<vagrantc>so, i run guix challenge on something i've built locally, it differs ... how can i download the substitute's version to compare against my local version (e.g. with diffoscope) ? <vagrantc>oh, i forgot that u-boot-tools fails to build on armhf (and aarch64) ***j-r is now known as jmr
***jmr is now known as j-r
<rekado_>I finally found the reason for my failing NFS system tests <rekado_>the VM uses overlayfs and you can’t NFS-export an overlayfs. <vagrantc>are there available tools other than guix/nix to generate the hash used for a source tarball? <tune>can anyone tell me what package to installed to fulfill the need for "pdflatex" when exporting an org file to latex and telling it to make a pdf? <tune>/bin/sh: pdflatex: command not found [3 times] <tune>I don't think I have that! I'll start there <tune>I was maybe overthinking it, searching 'pdflatex' directly <tune>oh my goodness... 2.42GiB <tune>yeah, I made a typo (installed) in my message a bit ago that may have suggested otherwise <Blackbeard[m]>I ask because sometimes latex can give me trouble with Spanish <rekado_>tune: you may also consider the modular texlive packages. <rekado_>pdflatex is provided by texlive-bin. <tune>I'm 80% done getting the full texlive so I'll probably just let it finish at this point <tune>oh.... that wasn't the only file <roptat>I'm trying to unbundle dependencies from opam <roptat>I've packaged all of them, but the last one did not install correctly <roptat>so it's not found and opam's build process tries to download it <roptat>this feels wrong: "LIBDIR = @prefix@@libdir@/ocaml/$(shell ocamlc -version)" <roptat>that's not the location where I expect it to install <roptat>yes, but it looks like libdir is somehow stripped because I don't get redundant paths <jonsger>is it intended that "guix system search" needs (gnu installer)? <roptat>but I already set libdir in the ocaml-build-system to point to the right directory, so the redundant part is the one at the right with "ocamlc -version" <jonsger>civodul: guix system search doesn't work then ***jamesrichardson is now known as Guest79622
<civodul>another quote for next time i give a talk about EDSLs: "SASS is a tool that empowers frontend engineers and provides a higher level abstraction for writing CSS files through capabilities such as mixins, variables, conditional statements, and so on." <civodul>variables and conditional statements FTW! :-) *rekado_ works on tensorflow again… <civodul>rekado_: re tensorflow, you have all my support! <roptat>I'm thinking about doing some actual OCaml contributions, to change the way modules are found <roptat>currently it's only possible through the OCAMLPATH variable, but that's impractical because it means we need to use propagated-inputs <roptat>I think if the format of the META file was changed to (optionally) include full paths of dependencies, we would not have that issue <roptat>and it would also improve the way OCaml packages work right now, because you would then be able to install packages that have incompatible dependencies for instance, even you only use opam and not guix <civodul>one contribution you could make is "alias opam=guix" :-) <roptat>I think I'd just need to modify the way findlib works, but I need to talk to an actual OCaml programmer to make sure I understand it correctly <roptat>what I meant is that I think this kind of changes would be accepted because it's not just good for us ;) <roptat>but yes I think guix makes much more sense for OCaml programming than opam <decent-username>Hi. There doesn't seem to be a guixSD channel on freenode, so I figured this one would be my best option to get GuixSD related help. <decent-username>I'm at the step where I need to create a system configuration file, but I'm not sure how to do that with my current file system set up. <kmicu>Hi decent-username, the default config presented in the manual has ‘;; The UUID is that returned by 'cryptsetup luksUUID'’. <mange>decent-username: How are your file systems set up? Are you using LUKS? <mange>Oh, wait, sorry, it's in that paste. I can't read, apparently. <mange>In that case, I'll point out that you have a typo to prove that I can actually read. "luke-device-mapping" should be "luks-device-mapping". <decent-username>It would be nice to have a more detailed explaination in the docs on how to set up an encrypted file system. <mange>It looks like you could replace the (uuid "...") with "/dev/sda2" (looking at "(guix) Mapped Devices" in the manual). I've never used LUKS with Guix before, so I don't really know anything more than what the manual says. <kmicu>It something is not clear in the manual then report back decent-username. There is a paragraph with ‘If you are instead planning to encrypt the root partition, you can use the Cryptsetup/LUKS utilities to do that (see man cryptsetup for more information.) Assuming you want to store the root partition on /dev/sda2, the command sequence would be along these lines…’ *kmicu has a working LUKS+Guix System setup (with encrypted root and that currently results in entering pass two times). <decent-username>computers aren't perfect and will never be, because software is written by and hardware designed by flawed humans. <kmicu>I avoided that issue by not using Qt xD <kmicu>(Though I still have a dormant style=GTK+ somewhere in ~/.config/Trolltech.conf. Fun times.) <decent-username>kmicu: Would you mind sharing the `mapped-devices' and `file-systems' portion of your config.scm? It could probably help me a lot. <kmicu>I use btrfs so sharing those snippets could be not helpful but sure, give me a moment. I need to boot into that laptop. <rekado_>supports OpenCL and can be used with Keras. <rekado_>Tensorflow (and people’s acceptance of it despite CUDA) is a big disappointment. <decent-username>rekado_: Is it even feasable to train a model on a laptop without CUDA? <jonsger>rekado_: is there already something Vulkan based for ML around the corner? <rekado_>decent-username: with Tensorflow? The answer is generally no. <rekado_>for some models Tensorflow with CPU only works fine. <rekado_>I spoke too soon: plaidml uses bazel as a build system, so building it from source will be tricky. <civodul>"(gnu packages tex)" -> "(gnu packages libreoffice)" ?! *civodul just spent a couple of hours trying to optimize the module graph <civodul>and also trying to get something useful out of gvpr <decent-username>civodul: It wasn't everything in vain. There is always something you can get out of it. <brendyyn>do they need to be optimised? I mean they're all a mess regardless :/ <roptat>why are propagated-inputs bad again? <roptat>ah, because it makes it hard to wrap a program that has a library with propagated-inputs, at least... <civodul>propagated-inputs lead to "profile pollution", and pollution is bad <rekado_>roptat: it also makes “guix package -i” confusing and can lead to conflicts. <rekado_>(e.g. guix package -i python-pandas; guix pull; guix package -i python-numpy —> conflict) <rekado_>I feel that something’s not right about the way we build pam entries. <rekado_>the order of records in PAM files matters <rekado_>but we leave it up to services to specify how they extend existing PAM files <rekado_>I think that this should be up to the user. <rekado_>for example, for authentication you may want to use pam_unix first, then pam_ldap, and then pam_deny as a required final fallback <civodul>rekado_: indeed, looks like there's something wrong here <rekado_>I don’t know what better user interface we could offer, though. <rekado_>we could let users extend the pam root service the hard way. <civodul>but perhaps we could specify how ordering works <civodul>or provide a PAM configuration knob to force ordering <rekado_>I guess we could have services provide lines or building blocks <rekado_>and a user-facing PAM service type would be used to assemble the blocks in the desired order. <rekado_>PAM files can include other PAM files, so we could have services generate a snippet in /etc/pam.d/snippets/ (or similar). <rekado_>the record types (auth, password, account, session) are well-known, so the PAM service type could take lists for each of the types. <rekado_>on Fedora they generate a single file “system-auth”, which contains the default authentication settings, which are then included in other files. <rekado_>I let this idea grow a little in the back of my head while playing with my nslcd VM. <civodul>i wonder if more commands should default to -v1 (instead of -v2) <civodul>for instance 'guix environment', 'guix pack', 'guix system vm' <civodul>ok, let's do that at least for the 3 commands above <civodul>rekado_: do you think you can look into the Guile-JSON migration? <jonsger>I really struggle with the documentation of the Dovecot service. Is there an example config file? Or even better is there an explanation how to translate this "dovecot-configuration paramater bla" stuff to valid scheme code? <civodul>jonsger: i'd suggest emailing help-guix and then getting people to add an example or two in the manual :-) <jonsger>so I don't get it how to nest them: e.g. add protocol-configuration (named protocols) for name "imap" <roptat>I don't have anything complex, but maybe it can help? <roptat>jonsger, my guess is : (dovecot-configuration ... (protocols (list (protocol-configuration (name "bla") ...)))) <jonsger>thanks roptat that works. very intuitive :P ***apteryx_ is now known as apteryx
<apteryx>sorry for cross-posting, but this question really belonged here, not in #guile: Has anyone experience using a guix pack in a Dockerfile? <apteryx>I'm interested in having it use an ENTRYPOINT or otherwise make it useful 'out-of-the-box', so that if a user run 'guix run guix-pack-image python', it just works (without the need to source some /gnu/store/...-profile/etc/profile first).