<nckx>Hi, ng0. Do you use opensmtpd-extras by any chance? I'd like to but have no idea how to point smtpd to the binaries. <ng0>isn't it in the searchpath of it? <nckx>I guess I can just add opensmtpd-extras as a system package and use /run/current-system/profile/libexec/opensmtpd/ explicitly in the configuration file, if that's at all close to what you mean. <ng0>No, I meant if there is something in the package definition doing that <ng0>if not, and it is not picked up by opensmtpd, that's a bug <CharlieBrown>I installed cryptsetup (as root) but "cryptsetup" isn't found, so I can't mount disks. <nckx>ng0: if you mean adding opensmtpd-extras as an input to opensmtpd, that's what I'd like to do but assumed you had good reasons not to. If you mean something else: sorry, I'm tired & 've been staring at mail server docs for 6h 25 straight ☺ <nckx>That's not true. I've also been typing and yelling at my computer. <ng0>well I meant like making it aware of the path of the -extras. I just looked at it and it doesn't include this <ng0>this: (native-search-paths) <ng0>but maybe this isn't what's needed <ng0>for the service it should be like cups does it.. but for the general case where Guix is on another system it needs to be solved too <dpg`>Is there a reason why `guix pull` doesn't use rsync? <lfam>dpg`: As opposed to downloading the entire source code snapshot each time? <lfam>dpg`: We're planning to rewrite / re-design `guix pull` to use Git, which I think will be even better than rsync for this use case <dpg`>lfam: yeah that would've been my second question, lol <lfam>Yes, ng0's link shows the problem statement and the discussion of how to improve `guix pull` <ng0>where are we with this? https pull works now <ng0>someone added a git module… what's left? <ng0>verificatin mechanism? <nckx>ng0: I thought native-search-paths were only for environment variables (which smtpd doesn't use — .../libexec is hard-coded at build time). <nckx>OK. I'll just try /run/current-system/... for now, then have a look at CUPS. Thanks! <lfam>ng0: The idea of "channels" and the potluck should probably be integrated into `guix pull`, but it's a lot of refactoring to consider... <ng0>right, that's what it's for… if you can solve this, would be great. I won't have the time <dpg`>semi-related to one of solutions on the bug rp; gpg signing doesn't seem to work properly on guixsd. Just says it can't sign the commit even after setting the right PATH <ng0>cool, fish completions will finally have a way to stop the local file search with <tab> when -f is given, in the release which comes after 2.6.0 <reepca-laptop>So correct me if I'm wrong, but it makes sense for the first thing to be installed with --fallback to take a long time to compile due to the need to make it not depend on the bootstrap binaries, right? <lfam>dpg`: It *should* work; several Guix developers work on GuixSD <lfam>dpg`: Can you share more info about how it fails? <dpg`>lfam: ahh. I left the "gpg GIT signing" part out <dpg`>i.e. git commit -S -m "blah" <lfam>Right, I figured you meant signing Git commits. <lfam>I can help you debug this <dpg`>lfam: [dpg:~/code/prog/guix]$ git commit -S -m "blah" (sign-test✱) <dpg`>error: gpg failed to sign the data <dpg`>fatal: failed to write commit object <lfam>dpg`: Are you able to make a signature outside of Git? That is, `gpg --sign foo`? <dpg`>Other possibility would be my pinentry messing it. <dpg`>I don't think git uses v1 <lfam>Did you set the Git configuration key 'signingkey'? <lfam>How about `git log --show-signature`? <lfam>Does it successfully invoke gpg to try signature verification? <dpg`>lfam: Yup, they all come up <dpg`>just realized I've been using `dpg` <lfam>I hope you'd use yourself rather than somebody else ***dpg` is now known as DoublePlusGood23
<lfam>Oh, you were talking about your IRC name. I thought you meant that you'd tried invoking `dpg` instead of `gpg` <lfam>Okay, and which pinentry are you using? <DoublePlusGood23>lfam: manually set to pinentry-program /home/dpg/.guix-profile/bin/pinentry-curses <lfam>In gpg 2.1, that's really the only thing you have to do to get gpg-agent working. On Guix, one also needs to set pinentry-program in the gpg-agent.conf file. <lfam>A big improvement over gpg 2.0 ***Guest46726 is now known as sturm
***andreh1 is now known as andreh
***andreh1 is now known as andreh
<j-r>Is this the place to talk about GNU Shepherd? <j-r>I'm running guix on top of Debian, and shepherd there. the shepherd deamon seems not to be reaping children. <j-r>Of course I don't really know what I'm doing and may have the services defined incorrectly. ;) <j-r>I may have it hooked into systemd incorrectly also... ***chipb_ is now known as chipb
