IRC channel logs

2021-05-21.log

back to list of logs

***rekado_ is now known as rekado
<rekado>zimoun: I guess we should upgrade R to 4.1.0 first and then do the Bioconductor upgrade.
<zimoun>hi!
<zimoun>rekado: yes. I should submit the remaining Bioconductor move today, I hope.
<rekado>excellent
<zimoun>and yes, I agree R should be updated first, before updating Bioconductot
<rekado>I feel like I should continue work on the GWL
<zimoun>yes, me too. :-)
<rekado>so far there was no feedback on the DRMAA branch
<rekado>so I guess I’ll just clean it up and merge it.
<zimoun>ah, talking about guile-drmaa, I had a discussion with Yann from CCIPL (Nantes) using Guix on their cluster. And an potential user of guile-drmaa should be Cuirass… but not on the list of mothacehe’s stuff, yet. :-)
<zimoun>I had not had the time to checkout out the branch and test it.
<zimoun>So merging it seems the best move. :-)
<civodul>we need training sessions! hint hint ;-)
<zimoun>civodul: yeah, I spent part of this morning to do that :-)
<zimoun>Almost full switch from one Conda user. ;-) Ah worse, I discovered that people are misusing Conda… the situation is even worse that I thought. ;-)
<zimoun>well, I have scabality issues though. :-)
*civodul .oO( how is "misusing Conda" different from "using Conda"? )
<civodul>yeah i think setting up sessions together with rekado, PurpleSym et al. would be sweet
<PurpleSym>What kind of sessions?
<zimoun>PurpleSym: it is related to the (private) email 86im40vycp.fsf@gmail.com. :-)
<civodul>PurpleSym: introductions to Guix for reproducible science/HPC, with hands-on getting-started sessions
<civodul>i think several of us did that already at work
<zimoun>somehow, people are using Conda, modulefiles etc. and they need help to switch. New “paradigm”, how do I do X with Guix? etc. At the end-user level, sysadmin level, etc.
<civodul>joining forces and reaching out to others could be fruitful in many ways
<civodul>yeah
<PurpleSym>Ah, I see. Yeah, I’ve been silent, because there’s really no-one here, who would use Guix without a big layer of GUI.
<PurpleSym>Our sysadmins are moving to Docker, as far as I know -.-
<civodul>right, you've developed a nice UI layer on top of Guix, so i suppose you're in a somewhat different situation
<PurpleSym>We’re hiding all the nice things about guix from the user :)
<civodul>perhaps that UI could also be part of a training session though, dunno
<civodul>heh :-)
<civodul>they don't even get to see parens, how sad!
<PurpleSym>They don’t even know they’re using GNU Linux ;)
<PurpleSym>Seriously though, the UI is being used to teach students and there’s an upcoming workshop. But we’re simply on a different level here, trying to convince people to use R instead of some proprietary statistics tool.
<zimoun>PurpleSym, here it is the same. I had to explain with details why SAS is not an option; for a better or a worse, the only argument which had “worked” was money: who will pay the license? :-)
<rekado>I’ve seen a number of people abandon Guix as soon as it became inconvenient
<zimoun>for instance, I would like to setup something as your solution. But I am running out of time. Therefore, I would be an attendee for a training, not about using it but about help in setuping. :-)
<rekado>e.g. some tool is provided via Conda and the developers of that tool recommend Conda
<rekado>bam! Back to Conda they go.
<rekado>I don’t see much value in converting users like that.
<rekado>they haven’t felt the pain yet.
<rekado>on the other hand: one of my colleagues has finally decided to go all in with Guix
<PurpleSym>zimoun: Yep, money is pretty much the only argument to use R here too. Free software? Nobody cares.
<rekado>this is the result of years of sharing Guix blog posts, demonstrating reproducibility benefits, and a couple of comments about “solved problem” whenever an R upgrade would wreck their projects.
<PurpleSym>In theory you should be able to read the documentation here https://github.com/leibniz-psychology/psychnotebook-deploy and set it up yourself. Obviously there are a few hurdles.
<rekado>luckily people here are rather intolerant of proprietary software
<rekado>they will use proprietary software in a pinch, but there’s a collective groan when a pipeline depends on proprietary tools, and an unwillingness to use it.
<zimoun>what I find interesting is that using “open source” (transparency about the source) *is* the only scientific method. Otherwise, the name is not Science but cooking, artisans, etc.
<civodul>+1
<rekado>exactly
<civodul>rekado: yeah, i've seen that too, people who just want to get something installed, so they pick whichever tool gets them to that point
