IRC channel logs

<civodul>hello!

<civodul>i don't see zimoun yet but hopefully we'll be starting the hackathon real soon now

<civodul> https://hpc.guix.info/blog/2023/05/reproducible-research-hackathon-let-redo/

<flypaper-ultimat>cool!

<flypaper-ultimat>the BBB will be recorded, right?

<civodul>i don't think so

<civodul>it's mostly an intro to clarify what the goal is, i guess

<flypaper-ultimat>ah ok.

<civodul>videoconf starting now: https://bbb.inria.fr/cou-pmi-g09-ute

<zimoun>hi!

<flypaper-ultimat>I think i'm gonna work on https://rescience.github.io/bibliography/Boersch-Supan_2021.html, an R one which i definitely like more than python.

<flypaper-ultimat>does it seems like a a good candidate? They have a swh id in their paper :D

<civodul>having a swhid is a good sign :-)

<civodul>i'm not familiar with R but it should be reasonable

<civodul>there are many R packages in Guix proper, and then there's guix-cran if needed

<flypaper-ultimat>yeah thanks to rekado's hard work, Guix is the best R package manager.

<flypaper-ultimat>and other people's hard work as well ofc :)

<civodul>yup, definitely :-)

<civodul>i started looking at https://github.com/ReScience/submissions/issues/10

<civodul>it's Python, so rather easy, but it's machine learning and pretty much needs to run on a cluster

<civodul>that's gonna be the hard part

<zimoun>I am doing old OCaml stuff.

<zimoun>if I do “guix shell -C libx11 --search-paths”, why there is nothing about include/ and lib/ ?

<civodul>because there's no C compiler

<civodul>see https://guix.gnu.org/manual/en/html_node/Search-Paths.html

<civodul>(the note in the middle)

<zimoun>ah thanks. I knew about Python, R, Julia and co. but did not know about C compiler as well. :-)

<bkm>This is nice https://www.nongnu.org/skribilo/

<bkm>Links in Codimd are to papers under review, rather than already published

<bkm>Is it ok to examine these?

<civodul>bkm: why not, though i guess it's "safer" to pick already-published papers

<civodul>because then the code is in a "known good" state

<flypaper-ultimat>yay, made a renv.lock->specifications converter, and all the R packages where in guix.

<flypaper-ultimat> http://ix.io/4zdE

<flypaper-ultimat>what's a repo from an earlier hackathon with a guix.scm? cuz i imagine the "making the paper" bit would be mostly the same

<flypaper-ultimat>ah, found civodul's old one, https://gitlab.inria.fr/lcourtes-phd/edcc-2006-redone , that will make for a good example.

<bkm>Getting codes to compile. Will start on Guix setup in the afternoon.

<zimoun>civodul: meeting before lunch break?

<bkm>Is there a discussion now or after lunch?

<bkm>Not much to discuss for me. May want to examine others at a later date.

<zimoun>bkm, ok.

<civodul>zimoun: oops sorry, i got stuck in a discussion

<civodul>everything's going fine?

<bkm>Yes. Some of the computational requirements are heavy.

<bkm>May want to setup guidelines for cloud use.

<zimoun>if people want to share, we are on BBB video meeting

<civodul>yeah, the one i picked was meant to run on a cluster

<civodul>ACTION joins BBB

<bkm>Connection dropped. Taking a break. Back in a bit.

<zimoun>why many people write “GNU GUIX“? All caps! p.3 of https://saschafahl.de/static/paper/reprobuilds2023.pdf

<zimoun>civodul: who are P17 and PS20 in Table 1 of the paper above?

<zimoun>PS203 48 minutes 27 RB Summit 2022, Venice Developer Operating systems Scheme, C, and others

<zimoun>P17 39 minutes 20 rb-general mailing list Developer Operating systems C/C++, Python, Scheme, and others

<zimoun>:-)

<zimoun>Scheme is intriguing ;-)

<civodul>zimoun: i don't know, but it wouldn't be hard to find out :-)

<zimoun>« “I think there’s this big gap between scientific research and programming. [. . . ] I think bootstrappability and reproducible builds could be an enormous boost for free software, and inspire people to move towards free software and free software practices.” - P17 »

<zimoun>ACTION is playing the detective :-)

<rekado>zimoun: I thought it’s like a mental caps lock. “GNU” is all caps and the next G is a capital letter. What’s three more?

<civodul>:-)

<bkm>Hardware is everchanging, which is another challenge.

<civodul>yeah

<bkm>Guix seems to be packaged for OpenSuse and Ubuntu/Debian

<civodul>feel free to add the issues you stumble upon to https://codimd.math.cnrs.fr/DZB3kDhZTT-HxwIm-wc0Dw?both

<civodul>(wrt. hardware)

<civodul>bkm: i'm sure the Debian packager did a great job, yet i got mixed reports ranging from "it's perfect" to "it's broken"

<civodul>i wonder why

<civodul>ACTION has the beginning of something: https://github.com/civodul/replication-recanatesi-2015/tree/guix-environment

<bkm>civodul: Trying out OpenSuse first, uses rpm format, so may be able to package something similar for Fedora

<zimoun>civodul: cool!

