RAIDs?  Community for NAS system say that HW is basically dead, but is it?
<ulfvonbelow>it's alive enough that you have to specifically look for SAS controllers that aren't in hardware-raid mode on ebay, aside from that I dunno
<nckx>With the collective hours wasted fighting the ci.guix RAID controllers I shudder to think what would happen if we'd actually *used* them 💀
<Guest69>Not sure but after some research HW RAID was basically offloading CPU and is only on block level.  In other words, not needed in a world with 60 cores on nearly 6GHz speed and file awareness as checksums, compression features and so on
<Guest69>but never seen a company that use ZFS, so there must be a reason
<lechner->EMC and NetApp probably use HW raid
<nckx>‘never seen a company that use ZFS’ wat
<Guest69>Yea I mean think about it.  Every server has a RAID controller
<Guest69>though I was not every country in the world* just some little ones
<CompanionCube>ACTION scrolled down and also saw, which seems relevant?
<Guest69>I wonder if those big cloud companies use SW or HW RAID, since they have the highest bandwidth probably except maybe YT or some adult sites
<Guest69>bandwith as of data storage io
<nckx>I'm not doubting your experience, just surprised. I'm sure there are whole sectors that do use the RAID controllers, just like they use the default Windows that came with their laptops.
<Guest69>Sounds like you would favor SW?
<lechner->just don't use btrfs to do it
<nckx>Companies where ‘storing data’ is key to remaining profitable/competitive? You bet. I promise you that YT doesn't use hardware RAID 😉
<nckx>…yes they are also not using btrfs.
<Guest69>Basically, I just want to validate the information I found online.  It is the same with ZFS and BTRFS.  They all prefer ZFS because some emotional relationship or something.  But I want neutral information to decide which is the best tool for my case.
<Guest69>(which is BTRFS btw, since I can use it with GNU Guix :))
<nckx>How sure are you about that ‘because’? I think you might be rather hasty there.
<Guest69>nckx: Yes, it is called GFS (
<Guest69>nckx: About what specifically?
<apteryx>what does "guix time-machine --branch=v0.15.0 -- describe" produces for you?
<nckx>‘They all prefer ZFS because some emotional relationship or something.’ Those people definitely exist, but you seem to have encountered one too many and assume that this must be the reason that ZFS is considered best in class. I think that's because it is. And I don't have an emotional connection to ZFS and would rather not use it.
<nckx>apteryx: Are you asking anyone?
<apteryx>no :-)
<nckx>‘For me’ it produces ‘guix time-machine: error: Git error: cannot locate remote-tracking branch 'origin/v0.15.0'’ but that's probably not useful data :)
<apteryx>isn't that bogus?
<apteryx>v0.15.0 is an existing tag
<nckx>I pull from ~/guix, where ‘origin’ isn't Savannah, so TBH I hadn't given it much thought.
<apteryx>ah, tags need to be provided to --commit
<apteryx>guix time-machine --commit=v0.15.0 -- describe -> guix time-machine: erreur : cannot travel past commit `6298c3ffd965' from May 1st, 2019
<Guest69>nckx: Ah I see. for example.  He doesn't like BTRFS because RAID 5/6 is not working, though in the manual it says it is not ready for production.  Most say use ZFS because BTRFS does't even support those RAIDs but probably using RAID 10 anyways.  For my use cases it is the better choice, though if I had a NAS
<Guest69>I would install TrueNAS since I know that it works correctly and I don't need to manage it.
<Guest69>nckx: Can you give me some arguments my it is best in class so I can learn?
<nckx>apteryx: That --branch=TAG is bogus is ‘obvious’ and yet I didn't spot it. I wonder how cheap it is to check whether it's a ref and print a hint. I'm guessing it's cheap.
<Guest69>Just to be clear: I don't hate ZFS nor do I hate BTRFS.  I like both and BTRFS just fits my case better.  My experience was that if you look online it sounds more like they echo ZFS as in actually giving arguments why it is better.  So I use both to make my own opinion.
<bumble_>unmatched-paren: if you are here, I've tried greetd-terminal-configuration again today and still have blinking cursor, but this time I switched tty and looked at /var/log/greetd-1.log and noticed an error message that wasn't there previously....
