***rekado_ is now known as rekado
<efraim>Related to this but not HPC, I've been thinking how to use Guix to build actual Debian packages, using a Debian chroot base and Debian packages. <zimoun>back to impure derivation, it would help for many intel compilers used by HPC folks, no? <zimoun>the large adoptiom is such more complicated because 1. not all the tools are there (when they almost trivially are otherwise) and 2. we do not have enough manpower for fixing and helping each use case. <zimoun>having impure would help the transition <zimoun>as for --tune, the fact that people must flag raises them a potential issue about reproducibility. <civodul>efraim: in Nixpkgs, there was a thing that would run a build process in a Debian (or other) VM and use checkinstall to produce a package <civodul>zimoun: isn't the oneAPI thing free? <civodul>i can see the "pragmatism" argument for the __impure thing, but i think opens the door to a new class of troubles <efraim>civodul: I was thinking of basically replacing pbuilder (which is what I settled on for Debian packages), which uses debbuild or something similar for the actual packages <efraim>so the actual .debs created would appear in %output instead of in /var/cache/pbuilder/results <efraim>I wanted to be able to hack on packages on my desktop and then offload the building of (legit but unofficial) .debs to guix <efraim>and of course using the offloading feature to offload to other architectures as needed <efraim>I might actually package some of those for science-nonfree as a proof of concept as requested by some people at Cornell <efraim>that last part regarding the c/c++/fortran compilers <efraim>yeah, as part of my extra 2 weeks in the States I met with some people there to discuss guix and how it makes everything better and can help them with their riscv hardware design stuff <zimoun>civodul: about impure, I do not think the issue is free vs non-free. My point is the pragmatic one: many things just work out of the box when it is really hard with Guix. Something like impure would smooth a bit. For sure, the reproducibility is lost and so on. But it would help for the machine learning situation, among many others, IMHO. Where we do not have enough manpower to be attractive, again IHMO. <efraim>impure would also help work around node packages <zimoun>efraim: oh you are packaging for Cornell? <efraim>zimoun: It mostly falls under the NLNet grant and the riscv/guix outreach part of that, but yeah I'm helping them <efraim>it's where spike and riscv-pk came from, and some other related stuff that might make it upstream depending on how well I can unbundle stuff <civodul>i'm using copy-build-system for the 1st time (i think?); WBN to have an example in the doc <civodul>right now it's a bit hard to follow IMO *civodul pushed a draft of a blog post on guix-modules <rekado>early comment: there are a lot of parenthetical bits, e.g. “like Guix is—or, perhaps more accurately, guix shell”, “be delighted, or surprised,”, “Modules were—and still are”, ““loads” GCC 11.2 in your “environment”—really: in your shell“ <rekado>this leads to a somewhat halting reading experience <rekado>there’s a break at the comma here: ‘[the module] might refer to different patch-level versions of Python 3.8, or built with different dependencies or different build flags.’ <rekado>‘built’ no longer refers to ‘the module’ but to Python, which is a little confusing <rekado>huh, I did not know that module files are written in Tcl! <rekado>I was never curious enough to want to know :) <rekado>‘The generated module files work equally way with’ —> ‘work the same way as’? <rekado>I’ve never seen ‘gamut’ used as a replacement for ‘plethora’; I only ever see it in something like ‘runs the whole gamut of transformations’ <rekado>is it ‘the command-line’ or ’the command-line interface’ / ‘the command line’? <rekado>oh… there’s an ‘interfaces’ at the end of the sentence; I failed to parse it because the link anchor binds it more closely to ‘programming’. <rekado>‘uncommon for clusters admins’ –> ‘cluster admins’