<jpoiret>there definitely is something wrong with polkit/dbus
<samplet>jpoiret: You’re right: I can’t do anything with the network.
<roptat>I'm trying to find out how to use guard*, but I get &non-continuable whatever I try...
<roptat>it works as long as the body doesn't raise a condition, but when it raises &opam-not-found-error, first it's not catched by the (opam-not-found-error? c) clause, and second I get &non-continuable after the execution of a #t clause I added to check if it went through the first or not
<zimoun>sneek later tell civodul: on machine A, julia builds. On machine B, fails at test LinearAlgebra/test/matmul.jl:155.
<GNUtoo>As I understand, the userspace part doesn't have this issue though
<GNUtoo>I found out about it because libvirt depends on ZFS (probably for the userspace part), but for me the kernel module failed to build with 'make: *** No rule to make target 'arch/x86/Makefile_32.cpu'. Stop.'
<GNUtoo>and make install has some (string-append "INSTALL_MOD_PATH=" moddir) argument
<GNUtoo>So that makes images with that module non redistributable
<GNUtoo>+ it's probably not a good idea to depend on something like that since we can easily end up in situations where our interest is that copyleft becomes void (to enable to use and redistribute the zfs module)
<GNUtoo>Should I raise that issue on the mailing list? or in a bugreport?
<GNUtoo>Or should I just try to send a patch to at least remove that compilation process?
<GNUtoo>And would I need to patch the source to remove the module code?
<GNUtoo>so that guix build --source would yield a source that doesn't have the module and that can safely be built
<apteryx>IIRC the way it is distributed in Guix is that it isn't; Guix provides you the package, but you have to build it locally
<apteryx>I'm not sure I'm following though; what triggers this? which package in guix is problematic? I don't see anything zfs about libvirt
<apteryx>sneek later tell civodul haha! I got the zombie reaping thing going by forking, calling my new (set-child-subreaper); and invoking the test by (execlp "tini" "--" ...)
<M6piz7wk[m]><PurpleSym> "No explicit license means non-..." <- I would argue for unlicensed so that anyone can just fork and set their license on the repo as the original author failed to provide any authorization that the code is their
<M6piz7wk[m]>I did this to Terraria modders they were furious, but had to adapt GPLv3 after I did it :p
<fcw>PurpleSym: I wonder if the lack of response is because the author works for Google. Maybe there's some kind of clause in the employment contract that complicates open source work? I am not familiar with these copyright stuff.
<M6piz7wk[m]>and i gave left them a lot of ways to prove that they own the copyright to the code..
<M6piz7wk[m]>So going through this legally would probably mean lose for me, but poke's a poke :p
<abrenon>M6piz7wk[m]: being non-confrontational is also what allows a community to find harmony and a common base to go forward
<M6piz7wk[m]>from my experience being confrontational is more efficient in finding common base to go forward e.g. what FSF and it's supporters did with returning windows keys in the 80s or protesting in front of an apple store recently
<PurpleSym>vivien: It builds fine on master. Let me check…
<M6piz7wk[m]><PurpleSym> "6piz7wk: You’re not being..." <- depends on the interpretation of the wording and it's consequences 🤔 I don't think i mean any wrong doing as the projected end result is avoiding reinventing the wheel in guix packaging
<abrenon>there doesn't seem to be anything specific on this package
<abrenon>it uses exitcode-stdio test mode with cabal
<rekado>M6piz7wk[m]: you won't get away with framing your behavior as noble. As far as I’m concerned we’re not operating on the same foundations. I only help and socialize with people who act in good faith.
<vivien>Suspiciously, the hang only happens on the machine that boots with a message about rdrand giving spurious warnings and that I should boot with nordrand…
<Guest58>its a pain waiting for packages like latex to download lol
<jpoiret>you can use `mtr` to see if there's any packet loss along the way
<rekado>re latex: I strongly suggest trying to make do with the modular texlive packages, and to forget that the “texlive” package exists.
<rekado>I get slightly better speeds than 1.5mb/s on average, but my connection and hardware is pretty slow.
<rekado>I’d be happy if network engineers with plenty of time would be willing to join the Guix sysadmins to make sure we aren’t doing anything silly on ci.guix.gnu.org
<rekado>it’s entirely possible that we did something stupid and aren’t able to see just how stupid it was.