<zimoun>I have error compiling old OCaml stuff. )-:

<zimoun>I am doubtful that the reviewer of ReScience C compiled and run the code; from their comment https://github.com/ReScience/submissions/issues/22#issuecomment-618279040

<civodul>oh! does it look solvable? maybe with help from #ocaml or something?

<civodul>newbie question: can i blindly copy and adjust https://github.com/guix-science/guix-science/blob/master/.github/workflows/build.yml#L12 and it'll Just Work?

<civodul>PurpleSym maybe? ↑

<zimoun>the issue is https://github.com/guix-science/guix-science/blob/master/.github/workflows/build.yml#L15

<PurpleSym>civodul: It should work, minus issues with the Guix Github integration.

<zimoun>Here, you can “guix system image -f docker -m manifest.scm”

<civodul>specifically, i need "Guix cache" and "Install Guix"

<zimoun>and use this image produced by Guix.

<civodul>zimoun: i don't think we use home-made images there, can we?

<zimoun>yes, uploading an image to DockerHub.

<zimoun>the spirit of https://gitlab.com/zimoun/zimoun.gitlab.io/-/blob/master/.gitlab-ci.yml#L9

<zimoun>from my point of view, it would be simpler to locally generate a Docker image using Guix and push it to DockerHub then use it for GitHub CI than “Install Guix” and manage “Guix cache” directly in the CI itself.

<zimoun>But maybe PurpleSym “actions” can work out of the box :-)

<bkm>CI is great. But things may change 5 to 10 years from now.

<PurpleSym>zimoun: Can you actually use custom images on GitHub?

<zimoun>you can on GitLabCI so I was assuming you also can on GitHub. Maybe you have right.

<civodul>the name "ubuntu-latest" is resolved on DockerHub, or is it a GitHub-specific image registry?

<flypaper-ultimat>zimoun: wrt locally generating a guixy dockerimage; yeah that was also my experience.

<flypaper-ultimat>at least on gitlab ci.

<flypaper-ultimat>any tricks for setting TZDIR in manifest.scm? just adding tzdata package doesn't do that.

<civodul>flypaper-ultimat: you cannot set it automatically, unfortunately

<civodul>well, you could create a fake package that has TZDIR in its native-search-paths

<flypaper-ultimat>yeah, i think i'll do that

<flypaper-ultimat>did a similar hack one time for emacs

<zimoun>I have something that run \o/

<zimoun>about Python not old OCaml :-)

<zimoun>civodul: how did you go about “actions”?

<civodul>zimoun: not done yet

<zimoun>argh I have forgottent why I have stop using GitHub. Because it’s painful only for psuhing… token blabla

<zimoun>One part is starting to compute something https://github.com/zimoun/mixed-coordination-models/tree/guix-environment

<civodul>there's no .github/workflow dir though?

<zimoun>not yet. :-)

<zimoun>how can I bin the directory ../images/ to ../images with “guix shell”?

<zimoun>because “guix shell --share=../images=../images” does not work.

<civodul>the right-hand-size in --share cannot be "below" the current dir

<civodul>*side

<civodul>sorry i meant the left-hand side

<zimoun>but “--share=../images=/images” works.

<civodul>ok

<civodul>ACTION makes progress: https://github.com/civodul/replication-recanatesi-2015/actions/runs/5390676910/jobs/9786385590

<zimoun>I do not understand why I cannot bind ’source’ to relative ’target’

<zimoun>for example “--share=../images=$HOME/images” is “below”

<zimoun>what I want is “--share=../images=$HOME/tmp/mixed-coordination-models/images” but without manually specificying the absolute path $HOME/tmp/mixed-coordination-models

<civodul>then remove $HOME; /tmp is fine too, no?

<zimoun>on my host, I am working in $HOME/tmp/foobar and then the container follows the same, no?

<civodul>you can map it to /tmp/whatever inside the container

<zimoun>hehe! But then I have to patch the Python code. The code expect ’../images’ from the directory baz.

<civodul>ACTION had to intervene on ci.guix to unbreak the "Install Guix" action...

<zimoun>on the host, I clone (whatever where), say ’foo’. Then foo contain bar and images. From foo/bar, I run “python3 run.py” which looks at ../images.

<civodul>45mn left!

<zimoun>Launching the container from ’foo/bar’, something appears to me missing because I have to know the absolute path for binding ’images’ correctly (only knows relatively from foo/bar).

<zimoun>Anyway, it does not matter much. :-)

<zimoun>civodul: the GitHub action you pushed 27 minutes ago, does it work?

<zimoun>ah no )-: « guix time-machine: error: failed to load 'channels.scm': No such file or directory »

<civodul>what directory is it supposed to run from?

<civodul>oh i guess i need the "Checkout" step...

<civodul>PR: https://github.com/c-torre/replication-recanatesi-2015/pull/1

<zimoun>In progress https://github.com/zimoun/mixed-coordination-models/actions/runs/5391150244

<civodul>yay!

<zimoun>and the computations seem OK on my local machine.

<civodul>going back to BBB?

<bkm>Seems to run. Have put notes on codimd.

<bkm>Can join BBB in a few minutes