IRC channel logs

<rekado>a manifest sounds easy enough to do. We have a predicate for bioconductor packages to select all relevant packages in the manifest, and then remove those that we don’t want to build (and whatever package requests them).

<rekado>I’ve been trying to build jaxlib with CUDA support for guix-science-nonfree, but I have to move that work to our HPC cluster as GCC keeps getting killed after running for hours on my laptop.

<rekado>would be nice if we could offload builds to an HPC cluster with a little adapter

<civodul>hey

<civodul>rekado: i often offload to the build machines we have at work actually :-)

<civodul>GUIX_DAEMON_SOCKET=ssh://… guix build foo

<civodul>that’s very ad-hoc though

<civodul>we should use something like a semi-public Build Coordinator instance

<civodul>or a daemon exposed over websockets

<civodul>guix-bioc is at 93%! https://hpc.guix.info/channels

<rekado>guix-cran on the other hand hasn’t received an update for a few days. The importer sometimes doesn’t do the right thing and generates packages that still depend on removed variables.

<rekado>not sure why.

<rekado>oh, never mind: 5 hours ago

<rekado>guess it just takes a long time :)

<civodul>you mean the auto-update script ran 5h ago?

<rekado>it finished running 5h ago, yes

<rekado>yesterday night I checked and it hadn’t have a successful run in 4 days

<civodul>why is that?

<rekado>I noticed that sometimes fetched DESCRIPTION files are empty (or contain #F)

<rekado>these packages aren’t removed, but they don’t result in package definitions.

<rekado>so sometimes we end up with missing packages that other packages still depend on

<rekado>and so the “guix pull” check fails at the end, and the script refuses to update the channel (which is correct).

<civodul>bah

<civodul>it’s your lucky day: we’ve had a bug report on the French-speaking Mattermost channel, with Bazel failing to build

<civodul>so i’ve invited them to report it at https://github.com/guix-science/guix-science :-)

<civodul>but CI says it builds fine, so dunno

<civodul>(speaking of which, the whole instant messaging thing is getting out of hands)

<rekado>there’s a Mattermost channel … about Guix?

<rekado>bazel builds “fine” for me (as fine as one could reasonably consider the build)

<rekado>I’ve been using it for jaxlib, jax, tensorflow, etc

<rekado>here’s the bug report: https://github.com/guix-science/guix-science/issues/32\

<civodul>i was referring to the “Café Guix” Mattermost channel: https://hpc.guix.info/events/2024/caf%C3%A9-guix/

<civodul>and yes, bazel builds fine on guix.bordeaux AFAICS so I don’t know

<civodul>a channel of ours broke due to changes in guix-science{,-nonfree}: https://guix.bordeaux.inria.fr/eval/7434396

<civodul>ooh, i think that’s because that channel does not explicitly declare a dependency on guix-science

<civodul>it’s only pulling it transitively

<civodul>ah, it’s more complicated

<civodul>rekado: why does guix-science-nonfree depend on a specific commit of guix-science?

<civodul>copy/paste error?

<civodul>ah no, i’m saying nonsense

<rekado>hmm?

<rekado>did I break something?

<rekado>I have been working on (guix-science packages python) until four days ago

<civodul>you didn’t

<civodul>i think we have uncovered a channel bug

<civodul> https://issues.guix.gnu.org/68797

<civodul>fortunately it’s easy to work around it

<rekado>huh, I guess we’ve always been lucky

<rekado>we have been using way too many channels for a long time, but people generally only used an explicit list of all channels that I had prepared.

IRC channel logs

2024-01-29.log