IRC channel logs


back to list of logs

<rekado>we’ve had a number of GWL releases last year. Finally something I don’t have to rush before the report :)
<civodul>well done :-)
<civodul>there's no canonical blog post or article about GWL, right?
<civodul>something about The Vision
<rekado>true. Even the manual’s introduction is very light on why we’d even bother doing this.
<rekado>I only ever discussed this in my presentations at FOSDEM
<zimoun>rekado: I was invited a seminar back on December where I presented Guix. I had questions about GWL. Yeah!
<zimoun>A The Vision would be worth, IMHO. :-)
<PurpleSym1>Hm, is Guix’ channels.scm order sensitive? I’m getting different profiles with guix pull -C vs guix time-machine -C and the only difference is that guix describe reports a reversed order of channels.
<civodul>PurpleSym1: could you tell what the difference is between those profiles?
<civodul>as in "diff $P1/manifest $P2/manifest"
<PurpleSym1>civodul: Like `guix describe` the order of channels differes.
<PurpleSym1>That’s all.
<civodul>PurpleSym1: ok, that's not great, but normally inconsequential
<civodul>we could/should canonicalize channel order
<civodul>+ implement hard collision detection as in
<PurpleSym1>civodul: I’m trying to figure out how to ensure a given Guix profile in /gnu/store/…-profile matches a given channels.scm and using cached-channel-instance I have to reverse the channel list. That’s how I noticed.
<PurpleSym1>(When sharing projects we cache the Guix profile for speed and need to make sure it’s “genuine”.)
<PurpleSym1>Which is harder than thought, because it takes a full minute to build a profile and then it’s only cached for one user 😕
<civodul>yeah, always per-user
<civodul>if you use time-machine, you get caching for free though
<rekado>I noticed a little problem: “guix time-machine --commit=…” will still get all the channels but only set the ’guix channel to the specified commit
<rekado>(or at least it looked this way)
<rekado>I had to give it a channel file to make it only use the 'guix channel
<rekado>not sure if that’s intended behavior
<civodul>that's intended, but i see it can be confusing
<civodul>by any chance, did one of you fearless scientific hackers try to build gloo and pytorch with rocm?
<zimoun>civodul: No. “Machine learning” is becoming against my ecological principles. :-)
<civodul>heh, same for HPC
<civodul>another thing i dislike about machine learning is when it's used as a substitute for analytical solutions
<PurpleSym1>civodul: re channel order: Should I open an issue?