<rekado>that said: I think my colleagues at the MDC haven’t done anything silly — not in the physical patching of cables, nor with data centre networking.
<jpoiret>now that i've got the hang of elogind/dbus/polkit shenanigans, ill look into why gdm won't let you shutdown
<jpoiret>iiuc the shutdown operation is restricted to administrative users, but then we should be able to authenticate them for that operation on the gdm login screen (or do other distros just let anyone shutdown?)
<rekado>this has been a problem since I upgraded a month ago.
<rekado>ekaitz: remember to pet the yak before you take out the scissors!
<ekaitz>rekado: haha yeah... i'm just trying to package libappimage
<ekaitz>all this cmake scripts that download code from the internet are killing me
<jpoiret>civodul: do you know enough about guile internals to see if we could pass whole syntax objects in the builder, to get back traces from where they're defined in the guix sources? although that may reduce the builder's performance
<rekado>ekaitz: you’re not alone. It’s a drag having to patch cmakelists.txt files to undo the damage.
<notmaximed>jpoiret: Possible problem: rebuilds would be needed if line numbers change. Possible work-around:
<notmaximed>only use syntax objects if requested (e.g. --with-backtraces=some-package and(package (arguments `(#:backtraces #true)) ...))
<notmaximed>The G-exp code would need to be modified to include position information
<jpoiret>that would look like a good compromise imo
<notmaximed>And package definitions would need to be changed to use G-exps (#:phases #~(stuff ...) instead of #:phases (stuff ...))
<jpoiret>yes, that's also why I was thinking about that, moving phases to gexps
<jpoiret>well, rather than that, isn't the whole argument supposed to be moved to a gexp?
<notmaximed>sexps can be used in most contexts were gexps are expected
<notmaximed>though it would avoid all these eval etc. in the backtrace
<notmaximed>So seems like an improvement, though without useful position information.
<notmaximed>jpoiret: Ignoring #$, #$@ and #+ for now for simplicity, what about keeping a table from ((position of character in string encoding of G-exp) -> position in the source code) and compressing it?
<notmaximed>Whereas a tree structure like (this (that ...) (p q) ...) would need relocations
<notmaximed>Keep in mind that if you work with a syntax object, it will need to be serialised & deserialized somehow to be sent to the builder.
<jpoiret>yes... i'll try looking into the definition of the syntax object, that'll give us hints as to what is needed
<robertp>Hi. I'm a Guix System user, and I'm trying to manage my homedir config files. I can see `guix home` could do this, but ideally I'd be able to manage these via my /etc/config.scm (for the transactional rollbacks). I can't see how to do that. Is it possible?
<jlicht>I still see the "error: integer expected from stream" when trying to build texlive /w substitutes enabled on c.u.f. Is this something that I can address on my end?
<nckx>Sigh. M6piz7wk[m]: Stop. Trolling. Consider this your last warning here or whatever. I'm tired of dealing with the community unease you generate for no reason. Your behaviour is so counterproductive I don't understand why you haven't been corrected out of it yourself, but it's not our full-time job to do so.
<civodul>jlicht: i suspect this has nothing to do with the branch, but rather something with your daemon, no?
<jlicht>civodul: Maybe? I'll try to reproduce on another (much slower) machine. Is there some nar(info) cache I can clear locally?
<jlicht>I also doubt it has anything to do with the branch, but from searching the archives, last time this error popped up it seemed something went wrong on the server (a broken/corrupted narinfo? just cargo culting here)
<nckx>You could try --max-jobs=1 and it should print each URL before the error it generates. Unless they were hidden in the great substitute URL hiding. Then, maybe, ‘guix system build’ will show them, analogous to how ‘guix install’ hides but ‘guix build’ shows substitute URLs.
<roptat>I have no idea what's an "escape continuation", and the manual is really not helpful when it discribes continuable stuff
<roptat>what does that even mean: "the handler is invoked in tail position relative to the raise-exception call", I don't know half the words in this sentence
<abrenon>sailorCat: it certainly depends on whether it's expected or not (i.e. if you're running something that you know may legitimately recurse a great number of time or if you're only running something you expect to run out of the box on some reasonably sized input data)
<civodul>zimoun: the easiest way to get a log is "guix build PKG --no-grafts --log-file"
<abrenon>"escape continuation" : isn't that a function you pass for something that is going to loop so that it knows what to run next once it's done ?