<bumble_>the first error to appear in the log during the timeframe of the last session is...
<bumble_>`2023-09-05 17:15:35 error: terminal: unable to get current vt: EIO: I/O error`
<bumble_>if you have any sort of advice I would be interested to try that...
<Guest69>rebooting system
<nckx>Guest69: For personal use, and especially on Linux, I actually recommend btrfs, so I'm going to pass on the ‘Why ZFS?’ presentation 😉 Btrfs works fine, it has some nice features that ZFS lacks that are interesting to home users, it's more resource efficient, and is better supported by distributions OOTB (‘roll backs’ etc.).
<Guest69>nckx: Wouldn't mind your ZFS presentation.
<nckx>My ‘best in class’ was still in the context of large enterprises (or ones where Storing Data is a large part of the bottom line). Take off your free software hacker glasses for a moment, then try defend shipping a known design flaw (raid56) in the *same* code base that users are expected to use to protect their data. ‘It was marked experimental and it *probably* didn't affect the rest of the code lol’ is just… insane, if you're not in
<nckx> front of a friendly or like-minded crowd.
<nckx>Alas, there will be no presentation, only sleep.
<nckx>ACTION 'night!
<Guest69>nckx: Good Night!
<ulfvonbelow>well now this is quite interesting... when trying to build llvm-for-mesa with -ggdb, I get: ld: error: ../../lib/libLLVMSelectionDAG.a(LegalizeIntegerTypes.cpp.o)(.text._ZN4llvm16DAGTypeLegalizer19ExpandIntRes_VSCALEEPNS_6SDNodeERNS_7SDValueES4_) is too large (0x42a bytes)
<ulfvonbelow>followed by ld: failed to set dynamic section sizes: memory exhausted
<ulfvonbelow>I suspect that particular error message is bogus, since this is on a machine with 64GB RAM and 72GB swap, sooo...
<ulfvonbelow>also, 0x42a == 1066 doesn't seem like a particularly crazy number
<ulfvonbelow>ahhhh, I see now, it's because it's a i686-linux build
<ulfvonbelow>address space is presumably rather more limited there
<bumble>I opened an issue/bug for the first time just now, sending an email to ``. Will my message appear at ``?
<mwette>maybe here:
<lnnk>Is guix similar to arch or gentoo?
<apteryx>yes and no
<apteryx>you can build from source all you like like on gentoo, but there's also binary substitutes available
<apteryx>there's no direct equivalent to USE flags, but you can rewrite the package graph, at a heavy computational cost (you're on your own -- no substitutes)
<apteryx>the biggest distinction in my opinion is that the package manager is transactional; it's easy to go back between upgrades
<apteryx>and it's not a file-hierachy standard (FHS) distro, so no /usr/lib and the likes
<apteryx>it also probably takes reproducibility and bootstrapping more seriously, although that's not verified
<bumble>guix uses a lot of disk space, compared to arch
<mwette>rlb: I assume it's the utf8 branch
<ulfvonbelow>re: "at a heavy computational cost", I've been tinkering with deep rewriting of large manifests, and found that it actually something like triples the amount of time guix spends before starting to build anything (1.5 minutes without any transformations, 4.5 minutes with ungrafting, -ggdb, and removing debug outputs)
<Guest53>Hi everyone;
<Guest53>I've been trying to install GuixSD, and everything seems to go ok (no errors at all, looks like everything gets put into place properly on the desired partitions) but it doesn't seem to create an EFI boot entry at all. My system has 2 SSDs (the other has another OS on it with its own ESP partition), and for other distros I've tried I've been able
<Guest53>to have ANOTHER ESP partition on the SSD I'm installing the distro on and will be able to navigate to the entry through BIOS rather than the Grub menu (which is fine)
<Guest53>Is this not possible, not recommended, or am I just doing something completely wrong?
<Guest96>I did sudo -i /path/to/current/bin/guix pull --commit=<commit corresponding to /path/to/current/bin/guix>. I did this to initialize the root profile so services can point to it. Strangely it still tries to git pull on upstceam. Shouldn't it notice that the commit already exists in the store and not dowload?
<Guest96>Guest53: did you specify an ESP parwitian in your system config?
