IRC channel logs
2017-04-20.log
back to list of logs
<quiliro>hello...is there a configuration to install guix with emacs to develop for guix? <quiliro>is it possible to use that machine with free software? <jonsger>quiliro: of course but not exclusively :P <apteryx[m]>What do people do to setup a build environment on a non Guix machine? I used to think installing tons of dependencies on Ubuntu globally was OK, but after using Guix, it feels very crude to mutate my system like that! <quiliro>i had problems with the parabola live usb <quiliro>then i tried triquel 7.0 and guixsd 0.11...they would not boot <jonsger>quiliro: slow mouse seems to be like not well configured but no wifi is "normal" if you only use free software <quiliro>jonsger: i would like to use it as a true hacker...with Emacs <quiliro>and no pretty window manager but a functional one <jonsger>yes. guix package -i i3-wm i3status dmenu <k`>quiliro: FWIW, I don't see any Macbook Air entries on h-node.org. <quiliro>jonsger: would you pass me your config.scm and the packages installed on your user? <quiliro>h-node has 9400m video card as not compatible <jonsger>quiliro: I haven't one I use LinuxMint :( I use guix just as package manager <quiliro>jonsger: what packages have you installed, just i3-wm i3status dmenu? <jonsger>for using i3 yes. and then you have to "enable" i3 in your config.scm <quiliro>how about viewing videos on youtube and using spreadsheet? <jonsger>just use icecat, it's a "gnu version" of firefox <ng0>humph. someone around with more insight in qemu-kvm servers and grub errors? <ng0>I get embeding error warnings <ng0>it could've been easy.. if the hoster support has no idea, I'll send them the server to dd <ng0>yeah, I'm at the point where when I log out now I know the server won't let me back in, damage done. Sent the hoster a notice for further questions. <ng0>The joy of being early testers :) <methalo_>failed to produce output path /gnu/store/..-pcre-8.40-bin', if I remove the (outputs '("out" "bin" "doc")) finish with success! <quiliro>impressive...32 packages more in guix in a week or two <quiliro>that means 1500 additional packages per year <Apteryx>Are some build tasks done directly under /gnu/store? <Apteryx>I put triggered a failure at the end of the build of gimp, and found out in the log that some operations such as generation of manpages were conducted straight under a /gnu/store/...-gimp-... location. <Apteryx>Oh, sorry that's false. The manpages are generated during the build phase. It's just some extra symlinks creation which happens later creates things in the store; normal as it's part of the install process. <quiliro>i could not boot guixsd usb on a macbook air <quiliro>civodul: is it good that root is in charge of nginx web server for a mirror of guix? <civodul>quiliro: re nginx, nginx has the ability to drop privileges <civodul>for instance, "user nginx;" in the nginx config file will have it switch to the "nginx" user <quiliro>why does mpv need to download the whole video before starting...i will read the manual to see what to do about it <rekado>quiliro: AFAIK it does not download the whole video. I have been using mpv to stream videos without having to wait for a complete download. <quiliro>civodul: i will test with other videos then <Petter>I'm curious to try the Icecat 52.0.2-gnu1 patch by Mark. But I fail at grabbing the patch without broken lines etc. How do you do this? <Petter>I don't have the e-mail in my inbox, just this web page. <Petter>Guess that suddenly answered how most other people do it. <civodul>Petter: i think Mailman also exports raw mbox files, but yeah, that's inconvenient <civodul>otherwise you could use a news client with gmane.org <Petter>I'll poke around a little, thanks! <snape>does anyone experience this warning while running git? <snape>fatal: read error: Connection reset by peer <civodul>snape: indeed, git.sv.gnu.org terribly slow right now <civodul>wingo: did you eventually manage to push? <wingo>civodul: i was able to push to guile-web an hour ago or so but not guile <wingo>or rather, right now i have a guile