<abrenon>I read a post about iterators comparing python and scheme recently which used stuff like that
<abrenon>roptat: I think "the handler" and "escape continuation" are coreferent in this situation
<roptat>abrenon, I think I need to learn more about continuations
<nckx>When docs make you use the word coreferent they are very good docs. /s
<roptat>so, as I was completely clueless, I just figured that &non-continuable meant I needed to raise a continuable exception, whatever that means, so used raise-continuable, which ludo didn't like, and I'm just as clueless as before (:
<florhizome[m]>I see gdm gets fixed in c-u-f but greetd is a good and clean alternative. as far as I understand the commit also leverages a new service configuration list without elogind. That’s win-win-win to me...
<abrenon>hmmm sorry, probably it has a meaning in comp.sci. too ? I meant they refer to the same concept ? as in "the handler" in second sentence is a phrase that means the same as the "escape continuation" from the previous one
<zimoun>civodul, from local log file, how do I construct Cuirass URL?
<roptat>I think I understand "the handler" is the exception handler, so the code that runs when an exception occurs
<roptat>but then what does it mean to run in tail position?
<sailorCat>abrenon: I'm going to build a package, that requires tones of derived packages, that requires tones of derived packages. So, roughly few hundred packages in total. Perhaps that's a lot but I have a mighty PC with 16Gb RAM, so I don't think it can't calculate because of resource limitation.
<abrenon>isn't that a stack-optimisation compiler stuff ? like it says it doesn't need to push a new function frame on it ?
<roptat>abrenon, that's the documentation for "#:continuable? #t"
<apteryx>that was the motivation to have a quick 'guix gc --invalidate' in my case
<zimoun>apteryx, yes. But your guard (remove GCOptions::gcDeleteSpecific) should not fix my issue, i.e., bypass the phase for few elements. I think adding the guard if (options.maxFree >0) should bypass the phase for few elements.
<apteryx>ah, so you want a 'guix gc --invalidate one two three', I see
<florhizome[m]>Is search paths/native search paths only affecting build or also runtime search paths?
<notmaximed>zimoun: delete from store (both /gnu/store & the database) I think
<nckx>Nor I, zimoun, I'm just guessing from context. The other interpretation I could see is not even deleting /gnu/store/hash-foo but only removing it from the DB, but that sounds like it could introduce subtle bugs/unexpectations.
<notmaximed>nckx: For the other interpretation, there appear to be code paths in the daemon that will delete /gnu/store/hash-foo in that case before building, but it would be surprising to the user (looking like guix gc --invalidate doesn't do anything)
<nckx>Yeah, i see those code paths more as robustness against unexpected failures than meaning we can be this ‘lazy’ during regular operation.
<zimoun>I would delete the item /gnu/<hash>-foo but not traverse .links. Basically do a clean ^C^C on the «deleting ununsed links» phase. ;-)
<florhizome[m]>Other than that they have more then doubled since July. Probably c-u-f?
<vivien>There’s a rumor that libreoffice is written is java. I can confidently say that this rumor is not propagated by guix people, because they spend their day waiting for those [build CXX] to complete ^^
<civodul>mothacehe: oh, that's the 2nd gc wave, lemme see
<notmaximed>The rumor is not propagated, because we prefer our rumors in 'inputs' or 'native-inputs' instead of 'propagated-inputs' :P
<vivien>So I had opened this issue: https://issues.guix.gnu.org/51890, but it got fixed by someone else in the mean time, but since we used different approaches the discussion is not completely finished. As far as I’m concerned, this is fixed, but I don’t want to close it in case people have more things to say. What should I do?
<roptat>trying to "guix build -f doc/build.scm", I get "Failed to produce PDF for language 'de'!"