<Guest53>Indeed, I did have to change the UUID manually to the new partition though as I noticed it used the UUID of the partition from the other drive
<Guest96>Could you try booting the machine with only the guix drive?
<Guest96>or have you?
<Guest53>Not yet. My plan if nothing else worked was going to be installing it with only 1 drive installed, so thinking that might be best
<Guest96>what does your bootloader entry look like?
<Guest53>1 moment
<Guest53>(bootloader (bootloader-configuration(
<Guest53>                      (bootloader grub-efi-bootloader)
<Guest53>                      (targets (list "/boot/efi"))
<Guest53>                      (keyboard-layout keyboard-layout)))
<Guest53>Then the file-system is simply (file-system (mount-point "/boot/efi") (device (uuid "..." 'fat32))
<Guest96>da you see tme efi pattition with populated folders, including the one from guix?
<Guest96>If so, you could manually load it by booting to your bios or whatever. if there's a "boot options", go there
<Guest96>by "load" I mean manually add a uefi entry and navigate to the one for guix
<Guest96>there should only be one selectable file in the guix folder on that uefi part iirc
<Guest53>finishing install up & will check
<Guest53>Seems to be a MSI issue
<Guest53># mv esp/EFI/grub esp/EFI/BOOT
<Guest53># mv esp/EFI/BOOT/grubx64.efi esp/EFI/BOOT/BOOTX64.EFI
<Guest53>Had to do the above
<ardraidi>I'm just so happy and want to say thank you to all the Guix maintainers and contributors. You guys are awesome!
<Guest96>ardraidi: <3
<Guest96>gnight guix
<bumble>unmatched-paren: please see where the cause of the wlgreet blinking cursor issue is identified
<bumble>I don't want to bother you and am messaging you from a different timezone where I might be offline when you come online, so hoping you get this message
<nikolar>Re zfs and btrfs, btrfs isn't even production ready in raid10
<nikolar>If you have a faulty connect or a drive that sporadically fails, it won't detected that they diverged and will silently corrupt both
<nikolar>If you're using btrfs on a single drive, it's probably fine
<nikolar>But any kind of raid is broken
<jpoiret>isn't it only raid5/6? there's an entry on the btrfs wiki regarding that
<nikolar>Raid 5/6 is completely broken
<nikolar>Raid 1 is subtly broken
<nikolar>Which is even worse probably
<rekado>In the past few days I’ve tried to switch to a more recent kernel on my RockPro64. Currently, I use 6.1.46.
<rekado>I tried 6.4.12 in various configurations but none of them would boot.
<rekado>has anyone tried a more recent kernel with the RockPro64?
<rekado>I want a more recent one because there’s no HDMI video output once Linux takes over.
<rekado>(I have HDMI in uboot and Grub)
<efraim>who can I ping about the lisp-team branch and sbcl? sbcl@2.3.8 on riscv64 fails on the contrib section again
<efraim>I'm using 6.4.12 on my pinebookpro
<rekado>efraim: hmm, that’s also rk3399, so I would expect it to work for rockpro64 as well. Which package variable do you use?
<rekado>I use it with a custom kernel config to allow booting from SSD and enable MIDI.
<efraim> linux-libre-arm64-generic
<efraim>rekado: it looks like towboot supports the rockpro64 if you want to try that
<ngraves>Hi guix! I'm having some trouble with primitive-eval a file content in a file. I have a very small example here : What happens is that the script works if it's not compiled, but fails if it's compiled. It boils down to guix records, the error being: In procedure make-struct/no-tail: Wrong type argument in position 1:
<ngraves>#<syntax-transformer <channel>>
<ngraves>If someone has already managed to make this kind of script, I would be happy to have some ideas about how I could do this. Thanks!
<efraim>looks like I do need to recompile full texlive on riscv64. I really need to put an SSD or something in my big machine so I can use that as /tmp
<ryan77627>Hey all, anyone have any experience in getting virt-manager to see the OVMF package and utilize EFI in VMs?
<nckx>arby: guix copy: error: SSH authentication failed for '': ssh_userauth_agent_publickey: Out of memory
<nckx>is a new one.
<tricon>ACTION :S
<nckx>Both server and client are very much not out of that sweet sweet memory.
