IRC channel logs
2021-09-24.log
back to list of logs
<civodul>(currently not linked from anywhere) <civodul>we need to set up a notepad somewhere <zimoun>hi! I was writing to you civodul about that. But you are too fast and efficient. Paul added an item. :-) <zimoun>the notepad, I do not know. I never use this kind of stuff <efraim>Before someone suggests an etherpad, it'd be nice to have something that we can point people to and edit from now through the meetings <zimoun>I cannot reach it. Probably networking issues on my side <zimoun>civodul: I would enlarge HPC to «Scientific Context». The name GuixHPC is what it is. Something along this line «Users and developers of GNU Guix in scientific context and reproducible research» <zimoun> civodul: do you know how to quickly reach GriCAD folks other than email? <civodul>zimoun, efraim: yes, i think we should have something that people can start editing now <civodul>or wait, were you suggesting to *not* use a pad, efraim? <civodul>zimoun: the page reads "Users and developers of GNU Guix in high-performance computing (HPC) and reproducible science" <civodul>zimoun: re BBB, i can set up a room on Inria's instance <zimoun>civodul: I would replace the first sentence by «Users and developers of GNU Guix in scientific context and reproducible research». <zimoun>civodul: the pad works for me :-) <zimoun>ah, there is a Inria’s instance of BBB. Let use it so. <civodul>ok, i tweaked that by taking all the words :-) <efraim>civodul: I was suggesting that we use something that won't disapear after 5 minutes if we all leave the webpage <zimoun>civodul: thanks perfect! Do you want we try and provide the BBB instance? Or do we do on Monday morning? <civodul>efraim: normally it won't disappear before 1 month <civodul>a bad actor could still erase it, but that's quite unlikely <civodul>at any rate, it'd be good to allow participants to write <civodul>we can't do that with a Markdown file in Git <zimoun>civodul: all works! Nice :-) Except recent chromium from Guix, ahah! ;-) <zimoun>civodul: typo with the Markdown, I guess. <zimoun>civodul: yeah, hoping you haven’t done “apt upgrade” in the meantime ;-) <civodul>i had mistakenly suggested on birdsite that conda didn't have transactional upgrades and rollbacks, so a couple of people were angry <civodul>(birdsite seems to be a lot about being angry) <zimoun>well sometimes says Conda is another Modulefiles but with fancy CLI <zimoun>where do you all these answers? How birdsite works is really mysterious for me <zimoun>I should be old because I find email threads much more easier to stay on touch :-) <zimoun>As I wrote elsewhere, these “angry” arguments are implementation detail bikeshedding and miss the core point about why there is a “scientific reproducibility crisis” and how to move forward to fix it, IMHO. (At least fix what we can. :-)) <civodul>zimoun: do i get it right that they defined the same workflow using several tools/formalisms? <civodul>sounds like zimoun & rekado will have to provide a GWL example :-) <zimoun>yep, that’s my understanding :-) <zimoun>in their Nature article, they write «They [workflow managers] simplify pipeline development, optimize resource usage, handle software installation and versions, and run on different compute platforms, enabling workflow portability and sharing» <zimoun>To make a point, we should demonstrate that «handle software installation and versions, and run on different compute platforms, enabling workflow portability and sharing» is wrong for all the tools they use. <zimoun>because these workflow managers are not connected to any package managers. Other said, their claim depends only on the underlying package manager. :-) <rekado>snakemake installs either via conda or downloads singularity images <rekado>that’s one way of doing deployment <rekado>and from the perspective of someone who sees value in version strings there is little more that’s missing <rekado>if we want to show that this is not good enough we need to find a way to get our message across more succintly <rekado>…and have a good answer to the question why in guix you can’t just say “use samtools version xyz” and have it DTRT <rekado>I wonder: can we have a guixify-condaforge tool that provides an environment, builds a conda package, bundles it up, records references, and then lets you install it with Guix as any other guix package. <rekado>slurp up all existing bioconda/condaforge packages out there <rekado>(discard those that can’t be built) <civodul>i guess we could have an "ugly-conda-hack-build-system", say, that spins up a VM and does all that <zimoun>yeah that what Katherine suggested on guix-devel, IIUC. <civodul>are there this many packages on condaforge? <civodul>apparently you need (?) one repo and a bunch of scripts for each package <zimoun>yes, it is a kind of PyPI or probably more something like MELPA <zimoun>rekado, the message that I try to teach here is that “use samtools version xyz” is not enough because it says “use samtools source version xyz” and what about the binary version you run. Using the Milk and Yogurt analogy to explain binary. It works more or less… Bio people seems receptive, I guess. <zimoun>Here some slides that I have to convert to text. :-) <civodul>ah, depends, but i manage to access it <civodul>zimoun: looks nice! this is the talk you gave at work, right? <zimoun>civodul: about 403, I should have missed to save the visibility switch. :-) <zimoun>yes, it is the slides of this very same talk at work] <zimoun>iPOP-UP is “bioionfo” folks from Univ. Paris, more or less <rekado>re the samtools recipes: conda has build.sh and meta.yml. For samtools they seem to have a bunch of different recipes for different versions, each in their own directories. <rekado>I wonder if we can just reuse *their* tooling without complicated modifications, build in a base guix system, and then copy the results to a Guix output. (Then scan the output for references and make these things available when the package is installed.) <rekado>PurpleSym: I fixed the rstudio-server-multi-version patches. <rekado>I’m running the build with the guix-science package definition now and will then push my updated patches. <rekado>I tested it with R 3.6.0 and R 4.1.1. <rekado>previously, running rserver as root would break the multi-version feature as the wrong home directory was accessed (namely that of the user running "rserver", not the user account running "rsession"). <rekado>R versions can be specified globally (in /etc/rstudio/r-versions) or per user (e.g. /home/rekado/.local/share/rstudio/r-versions). <rekado>every version that should be available in the GUI gets a record consisting of “Path: /tmp/r-test/.guix-profile” and “Label: whatever you want”.