push that is hanging <wingo>could be worse, we could be using cvs :) <wingo>ACTION had to touch the beast today already <efraim>i'm getting error 502 from the repo <efraim>mbakke: i'm still getting the 'Unexpected pages_huge()' error <efraim>when I checked `cat /sys/kernel/mm/hugepages/hugepages-2048kB/nr_hugepages` i got 0 <efraim>to clarify, i tried with the --disable-thp configure flag <efraim>interesting, 'thp' in the configure printout says 0 <snape>ACTION is building Icecat 52 <efraim>i was going to try to bump moreutils, but 0.61 came out ~19 hours ago and debian hasn't packaged it yet <Petter>snape: Did you have any problems applying the Icecat patch? <Petter>Once you have a patch, it should just be running `git apply icecat.patch`. Right? <mekeor>Petter: maybe the patch was made based on a different commit than you have checked out? <mekeor>Petter: did you try to git-pull? <Petter>Could be. I haven't been able to update yet. I get 502 responses when I try. <brendyn>Anyone know where I can get a vps I can setup guixsd on, and pay for with bitcoin? <Petter>mekeor: You're right. I had quickly compared the file with the patch, and it looked ok. Looking again I see some changes. Thanks! <efraim>brendyn: people have had relatively good success converting existing VMs to guixsd VMs <amz3>I have 502 "Bad Gateway" when running 'guix pull', is it just me? <ng0>uh. maintenance at the git host of savannah? all repositories are unreachable on all channels <ng0>yeah.. maybe best leave a message after the beep :) <ng0>unfortunate time for me, setting up a server.. but it can wait <ng0>oh. we can set the sourceurl for pull, so I can temporarily pull from my own master, right? <ng0>is there some generic way to address cgit for the tarball of master? <ng0>append "snapshot/master.tar.gz" to the cgit url, if the feature is enabled <ng0>only problem, certificate issues. I'll just wait until savannah comes back <civodul>sneek: later tell alezost thanks for the online manual! could you update doc/htmlxref.cnf in Guix to point to it? <snape>Petter: yes, the patch does not apply, there are three lines to add to local.mk (icu4c CVEs) and end-of-line comments in gnuzilla.scm. <snape>*that should be removed, so the patch applies <civodul>rekado: do you have plans to look at the Conda patch set? :-) <wingo>civodul: any thoughts re: guile 2.2.1? i guess we just push an update when git recovers/ <wingo>once that lands i can prep a proper potluck patch pull-please (phew!) <efraim>mbakke: without the --disable-thp config flag jemalloc thinks my aarch64 machine does have thp <wingo>updating guile package to 2.2.21 <wingo>ACTION having trouble git pulling :) <wingo>it was a somewhat hasty release <civodul>yeah it works fine, at least on x86_64 <civodul>wingo: the 'thunk?' optimization worked as expected, BTW <civodul>i no longer see elf.scm in the profile \\o/ <wingo>i still feel like we should be able to improve that <snape>Petter: Actually, Icecat patch applies when guix repo is up to date :) It's just.. I could not fetch a few hours ago <wingo>like what would it take to get to 200ms <wingo>would need guix changes obviously <wingo>but if we could do it, it would feel magical <civodul>caching is what we should focus on in Guix <civodul>that's where we spend most time currently <civodul>because we use equal?/hash caching (!) for packages <civodul>but that's still faster than eq?/hashq <civodul>it's complicated, but there's room for improvement! <civodul>the other thing that takes time is 'write', which is called when writing derivations <wingo>so loading up all the package modules isn't a significant part? <wingo>i guess your test didn't include that <wingo>b/c the all-packages hash was already created <civodul>i don't think so, because it's a promise <wingo>well yeah but it loads like hundreds of modules <wingo>i guess if you measure from a fresh repl <wingo>in that case you could measure the cost of that by diffing between the first guix-build and the second <wingo>if i could ever get git to pull :P <civodul>not really, because the second one is instantaneous due to