<apteryx>civodul: hi! do you see a way making the list of guile modules loaded when using %load-verbosely hierarchical? (like the 'tree' commands when showing directory contents)
<alethkit>Are there any tricks for building a Go package as a module? using the GOPATH mode seems to cause a duplication error for a function symbol
<mirai>can anyone try this command out and see if the 'check phase passes? (guix build xfig --no-substitutes --no-offload)
<mirai>I'm scratching my head at this… it passes if offloaded (to a machine with an AMD cpu) but not locally (with intel cpu)
<nckx>free(): double free detected in tcache 2
<nckx>Intel (i7-3615QE).
<mirai>is this an intel machine?
<apteryx>yes (i7)
<apteryx>works for me
<andreas-e>mirai: It works for me.
<apteryx>Intel Q6700 Core 2 Duo
<mirai>Just to be sure, when I do an offloaded ./pre-inst-env guix build, does it matter how recent guix is on the building machine?
<nckx>To be clear, my ‘double free’ was fatal.
<apteryx>mirai: no
<andreas-e>Intel(R) Core(TM) i5-6300U CPU @ 2.40GHz
<apteryx>it just builds the derivation, which should behave exactly (unless of breaking changes in the daemon on the remote, which are unlikely)
<mirai>other than some grafts done to ghostscript in the meantime (didn't do a reconfigure on remote yet) I don't see why this should occur
<nckx>It just succeeded on Berlin (AMD EPYC 7451) but this whole manufacturer thing might be a red herring.
<apteryx>mirai: could be just non-deterministic
<mirai>this contraption is yet to give me a successful check './pre-inst-env guix shell parallel -- parallel -j1 --joblog /tmp/build-parallel.log -N0 './pre-inst-env guix build xfig --no-substitutes --no-offload' ::: {1..100}'
<mirai>in 100 local builds, I've yet to see a '0' exit code
<mirai>hmmm… strange since upstream can't reproduce this observed double-free
<nckx>Is that what kills it for you?
<mirai>I think the test is dying there, no?
<apteryx>ACTION tries
<nckx>mirai: It is for me.
<nckx>I'm asking if it is for you.
<mirai>though the main question is “why does the testsuite fail/succeed erratically”
<mirai>all I'm certain is that it fails at test 3/3, and I see something something double free in the build logs
<nckx>It's not erratic, so far machines are reproducibly cursed or blessed ☺
<nckx>Until the berlin failure it was obviously based on CPU model number being even or uneven, but that turned out to be false.
<nckx>-{M,c}1 doesn't seem to help either.
<mirai>well, there's another datapoint (which _shouldn't_ matter): the remote isn't running X11
<mirai>I presume berlin isn't as well
<apteryx>mirai: that can't matter
<apteryx>(unless containerization is broken in Linux)
<apteryx>does it happen if you build with --cores=1 ?
<apteryx>or --cores=4 ?
<apteryx>(my CPU has 4 cores)
<apteryx>mirai: you no about --rounds=, right?
<apteryx>instead of using the external parallel
<mirai>yea but it doesn't do what I wanted here, even with --keep
<mirai>the --keep going option I mean
<apteryx>ah, right
<apteryx>that's #51392
<mirai>it stops if the package in question fails (whereas I wanted to see if this was some nondeterminism)
<Rovanion>On the subject of containers: Does anyone know of a tool that allows the creation of a mount namespace without also mapping the current user to root? `unshare --mount` for example cannot be run without `--map-root-user`.
<elevenkb>Hello there, due to a change in the interface of consult, the package emacs-consult-eglot no longer works at all.
<elevenkb>Currently the package is pinned to the tag v2.0 (which is the only stable tag), should I upgrade it to master?
<apteryx>unstable is broken, I'd say yes. especially that's what most people must use when installing from Melpa
<apteryx>unstable >> broken
<nckx>The ‘stable releases only’ rule has many exceptions.
<elevenkb>Thanks will do!
<ngz>andreas-e: I could use some help wrt a TeX Live package, if you're around and have some minutes to spare.
<ngz>(or anyone else, of course; it's a problem `ld' returning errors)
<tylerm>Does anyone know how to get the OVMF firmware in libvirt for virt-manager?