<efraim>sneek: later tell ng0 how about inxi source #f, replace 'unpack, one input of github/inxi.../raw/inxi.tar.gz/gitcommit <rekado>reepca-laptop: that depends on which substitute failed to be downloaded <rekado>reepca-laptop: if it’s a package somewhere low in the graph, then building everything at this point from source might take a long time. <rekado>reepca-laptop: the problem here appears to be that hydra hasn’t build enough for i686 at this point <CharlieBrown>"I'm pretty sure Guix is entirely proprietary, because they only include the bytecode." Huh? <rekado>CharlieBrown: says who? Bytecode of what? <CharlieBrown>rekado: Oh, just some random guy who works with NixOS/Tryton. <CharlieBrown>rekado: I assume the form the Guile VM interprets the Scheme code into. <rekado>I guess without context I cannot say anything about this that would get us and that person to a better understanding. <rekado>ACTION re-enters the Java rabbit hole <Petter>ACTION puts some Java (band) on the stereo and hands rekado a cup of Java coffee. <_root_>but I am fuzzy what to download and how to go on with installation <Petter>Are you putting together the configuration? <rekado>_root_: to install GuixSD just download the image for your architecture and follow the linked installation instructions. <_root_>rekado, The full OS not installing the package manager on another os <rekado>_root_: yes, that’s called GuixSD. <rekado>‘GuixSD’ = Guix System Distribution. <Petter>And Guix is the package manager. <CharlieBrown>Petter: I'm not going to lie: it's hard to not call the OS "Guix." <CharlieBrown>I mean, the visual arrangement of the letters "Guix" makes it look like a Unix. <artyom-poptsov>civodul: FWIW, I've sent Guile-SSH 0.11.0 release announcement to to the guile-user ML. <brendyn>civodul: Could Guix do with a Roadmap? <brendyn> #:search-paths ',(map search-path-specification->sexp <brendyn>What's the point of ',? Don't they cancel each other out? <civodul>brendyn: it has something like that in the file 'ROADMAP', though a bit abstract and outdated <civodul>otherwise i usually mention the road to 1.0 in my talks <civodul>but yeah, we could/should formalize this <brendyn>You said to me that it's hard to stick to a schedule with a volunteer project, so I thought it would be a good idea <brendyn>I wonder if it can be org-mode based? If we have a repo with an org-mode file in it with free commit access, devs could add TODO's all they like with schedules and the like <civodul>ROADMAP and TODO are org files, but it might be more convenient to have something outside of the repo <jlicht>anyone else having issues with the latest master checkout about <syntax-transformer mapm>? <civodul>and maybe rm -rf ~/.cache/guile/ccache <jlicht>ah, shame on me :#. thanks civodul! <civodul>there was an ABI change recently in (guix monads), hence this error <jlicht>civodul: How would one figure out that this error means that the ABI changed? <civodul>normally, there's no way you can write code where a macro ends up being applied <civodul>because macros are supposed to get expanded at macro-expansion time <civodul>so if you end up in a situation where a macro is applied, that's a sign that an identifier that used to be bound to a procedure is now bound to a macro <civodul>ofc it'd be better if Guile could automatically figure it out and rebuild what needs to be rebuilt <jlicht>yeah, I have to admit that guile stack traces are still hard to read for me as well <efraim>Based on online instructions for building gccgo and the error message I might have to rebuild gccgo-7 with languages 'c,cpp,go' <civodul>my Emacs went out of memory a couple of times over the last few days <civodul>did that happen to anyone else here? <civodul>i don't understand what's causing this <catonano>civodul: it's working properly here. But I only have 5 to 7 buffers open, usually. You probably use it more intensively. <civodul>perhaps that was a long-running build in shell-mode that filled up the buffer ***andreh1 is now known as andreh