memoization <civodul>to benchmark module loading, we can just do a 'fold-packages' call in a fresh process <Petter>snape: I haven't been able to pull from git either. But now I have Icecat building anyways. Thanks! <ng0>i think boston is back, or at least there is now a timeout instead of an error <ng0>I guess it will be an timeout <thomasd>do we support cross-references in package descriptions? (like “x relies on @command{y} provided by package @pxref{z}”) <thomasd>ah, yes, I'm a texinfo newbie and can't remember the proper ref commands yet :) <civodul>how this is rendered would depend on the UI though <civodul>what would you do in a terminal? :-) <thomasd>it's probably a rarely needed feature, usually such dependencies would just be inputs <thomasd>in a terminal you could just type the package name? <thomasd>anyway, I'll take that as a no(-t yet) ;-) <wingo>probably you just want to mark a package as being a package <wingo>then figure out what presentation you would like depending on the medium <wingo>i guess texinfo macros could do that <ng0>It would be great if texinfo would have translation support one day <ng0>like for documentation generation <ng0>I'm not there yet with my documentation fixes, but I'm not aware of any way to have more than one language output to files from an full documentation "book" <ng0>unless texinfo can do something similar to sphinx with rst <thomasd>wingo: yes, we could just mark packages with “@package{...}”. Maybe I'll experiment with it at some time. <wingo>probably you'll have to add some hack to the texinfo parser, but that's not that bad <wingo>i think it might not know anything about macros but i could be wrong there <wingo>anyway a scheme parser not knowing about macros is unforgiveable :) <thomasd>and semi-related: some projects use a lower case name on purpose, but the guix package lint does not accept that. I suppose an exception is ok in such case? <wingo>ACTION is more of a victim of lint than an authority on it :) <civodul>we could make the texinfo parser extensible, such that we can define "texinfo macros" in Scheme <civodul>thomasd: it does not accept that at the beginning of the description, right? <civodul>you could do @command{foo} if that's appropriate <anthk_1>hello, can the GuixSD image be burn on a DVD? <civodul>or otherwise, just ignore the warning <ng0>but there are people working towards an iso <civodul>anthk_1: currently the image is meant to be stored on a USB stick <thomasd>ok :) indeed, sometimes the name of the package/project is not the name of a command, but still written in lower case. <anthk_1>can it be bootstraped from another distro? <ng0>in theory. some systems work, like minimal Debian <anthk_1>in example, booting from systemrescueCD and install GuixSD from here <ng0>recently wingo documented an Debian -> GuixSD process on the mailinglist for digitalocean cloud server <efraim>Do we have any plans to depreciate mysql in favor of mariadb? <efraim>It should work, I'd use the binary install as a guide to getting it and the guixsd install guide for actually installing, but being familiar with the install process first should be helpful <ng0>and also what needs to be cleaned up, etc <efraim>wingo: did you try making a swapfile before you started? I get a lot more bang for my $5 droplet with a 1 GB swap file <wingo>efraim: no but i did that later, and it is much better! <wingo>i did it manually tho; do you have a guixsd snippet for that? <ng0>doesn't exist (yet) :) <wingo>when i reboot i will have to turn it on again manually i think <ng0>oh.. you can specify it in the config <wingo>yeah that is what i was looking for, something to specify in the .scm <civodul>how do we deal with cmake not adding $out/lib to RUNPATH again? <ng0> (swap-devices '("/cryptoswap")) <ng0>is cURL okay this time, no immediate bug smashing release in sight? <ng0>guess I'll do the fork release then. <ng0>ACTION curses to no automation <ng0>my server hasn't catched up. <ng0>it's still 502 in iceland <wingo>ACTION should switch to guix pull, this guix environment guix stuff is madness <ng0>okay, done. I'll send