<andreas-e>ngz: Sure, which one is it?
<ngz>andreas-e: Oh! I need to go for a few minutes. This package: is failing to build, I'd like to understand the problem.
<ngz>(the missing part is at <>)
<apteryx>mirai: xfig never fails here
<elevenkb>What should the revision parameter of (git-version ...) refer to?
<cbaines>my computer keeps suspending, and I have no idea why :/
<andreas-e>cbaines: The heat? Coupled with dust in the case?
<cbaines>andreas-e, it's a fanless case with a fan pointed at it, so I don't think it's a hardware think, I think it's a software thing
<cbaines>but I have no idea what software is indicating for it to suspend, or how to stop it :(
<cbaines>unfortunately most of the documentation on the internet talks about dealing with this via systemd, so I'm not sure what's applicable to Guix
<ngz>andreas-e: sorry about that.
<ngz>I'm back
<ngz>andreas-e: So, apparently, the build process is not able to find texlive-libkpathsea, even though it is propagated as a native input, along with pkg-config.
<andreas-e>No problem, I have just started to build the package.
<ngz>andreas-e: I checked the Makefile, and configure scripts, and they apparently handle the case of an external libkpathsea.
<ngz>hmm. I may be missing "--with-system-kpathsea"..."
<andreas-e>It does not find kpse_* functions.
<andreas-e>And when you look at the linker line, it has -lm and nothing else.
<ngz>It fails with "--with-system-kpathsea"
<ngz>(the same way)
<andreas-e>tex.scm now takes 1.5 minutes to compile...
<ngz>100k locs
<andreas-e>The does not mention --with-system-kpathsea.
<ngz>"configure" does
<andreas-e>It comes from the KPSE_ macros in, which come from code in the m4 subdirectory. This will not be fun...
<ngz>I wonder why $(KPATHSEA_INCLUDES) is not set correctly (or so it seems)
<andreas-e>There are texlive-kpathsea and texlive-libkpathsea. A mistake? Do we need one and not the other?
<ngz>We need both at the moment.
<andreas-e>I mean, as package input.
<ngz>Ah, not, texlive-kpathsea is not needed.
<ngz>err no*
<andreas-e>Maybe we should add "-lkpathsea" to some linker flags?
<andreas-e>configure does not seem to do it.
<ngz>It should be in KPATHSEA_INCLUDES, shouldn't it?
<ngz>i.e., `$PKG_CONFIG kpathsea --cflags` and `$PKG_CONFIG kpathsea --libs`
<ngz>I think those return the empty string.
<andreas-e>The Makefile in build/texk/bibtex-x contains a line "LIBS = -lm".
<ngz>Ah. You're looking at the failed compilation while I'm looking at the source.
<andreas-e>pkg-config kpathsea --libs returns something for me in the build directory after sourcing the environment variables.
<andreas-e>-L/gnu/store/h8vmxcbvk8n25y6zjj17qsq0fncansxs-texlive-libkpathsea-20230313/lib -lkpathsea
<ngz>which looks correct.
<andreas-e>It looks like what we need. So the configure process does not seem to call it and add it to the Makefile.
<ngz>Yet there is "KPATHSEA_INCLUDES=`$PKG_CONFIG kpathsea --cflags`" in "configure"
<andreas-e>Yes, but we need the libs, not the includes! (Or maybe both; but here, the libs.)
<andreas-e>The Makefile has: KPATHSEA_LIBS =
<andreas-e>And later: bibtex8_LDADD = $(KPATHSEA_LIBS)
<andreas-e>And also KPATHSEA_INCLUDES =
<andreas-e>but this should not be our problem here.
<ngz>sorry I need to go AFK another 15 min =/ brb
<mirai>apteryx: very strange, over here the _non-offloaded_ builds always fail
<mirai>but the offloaded ones always succeed
<nckx>sneek: later tell elevenkb: Nothing. It's a monotonically increasing number. Start with (revision "0").
<nckx>sneek <botsnack
<acrow>Does the package ruby-nokogiri.patch have some special significance during an upgrade? or, perhaps more accurately, why does it's absence cause a routine 'guix package -u' to error out? Hmmm...