<civodul>i had "make check-system" running in M-x shell <Petter>Maybe you could do with a swapfile. <civodul>i don't have swap, i don't think it really helps ;-) <civodul>it might become useful again now that SSDs & co. are becoming faster <brendyn>I think there is an additional problem with SSD dying after a certain number of read/writes that makes swap still undesireable <cehteh>an you never swap as much that swap will wear out any SSD soon (just do some math, basically when you swap so much, your computer is unuseable anyway) <cehteh>and swap write have a much lower ratio than swap reads <brendyn>In any case I have 16GiB of RAM. I reconsider my choices when I start using it all up. <cehteh>brendyn: even on 16G machines I add zswap <cehteh>then swappiness=100 .. works pretty well <cehteh>(and some other configs) ... almost doubles the 'useable' ram capacity without noticeable performance impact <rekado>kaffe built but 27 of 149 tests failed. Not too bad! <cehteh>brendyn: because swapping becomes extremely cheap with zswap, so you can instruct the kernel to swap out (compress ram) early for rarely used pages, which in turn leaves more ram available for hot pages <brendyn>Maybe when I buy a new SSD I'll use that. How do you set it up on GuixSD? <cehteh>and that works as well with HDD based computers <brendyn>Should it be used by default in GuixSD/ <cehteh>maybe not default, but easily accessible <cehteh>there are some more configuration knobs to tweak <brendyn>I'd like to push GuixSD to go beyond 1.0. I don't want to just provide tens of thousands of packages and see how the fit together, but build a fully polished desktop distribution, so things like zswap interest me <efraim>I had swap enabled with my 128 GB ssd, enabling swap and building constantly for guix caused smart to report about 1% health decrease every 3 months <brendyn>I have not seen any clear explanation of the swappiness metric. Is not the default of 60 ok? <cehteh>without zswap you may want to keep 60 when you want a low latency system <cehteh># cat /sys/kernel/debug/zswap/{written_back_pages,stored_pages} <cehteh>... nothing hit the disk and still swapped out <brendyn>Seems like it is compression in ram followed by swapping <cehteh>and you need to enable zswap cat /proc/cmdline <brendyn>How would one configure a guix system to do that, instead of manually editing conf files <civodul>we should add cross-compilation support to cmake-build-system <civodul>because supposedly CMake does support cross-compilation <cehteh>brendyn: dunno, that should be somehow added to the system config <cehteh>i just show the values i use, havent tried to fiddle that into guix <brendyn>TIL I'm working on software that a Medical Center is using <cehteh>i just started 'gitk -all' on the linux tree here multiple times .. swaped 14GB (with 16GB ram) without any noticeable performance impact <brendyn>But is that actual swap or in-RAM compression? <cehteh>still in ram compression, if thats swap or not deepens on your defintion, its a subsystem which hooks inbetween the swap stuff <cehteh>doesnt write anything to the backing store yet, well thats the goal <civodul>artyom-poptsov: upon autoreconf of guile-ssh 0.11.0, i get: <civodul>configure.ac:96: error: required file 'examples/ssshd.scm' not found <civodul>configure.ac:97: error: required file 'examples/sssh.scm' not found <artyom-poptsov>civodul: Hmm, seems that my "solution" of a failing tests is not a complete solution. <civodul>i think it should be AC_CONFIG_FILES([examples/sssh.scm]) <civodul>because you really need to instantiate these files (i.e., turn the .scm.in files into .scm) <civodul>i'm testing this in the Guix package <civodul>rekado: did you try GUIX_DAEMON_SOCKET=guix://something at your workplace? <civodul>with commit 950d51c9d9a5107c5dac279da7d2e431134b5f43 it should perform reasonably well <rekado>I’ll take a look at this commit in a couple of hours. <brendyn>Great, I tried deleting corrupt packages from /gnu/store, now I have SQL error foreigh key constraint failed <civodul>brendyn: you should never ever touch /gnu/store files directly <civodul>that said, you can try "guix gc --verify" <brendyn>That didn't work so I just ran guix gc. Now I have to download 4GiB of substitutes again <civodul>also, what makes you say you had corrupt packages in /gnu/store? hardware failure? <brendyn>Because it failed on wayland, with gzip saying the data was corrupt <civodul>oh, but it's data coming from hydra.gnu.org that's corrupt, no data on your disk, i think <brendyn>Actually it seems to happen a lot so I just rerun the command and it gets a bit further each time <civodul>it's a problem with our servers that should be addressed now <civodul>however there are still corrupt items in cache <brendyn>x86_64, but this is with the GuixSD 0.12 image. I had worse issuse when I tried using my own image build from guix. <brendyn>I don't really understand how it works but the image did not have the same versions of packages that my main laptop has. like I have serf 1.3.9, but when I build install.scm, it has serf 1.3.8 on it <civodul>DoublePlusGood23: swish-e fails to build indeed, a zlib-related error <brendyn>If I build program A from source and it