<rekado>there’s a little grey area, though. There are many tools that are “academic use only”.
<civodul>which is understandable
<rekado>that’s transparent, but restrictive
<civodul>would an editor be able to run the "academic-use only" software? probably not
<rekado>civodul: what’s unfortunate about these users is that they attempt to mix and match — and that usually won’t work well.
<civodul>yeah
<rekado>so they abandon Guix completely
<civodul>i think we need to do more outreach and more reflection/self-criticism
<rekado>my guix.install procedure for R was the final motivator for my colleague to make the switch
<civodul>ah that's nice
<rekado>I think we need a little more of that kind of thing. To bridge the last few gaps between tools.
<rekado>even though they are not strictly part of Guix
<civodul>true
<civodul>"my" new intern started this Monday!
<civodul>to work mostly on repo2docker
<rekado>neat!
<PurpleSym>Nice, a bridge into the binder world.
<rekado>we’re working on a new PiGx pipeline, and this time every developer uses Guix (we have a manifest with inferiors to all have the exact same environment).
<zimoun>yes, I agre. On wednesday, explaining Guix a colleague said me, «I need more flexibility at the R level» and I said, yeah something exists but I was unable to reach guix.install. Well, I agree that “bridge” are missing to better fit how people concretly works.
<rekado>civodul: one of the biggest problems in my experience is the glibc ABI incompatibility. There doesn’t seem to be much we can do about it.
<rekado>zimoun: I want to add it to our r-minimal package.
<rekado>bold move, but … that’s the best place for it, I think.
<zimoun>And this same colleague installed Guix on RedHat… then “guix import cran -a bioconductor” failed (leading to a discussion on #guix about certificates ;-)), then “guix install nss-certs” failed requiring --fallback, then even with it is still failing for mysterious reasons I had solved yet. Well, if I am not at the next door, then they abandon Guix.
<rekado>I often find myself posting the Application Setup URL
<rekado>perhaps there’s something we could automate.
<rekado>(not automate posting the URL…)
***zimoun` is now known as zimoun
<civodul>rekado: you mean the glibc locale incompatibility? or the NSS?
<civodul>probably both i guess
<rekado>no, I mean the fact that mixing Guix things with non-Guix things is not possible.
<rekado>Guix is completely separate from the system, which is both great and annoying :)
<rekado>I’ve had a lot of problems switching to Rstudio from Guix because even with my changes to support multiple R versions I can’t let users select the system R.
<rekado>this is, of course, a problem with Rstudio’s design (because they link with libR.so instead of simply spawning an R process like ESS)
<rekado>I was wondering if … a “leaky build mode” could ever be made to work.
<rekado>for Rstudio I’d have to build one component (the one responsible for launching the user session) in two variants: one using the system’s toolchain, and another one using the Guix toolchain.
<rekado>whatever — this is just an example of the general problem, which is that the hard ABI-incompatibility boundary makes Guix rather special among package managers.
<civodul>rekado: ah yes, true
<zimoun>does this “leaky build mode” could help for icc?
<civodul>i agree that can be annoying, but that's also a strong point, depending on how you look at it
<rekado>yes, for sure
***Server sets mode: +nt
<civodul>zimoun: i believe all the slides are now available at https://hpc.guix.info/events/2021/atelier-reproductibilit%C3%A9-environnements/
<zimoun>civodul: \o/
<zimoun>non-french reader could be interested by https://hpc.guix.info/static/doc/atelier-reproductibilit%C3%A9-2021/arnaud-legrand-debian-docker.pdf ; a really insteresting attempt with Docker+Debian for the 10 Years Challenge :-)
***Server sets mode: +nt
<zimoun`>rekado: Move done! See bug#48575.
<zimoun`>Tomorrow, I will start the Bioconductor update if you have not yet.
<zimoun`>civodul: Debian science discussing CUDA: <https://lists.debian.org/debian-science/2021/05/msg00030.html>
<civodul>bah
<civodul>so we had: texlive-texmf-20180414/share/texmf-dist/fonts/enc/dvips/fira/fir_7gpamp.enc
<civodul>where is it in texlive-2019*?
<civodul>is terrible, i'm stuck on texlive-2018, and the big one, even
<rekado>it’s not in the tlpdb
<rekado>the fira package looks like the most likely candidate.
<rekado>it’s got similar looking files
<rekado>e.g. texmf-dist/fonts/enc/dvips/fira/fir_7gskcl.enc
<rekado>we have no fira package yet
<civodul>it's surprising that it's not in the big texlive package
<civodul>"guix import texlive fira" produces something though
<civodul>but without 'source'