<acrow>Actually, it is a patch.
<andreas-e>ngz: I get the same error when just compiling the source code "by hand", without passing through "guix build".
<acrow>I should also mention this is only on a guix foreign disto installation.
<andreas-e>ngz: I have no idea, except for patching the Makefile "by hand", using a substitute* rule.
<andreas-e>Strange thing also, "make" runs "configure" in subdirectories.
<andreas-e>acrow: What do you mean by "package ruby-nokogiri.patch"? It is not a package, but a file in gnu/packages/patches. So are you building from source?
<andreas-e>It is required for building ruby-nokogiri, but it should simply be there in the right place.
<acrow>andreas-e: Yeah, it's a patch but why does it cause a guix package -u to error out? I was expecting the upgrades to simply pull in the substitute.
<nckx>Its absence where?
<ngz>ahhh, back. Minutes are so long round here.
<andreas-e>What is the error message?
<ngz>andreas-e: I don't mind patching Makefile "by hand", but I don't know what change to do.
<andreas-e>Append to "LIBS = -lm" the result of "pkg-config kpathsea --libs".
<ngz>I don't think there is LIBS = -lm in the source.
<ngz>It is generated at build time.
<nckx>ngz: Can you insert a phase between generation and execution? Usually, you can.
<andreas-e>ngz: Normally, between configure and make. Except that here make also does its share of configuring.
<jpoiret>acrow: patches are used to create the derivation itself, and the substitute is fetched only after the derivation is created
<ngz>OK. I can check what is the state of the build directory after `configure' phase.
<andreas-e>ngz: Or maybe, more easily: in the source, texk/bibtex-x/, replace KPATHSEA_LIBS by "-L/gnu/store/.../lib -lkpathsea", where the thing following -L comes from the texlive-libkpathsea package.
<ngz>After `configure', there is LIBS = (nothing) in the Makefile.
<nckx>acrow: Can you pull and try again?
<nckx>Substitution isn't relevant here.
<acrow>jpoiret: Thank you, that explains why the failure. But I still don't understand how to proceed because I have no ruby packages in my guix profile. I still need to be able to do updates.
<acrow>nckx: yes, I'm currently doing an new pull.
<andreas-e>ngz: Unfortunately changing the does not seem to help.
<andreas-e>ngz: Probably because we are not running automake.
<nckx>It's not related to derivation creation or ordering.
<nckx>This is not a build.
<nckx>It's a missing Guix source file.
<ngz>andreas-e: Trying to autoreconf first
<andreas-e>ngz: I have a simpler solution.
<andreas-e>In texk/bibtex-x/, replace the line
<andreas-e>KPATHSEA_LIBS = "-L/gnu/store/4dlkddc4hfbiswbfs4q71z59yy0msf9d-texlivebin-20230313/lib -lkpathsea"
<jpoiret>nckx: i think is not used at all for guix pull builds so it shouldn't matter
<jpoiret>could be wrong though, only acrow can tell us there
<andreas-e>where the path needs to be computed in the package definition from the texlive-libkpathsea native input.
<ngz>andreas-e: Let me try
<jpoiret>still that was missing
<acrow>nckx: Fresh guix pull yields 6113xxx and package is moving ahead without that failure. Problem resolved.
<jpoiret>huhh, well color me surprised
<andreas-e>ngz: The> transformation is done by automake and>Makefile by configure, I think.
<nikolar>cbaines: might be something to look into potentially
<ngz>andreas-e: It looks better! It's failing during tests, but I expected that.
<ngz>andreas-e: OK I'll investigate from here. Thank you for your help!
<andreas-e>ngz: Ah, I had forgotten to run the tests.
<andreas-e>Is it worth the effort compared to current texlive-bibtex8?
<ngz>andreas-e: The point is to move all non-web2c binaries out of `texlive-bin'
<ngz>andreas-e: and put them in their respective package, of course. For example it's strange that texlive-bibtex8 doesn't install bibtex8, but texlive-bin
<lechner>Hi, does Guix build for anyone from Git? " @menu reference to nonexistent node `Configuring Git'" and others
<nckx>It should build if your (Guix) environment is up to date.
