IRC channel logs


back to list of logs

***rekado_ is now known as rekado
<PurpleSym>rekado: If you want to take a look at the patchset:
<rekado>random thought: let the GWL import Galaxy tool description files to build workflows.
<rekado>PurpleSym: I’m in a meeting now, but I can check it out later tonight
<PurpleSym>Sure, it’s not *that* urgent :)
<rekado>(I’d really love to hack on GWL again… so many ideas…)
<rekado>PurpleSym: just two comments about your patch set (I need to buy some food): “(if tests? …)” should be “(when tests? …)” and: I agree with both decisions that you pointed out as controversial
<bavier[m]>if I'm being asked about my take on the easybuild-vs-spack debate, it's a good time to respond with "Guix HPC", right? :)
<rekado>there *is* an easybuild-vs-spack debate…?
<bavier[m]>apparently in some circles where system admins are trying to decide how to provide their users with the software they need.
<rekado>“Guix HPC” is always a good answer; even if it’s just to figure out why they would *not* consider it.
<bavier[m]>yes, good point.
<civodul>bavier[m]: one practical consideration i find useful is portability across machines
<civodul>you don't quite get that with spack/easybuild because their rely on the underlying distro (aka "CentOS")
<civodul>and then, 'guix pack' when targeting machines with Singularity/Docker
<bavier[m]>I'll try to bring those capabilities up in the discussion.
<zimoun>bavier[m]: do you have pointers about these circles? I am thinking to prepare something for a French “sysadmin conf” (among others) about why Guix.
<bavier[m]>zimoun: I'm currently talking with just a single customer contact; I don't know where the rest of this conversation is happening.
<bavier[m]>besides "multiple customers in Europe", at least one being CSCS.
<zimoun>thanks for the information. Good to know. :-)
<zimoun>Wow, “guix environment --ad-hoc r-minimal r-seurat -- R --version” took more than 30min on my laptop, desktop and workstation; all running different version of Guix and Guix daemon, where most of the susbtitutes are available. “guix size” says 1.2G and assume I download at 1M/s, it means 1/3 of time is for computing stuff. Well, r-seurat seems a good use-case to check performances. ;-)
<rekado>seurat has a lot of inputs and all of them have a lot of inputs themselves
<rekado>so I bet a lot of the time is spent building the profile
<zimoun>I did not profile, it seems also spending time between downloadds, and last on the graft part.
<rekado>oh yeah, grafting…
<rekado>I sometimes wonder if we could make grafts substitutable and just download the grafted thing right away if the original doesn’t yet exist locally
<zimoun>Maybe. On the other hand, I did some stats about the number of grafts over the time compared to core-updatesor staging merges. Now, some CI bottlenecks are unknotted, maybe another process drastically recuding the cycles could be worth.
*rekado nods
<civodul>rekado: did you get more data regarding your substitute issues?
<civodul>we were getting real close to... something
<civodul>perhaps it could have helped understand what Florian Pelz reported in that area
<rekado>no, I ran out of steam and haven’t been able to get back to it yet
<rekado>when I stopped I felt that there wasn’t anything obvious I could try next.
<civodul>i'd say finding the difference between "works" and "doesn't work", but that's easier said than done
<rekado>yeah, I looked at the strace of the two cases and I didn’t see anything obvious.
<rekado>it doesn’t help that I don’t really have an up-to-date mental model of how this all works.
<civodul>maybe we'll find other leads eventually
<rekado>the daemon launches “guix substitute --substitute” and then pipes strings to it
<rekado>that seems simple enough
<rekado>but is that really all there is to it…?
<civodul>the daemon also passes options by setting the _NIX_OPTIONS env var
<civodul>but that's it