the gnurl guix patch soon <ng0>washing machine calls <Apteryx>(the error downloading gmp). Seems the nar was corrupted. <amz3>does anyone has a "howto" for building a guix rootfs for an lxc ? <amz3>I think it's just a matter of put'ing together a os configuration and building the image and unpacking that image to a directory <anthk_1>where could I set up grub boot options? as video=1990x400 <lfam>ng0: What certificate problem did you have with `guix pull`? <lfam>efraim: Are you testing the new QEMU? If not, I'll do it <lfam>efraim: I've been tracking the release candidates and fixed bugs, so I have a commit message if you just needed that <efraim>Sounds like you're more on top of it than me. I just noticed that they tagged 2.9.0 <ng0>lfam: guix pull --url= with an LetsEncrypt url of my master.targ.gz of guix, but this was mostly 0.12 release, I'm past that now and bootstrapping my way to GuixSD <efraim>You can get it, I'm still working on jemalloc for aarch64 <lfam>ng0: Okay, I think that's expected with the 0.12.0 release unless you set an environment variable to help Guix find your local certificate store <mbakke>efraim: thanks for checking. if it builds --without-thp, I guess we can just pass it to all non-intel arches <efraim>It builds --without-thp, and neovim and mariadb both built successfully with --without-thp and tests disabled <mbakke>great! one further step towards ceph on aarch64 ;) <mbakke>ceph has a strange test failure on x86_64 though.. are the hardware specs of current build slaves available? <civodul>the results Should Not Depend on Hardware Details :-) <lfam>Should not, but sometimes do :) <civodul>in theory, the model matches reality ;-) <mbakke>it looks as if the compiler thinks SSE3 is available, but the test can't find it in /proc/cpuinfo <mbakke>all 64-bit intel cpus have sse3, but the very first 64-bit AMDs did not have it <mbakke>I'll try to hard disable SSE3 instructions for ceph <civodul>efraim: could you try to capture details about this? <efraim>@ substituter-started /gnu/store/whq5fsb6sdyxj8vycmw2kwgs38p7mpxx-lrdf-0.5.0 /gnu/store/2y1jx6lq4w61gz3ys8a29r6zxyajr0xm-guix-0.12.0-8.d72b/libexec/guix/substitute <efraim>@ substituter-failed /gnu/store/whq5fsb6sdyxj8vycmw2kwgs38p7mpxx-lrdf-0.5.0 256 fetching path `/gnu/store/whq5fsb6sdyxj8vycmw2kwgs38p7mpxx-lrdf-0.5.0' failed with exit code 1 <efraim>also re sse3, in qt some flags are -no-ssse3 and some -no-sse3, we should probably chose one or the other <mbakke>we should probably disable both for maximum compatibility <efraim>civodul: when I went to restart it it correctly started building it locally after failing the substitute <efraim>from qtbase configure: Enabled ones are still subject to runtime detection. <civodul>efraim: so it keeps restarting the substituter on that one? <Gareth422>Hey. I was wondering if burning the USB image to a DVD will work. <efraim>i didn't try --no-grafts, after I stopped it and restarted it it worked as exptected, failed and then fellback to local compilation <mbakke>efraim: In that case I'd guess it's okay. As long as it doesn't pass -msse3 to the compiler. <mbakke>Gareth422: I don't think so. But there is some discussion currently about generating an ISO image which should work. <Gareth422>I have an infinite stack of DVDs and pretty much no flash drives. <lfam>The staging jobset queue is decreasing rapidly :) <lfam>So far, no failures except for libressl, which is fixed on master <efraim>i should test mesa from staging on aarch64 to make sure it builds fine <Gareth422>Ah, man, I can't do any of that right now, since I'm on Windows. <mbakke>efraim, civodul: I just hit an infinite download loop too, for "sdl-union". <mbakke>lfam: tuxpaint fails the same way on "master" <bavier>lfam: I wonder if '#:parallel-build #f' would help tuxpaint <lfam>Find orphaned patch files: <lfam>for patch in gnu/packages/patches/*; do result=$(grep -rI ${patch##*/}) && test ${result%%:*} != "gnu/local.mk" || echo $patch; done <mbakke>is there anything inherently tying nss-certs to the nss package? <mbakke>AFAICT updating them independently works fine <lfam>From what I can tell, you're right <lfam>It would be nice to confirm that and really decouple them. I don't like the default certificate store depending on the flaky NSS build <mbakke>I'll reconfigure my system on the nss-certs update to be sure. <lfam>We should probably ask the NSS maintainers about it (or find some documentation), just to be sae <mbakke>since nss-certs only contain x509 certificates, and the nss ABI is guaranteed to be stable, I can't see what the relation should be <mbakke>the rare case would be if a new cert requires some new crypto primitives in nss <mbakke>but then we'd probably have problems with other TLS libraries too <lfam>Right, that's my understanding, too <mbakke>from following nss a while, it looks like new releases update either code or certs, so it makes sense to update them invidually <mbakke>I'll push the update. Geronimo :) <lfam>bavier: Can you take a look and decide what to do about these stray patch files? <bavier>lfam: I can remove them. I think the patches were used for a previous version that I had originally packaged, but I bumped the version before pusing <lfam>Cool, let's reduce that `guix pull` snapshot size :) <civodul>mbakke: if that's not too late, could you try to gather strace or whatever of the substitute loop? <bavier>might be fun to add a test that checks that all dist_patch_DATA patches are actually used in packages :) <lfam>bavier: Yeah, and I don't want to do it with POSIX shell! <lfam>So tedious building up lists and processing them there <civodul>Steap actually did that looooong ago as a shell script (i think?) <lfam>I punted with my script, to be honest. It only works because grep seems to check in 'gnu/packages/' before 'gnu/local.mk' <bavier>shouldn't be to hard?: diff `guile -e '...'` `echo $(dist_patch_DATA) | sed 's/ */\\n/'` <lfam>Ah, but you're cheating with guile, right? ;) <bavier>we do like to use it everywhere ;) <civodul>bavier: looks like you're ready for the challenge! :-) <bavier>oh, are we talking POSIX shell so that we could add it as a git hook? <lfam>No, I just did it with POSIX because that's what I know <lfam>I haven't bothered to learn any shell-specific scripting features <lfam>Oh well, it gave me something to do while I waited for a build to finish :) <mbakke>civodul: unfortunately it's long gone. will do if/when it reoccurs <rekado>civodul: I’ll take a look at the conda stuff soon. <rekado>ACTION apologizes for the late reply; parts of rekado are on holidays <mbakke>lfam: is it possible to cancel pending evaluations on Hydra? It would be good to have the nss fix in first. <mbakke>although, there is probably little overlap between ffmpeg+qemu and nss <mbakke>in more positive news, the "qtbase" package is reproducible across three different machines <reggggieee>long question: i'm having trouble using 'guild compile <file>' with guix's guile. it runs fine, except that it puts the compiled file in "$HOME/.cache/guile/ccache/2.2-LE-8-3.9/<the path to the scm file>.go", and I don't know what to assign GUILE_LOAD_COMPILED_PATH since it changes for every file I want to compile <lfam>mbakke: I don't think I can do it, but I did ask for something similar a little while ago on guix-sysadmin, so maybe somebody will take care of it :) <thomassgn>daviid: Hi, thanks for pointers a few days ago! Didn't check back in till now, forgot :) <didi>Can I install guix in /opt? <bavier>didi: yes, but you will not be able to take advantage of binary substitutes <didi>bavier: I am sorry, what is a binary substitute? <bavier>didi: a copy of a packages build result served from our build farm <bavier>didi: so, you'd need to build packages locally <Gareth422>I'm getting a 404 installing Flex, even after pulling. I'm trying to install GuixSD. <didi>bavier: oic Fair enough. I imagine because the paths are hardcoded. <bavier>np, let us know if you have other questions <snape>didi: what do you mean by "installing guix in /opt"? <lfam>Gareth422: A 404 for what resource? <Gareth422>What should I do to fix this? I tried using --fallback, but when I did that, it tried to install the entire thing