is not reproducible, then I build B which depends on A, compiling it with A, then there is an update to B with a substitute also available and i try to build it, will guix build it from source with my A, or download the substitute A and B, discarding my A? <brendyn>I'm updating Calibre atm. It's a bit of a nightmare. The developer is one of those people that doesn't care for package managers, and so everything is a nightmare <civodul>brendyn: it will take what's already in the local store <brendyn>Last time I tried to install, it looked like everything was being build from source. It was like my tree of dependencies had departed from hydras, and so I could never use a substitute ever again effectively <bavier>yesterday I had a problem running 'guix pull' on i686 because a bzip2 substitute was not available/corrupted. I was surprised to find that 'guix pull' does not honor --fallback <civodul>i fixed the corrupt bzip2 thing as well <civodul>there was a discussion on help-guix about the exact same story <DoublePlusGood23>civodul: yeah I see the zlib related parts. seems to be using uncompress2 incorrectly <DoublePlusGood23>Even if I remove it from the `inputs` the configure script is still picking up zlib.h . Is this correct? <bavier>DoublePlusGood23: zlib is an implicit input of gnu-build-system <DoublePlusGood23>bavier: ahh. interesting. Is there a way to blacklist it from a package expression? Or I guess I could try passing something to the configure script <bavier>DoublePlusGood23: passing something to configure would be easier <bavier>there's probably an 'ac_cv_<foo>' variable you could define <DoublePlusGood23>Looks like it passes all tests on ubuntu, but in a guix env it fails all four. <bavier>DoublePlusGood23: do you have a paste of the test logs? <hurdle>Will GuixSD work with Libreboot? <hurdle>And how is the general user experience for a day-to-day machine? Will xfce4, firefox, etc work? <janneke>ACTION is not 100% certain on the `etc' bit <janneke>hurdle: no, there are guix-usb img'es to dd onto a stick <Lambda_>I'm getting ready to install it, any pointers when configuring my gui? <bavier>Lambda_: do a minimal installation first, then reconfigure with a DE once that's finished <janneke>Lambda_: what i would do is install something very close to the vanilla desktop.templ and after booting that change add more personalized [desktop] stuff <efraim>do we suggest guix pull first? just saw the changes Ludo pushed today <hurdle>janneke: I was only able to find .xz file? do you build that into a bootable drive somehow? <janneke>efraim: i never do that, but it's great if we can [start to] suggest that <Lambda_>What do you mean when you say minimal install? <bavier>DoublePlusGood23: or send to bug-guix@gnu.org <bavier>DoublePlusGood23: it's "my" package, so I might take a look ***nyberg_ is now known as nyberg
<rekado>recently at work I get this error pretty often: guix package: error: corrupt input while restoring archive from socket <rekado>it happens with ‘guix package -p /path/to/something -i a b c’ <wingo>rekado: i think it means there is a corrupt nar in the cache <wingo>find the file it is talking about, download it, and try to unzip it -- that is where the error comes from i think <rekado>hmm, but the file already exists in the store <wingo>but it's hydra's version of the file that's borked <rekado>oof, I also get it without any involvement of hydra <rekado>./pre-inst-env guix gc --verify=contents,repair <rekado>and it immediately spat out: guix gc: error: corrupt input while restoring archive from socket <rekado>weird. Restarted the daemon and it’s fine now. <rekado>ah, simple explanation: ran out of space on the NFS share… <rekado>running ‘guix gc --optimize’ now… <DoublePlusGood23>bavier: bug rp sent. still getting used to gnus hopefully it's formatted and signed right <DoublePlusGood23>on an foreign platform (ubuntu 17.04) Any guix command returns `warning: failed to install locale: Invalid argument` this is even after setting up the `GUIX_LOCPATH` for the init file and exporting it in my .zshrc (though my understanding is the error comes from the builder) Any ideas? <rekado>DoublePlusGood23: is GUIX_LOCPATH set in the *daemon’s* environment? <DoublePlusGood23>line is `Environment=GUIX_LOCPATH=/root/.guix-profile/share/locale` and I did install glibc-locales under root <paroneayea>I played so much Liero with friends as a teenager <reepca-laptop>So does --fallback fall back on a per-derivation level (build locally only if this particular derivation doesn't have an available substitute)? <slyfox>should be per-derivation. global kill-switch is --no-substitutes i believe