<roptat>but it's not a fatal error, so the PDF is just missing
<roptat>maybe (@end occurred when @iftrue on line 19292 was incomplete)
<podiki[m]>been a while since I updated my "desktop" profile (c-u-f waiting the past weeks), but I thought the cache step finished quickly and it was just where it says "building a profile with 15 packages" or whatever
<podiki[m]>don't think I can do the upgrade right now (will report back when I do), but I see gnome-themes-extra, adwaita-icon-theme, breeze-icons, arc-icon-theme, arc-theme, orchis-theme, papirus-icon-theme ....which is probably more than I need
<nckx>But you'll be building N packages at once (when possible) which not all machines/packages like.
<KarlJoad>Ahh... So Guix does not do it automatically... This machine should be fine with it; I don't know why packages would care which is downloaded first (in parallel) when they are all substitutes.
<notmaximed>Barring out-of-memory problems, building & substituting in parallel should always be fine
<nckx>They don't. I meant that some package combinations (I dunno, say webkit + chromium + icecat + something else in C++) will OOM some machines if built at once.
<notmaximed>--cores=more-than-one can sometimes be problematic
<nckx>Non-OOM --cores=N failures are bugs, though.
<nckx>Anyway, consider yourself warned, etc. Don't do this again.
<M6piz7wk[m]>what do you expect from me then.. i am politically active so you will probably find lot of things justifying hiring someone to beat me up irl without understanding the implications and motivation for my actions
<M6piz7wk[m]>nckx: i feel warned it's uncomfortable making me think that i just spend lot of time and effort learning guix and guile only for you to ban me because you disagree with my politics >.>
<nckx>I expect you to behave in a way that doesn't reflect poorly on this community to outsiders, and in a manner that doesn't poison the atmosphere inside of it.
<nckx>I don't care about your politics or even know enough about them to know whether or not we disagree ☺
<M6piz7wk[m]>I am happy to disclose that i am not affiliated with the projected and i don't want to be to this level where my behavior would be representative of GNU excluding my agenda for FSF member conference and activities i cooperate on with FSF and FSFE
<M6piz7wk[m]>but i am not trolling and being controversial and aggressive is kinda in my nature >.>
<samplet>jpoiret: So now that polkit can talk to elogind, you discovered that elogind can’t talk to polkit?
<M6piz7wk[m]>maybe you can create a #guix-dev that is only for development and keep #guix as offtopic and for novice help from which you can ban me at any time without preventing me to contribute to guix? ^-^
<roptat>M6piz7wk[m], also technically taking someone's project and applying a license doesn't make it yours. On the contrary, this is a violation of their copyright :p
*M6piz7wk[m] can already tell that if you have this approach that he will get banned soon
<M6piz7wk[m]>roptat: only if they can prove that they are the rightful owner of the work :p
<nckx>M6piz7wk[m]: ‘being non-confrontational is boring’ I strongly suggest you give it a whirl anyway. Bye! No, #guix is #guix, #guix-whatever is #guix-whatever, and there is not a single #guix-edgy-trolling.
<roptat>M6piz7wk[m], well, commits from 12 years ago vs a commit from today... I don't think the judge will be in your favor
<nckx>M6piz7wk[m]: They don't have to prove anything. You know this. This is not a national thing. The laws are the same where you live.
<M6piz7wk[m]>nckx: they do.. i had a long discussion with a lawyer about it O.o it's just generally accepted as that the original creator is the copyright holder as it's very difficult to prove otherwise
<apteryx>if they are command line tools we can wrap them, but otherwise
<apteryx>(they already would by the python build system in fact)
<apteryx>oh, I guess it was the fix to polkit that caused a massive rebuild
<rekado>I’ll add the 1.20 variant and brace myself for conflicts then :)
<apteryx>yeah, I can't think of a good solution; revert doesn't sound to appealing. If we were to follow such strategy for each package that negatively impacts a few, we'd be stuck in guix-past
<rekado>we can tell users to make use of package-input-rewriting (to rebuild their environments with the older numpy if they must use other packages together with numba), and to have separate profiles.
<rekado>I’m biased towards not breaking many bioinfo packages, of course, but I think in this case we can let users work around it.
<samplet>apteryx: Really? I did check with ‘guix refresh’ and it said only about 2000. Did I miss something? We are looking at a similar situation with elogind now.