from source. <bavier>snape: presumbably `./configure --prefix=/opt --storedir=/opt/store` or similar <lfam>It should print a URL that is returning 404. Can you summarize the URL? <didi>snape: I mean, not creating a /gnu directory. <lfam>Gareth422: I think you'll have to build flex from source with --fallback :/ <Gareth422>When I did that, it went for like an hour, building all of the packages from source. I eventually cancelled it. <lfam>Gareth422: Hmm... let me check if we still have the substitutes for 0.12.0. I'd hoped we would <lfam>Gareth422: If you use --fallback to workaround this issue with flex, you will still use binary substitutes for whatever else they are available for <Gareth422>Is it 404ing because it's been replaced by a newer package? <Gareth422>Is it possible to just install the newest packages, instead? <lfam>--fallback only takes effect when Guix expects to get a substitute but the substitution process fails. If Guix doesn't expect to get a substitute at all, then it will build from source <buenouanq>I have to use --fallback all the time - Is this a problem with the Guix repo server? <lfam>buenouanq: Yes it's a problem, please report it :) <buenouanq>I visited the FSF once and saw the old Guix laptop - I think y'all have a different one now though. <snape>didi, bavier: I think there is some work going on to allow Guix to be installed in non-privileged places. There was a talk about it at Fosdem. <lfam>Gareth422: Does that hash match the one that's failing for you? <Gareth422>Oh man. It's fallen into the framebuffer void. I'm not sure. <catonano>civodul: why are youuu rollingg back to guile 2.2.0 ? <sneek>Welcome back catonano, you have 1 message. <sneek>catonano, efraim says: only the older macbooks support libreboot. x32 is a hybrid/pseudo architecture of 64 bit kernel and 32 bit memory, in general improvements across the board if you are limited to under 4 GB of RAM <lfam>Gareth422: Sometimes you end up building everything from source becaues the name service cache daemon caches a lookup failure for gnu.org. It's similar in spirit to <https://bugs.gnu.org/22209>. So, you might double-check that you can reach gnu.org and, if not, `herd restart networking` <catonano>sneek later tell efraim thanks for explaining the old macbooks sense to me ! <lfam>Gareth422: Sometimes you end up building everything from source becaues the name service cache daemon caches a lookup failure for gnu.org. It's similar in spirit to <https://bugs.gnu.org/22209>. So, you might double-check that you can reach gnu.org and, if not, `herd restart networking` <Gareth422>It's weird. It downloads the first few packages, and only starts building after the first failure. <bavier>Gareth422: that's expected behavior <lfam>Ideally, later packages would not need to be built if substitutes were available <Gareth422>Why does it build everything, and not just the one package? <lfam>My understanding is that Guix can't find a substitute <Gareth422>So, Guix uses standard compile flags, so you end up with the same binary by building or downloading, right? It's not like Gentoo? <civodul>catonano: an obscure issue with "syntax objects" and my expectations about them <civodul>we'll see investigate later with wingo :-) <lfam>Gareth422: The idea behind offering "binary substitutes" is that because we account for the full dependency graph when building a package, we can be relatively confident that a package built on our build farm will be bit-identical to what you'd build yourself. <Gareth422>Is it possible to pass your own CFLAGS for packages? <bavier>Gareth422: it takes a bit of work <lfam>Sure, but not too much ;) <bavier>Gareth422: you'd need to create your own custom packages that override the CFLAGS used. <bavier>Gareth422: not necessarily, depends on what you want. <Gareth422>march=native, in case I wanted AVX for something. Other than that, I can't imagine any situations. <bavier>Gareth422: you'd probably want to recursively patch higher-level packages to refer to those optimized packages. The package-input-rewriting procedure in (guix packages) can help with that