<lechner>How recent, please? I run my own Guix
<lechner>I am on August 15
<nckx>August 16 ☺
<lechner>okay, thanks!
<Guest69>I installed browserpass-native but it doesn't work.  After some research I found which mentions "make -C $(guix build browserpass-native)/lib/browserpass hosts-chromium-user".  Now it works.  This should be mentioned in the package for a better user experience.
<nckx>The ‘out of memory’ SSH error just went away after several hours. Weird.
<nckx>Guest69: The ‘perfect first patch’ smell coming off of this is overwhelming, you realise that, right? ☺
<Guest69>nckx: Will do, but I want to finish my mail server first to use that email instead of my old proton.  Does that mean it would be okay to put it in the package description?
<Guest69>Also just to be sure,  I assume debbugs does not care that my mail isn't from some big company?
<nckx>Guest69: Sure, the package description would be better than an obscure bug report.
<nckx>Guest69: It cares not, and if anything I'm glad to see ‘real’ domains when moderating. They are more trustworthy.
<Guest69>Perfect, that means I can do it how I imagined it.  Will take some time since I didn't even read the GNU Guix manual.
<Guest69>Basically did learning by doing
<nckx>For the Texinfo mark-up, use @code{make -C …}, not @command{}.
<Guest69>Noted.  Could you explain why exactly, too please?
<nckx>Sure. @command{} is intended for the actual command name, that is the binary: @command{ls}, @command{cat}. @code{} (and @samp{}) are for ‘literal programme text’, a fuzzy term that covers things like shell command lines.
<nckx>Mind you, I'm almost certain that I've erroneously committed a @command{ls -lR} at least once :)
<nckx>Does it matter? Good question, but no point in knowingly doing things wrong.
<Guest69>Ah good to know.  Makes sense
<apteryx>when I run the REPL with 'COLUMNS=800 ./pre-inst-env guix repl', (getenv "COLUMNS") returns "106"
<apteryx>is there a way to force that value? I guess xterm uses that for itself
<martin2020>Hi, I installed virt-manager, qemu, and libvirt, and added the libvirt service to my /etc/config.scm, but libvirt still tells me: "Could not detect a default hypervisor. Make sure the appropriate QEMU/KVM virtualization packages are installed to manage virtualization on this host." I think maybe I need to enable KVM somehow? How can I do that on Guix System? Thanks!
<martin2020>*virt-manager tells me:
<lilyp>martin2020 IIRC you just need to add your user to the kvm group
<lilyp>apteryx: might actually be an ncurses thing
<martin2020>Thanks, I tried adding "kvm" to my users supplementary-groups in /etc/config.scm, but it didn't have an immediate effect. I'll reboot and brb
<martin2020>Hmm, still same result. lilyp, do you have virt-manager working on Guix system?
<mirai>martin2020: you need to reconfigure the system
<mirai>merely editing the file isn't enough
<mirai>(by reconfigure I mean 'guix system reconfigure …')
<mirai>ACTION afk
<Guest69>martin2020: I have libvirt instead of kvm and virt-manager works like it used to be for me
<martin2020>mirai, thanks. I believe I did that, but I can double check
<martin2020>Guest69, did you have to do any configuration? I have virt-manager and libvirt installed. Is your user part of any other virtualization groups, besides kvm?
<Guest69>martin2020: I have only this and added my user to the libvirt group by added it to the supplementary groups
<Guest69>This should work.  If not send us your config
<Altadil>martin2020: I may be remembering wrong, but I think I had to launch virt-manager as root with sudo for the initial configuration.
<martin2020>Okay, I will be right back after reboot
<martin2020>Hmm I edited+reconfigured again to add libvirt as unix-sock-group and add it to my supplementary groups, but that didn't work. Running as root didn't work. Still says, cannot detect default hypervisor
<martin2020>I will try to paste my config
<nikolar>i an running guix/hurd as childhurd
<nikolar>i want to know where the default config.scm is
<martin2020>Guest69, here is my config:
<Guest69>martin2020: Where is virtlog?
<mirai>apteryx: re enblend, do you want a v3? About the comments, I do intend to forward them upstream, just haven't done so yet
<mirai>the whole docs being rebuilt over and over already reeks by itself