IRC channel logs


back to list of logs

<rekado>finally pushed a new version of guix-bioc
<rekado>it removed 10 or more very large packages
<rekado>if that’s not enough please let me know and I’ll lower the exclusion threshold
<civodul>rekado: do you know which were the problematic guix-bioc packages?
<civodul>so i can cancel them on guix.bordeaux
<rekado>here’s a list:
<rekado>the threshold now is 1GiB, but we could lower it
<rekado>anything above 1GiB will be rejected
<civodul>i think i managed to cancel the right ones after figuring out package names
<civodul>i’m running GC on the build machines, let’s see how far it goes
<rekado>I noticed that all the bazel packages like tensorflow, jax, etc behave … oddly when grafts are enabled
<rekado>perhaps that’s expected, but I wonder if grafts should have any impact at all on the fixed output derivations made by bazel-vendored-inputs.
<civodul>by definition grafts have no impact on the result of fixed-output derivations, but whether they should be enabled when building them is debatable
<civodul>in some cases, you’d rather have the security fix for some client that’s used when downloading stuff
<civodul>in other cases it’s moot
<civodul>BTW, guix-bioc uses exclusively tarballs, right?
<civodul>would it be a good testbed for using git-fetch for everything?
<civodul>(also because those tarballs aren’t going to be at
<rekado>civodul: oh, that’s a good idea.
<rekado>We could use git-fetch, yes
<rekado>re grafts & bazel: the bazel case is a bit unusual in that many inputs are available in the fixed output derivation when bazel fetches stuff from the internet
<rekado>they only have an impact in that their presence pacifies bazel into not fetching otherwise vendored dependencies
<civodul>rekado: is blank, hpcguix-web must be trying to tell us something
<civodul> is fine
<civodul>prolly due to the lack of a channel description in the config
<rekado>fixed it
<rekado>there was an additional location where I needed to add a guix-bioc description.
<civodul>yeah, metadata is separate
<civodul>good that hpcguix-web hasn’t collapsed with 51k entries :-)
<civodul>we’ll have to find a different strategy at some point because that packages.json file is becoming too big
<civodul>looks like CI workers are still consuming a lot of space
<civodul>not sure if it’s due to past guix-bioc builds that i failed to cancel
<rekado>I can lower the threshold to something like 300 MiB
<civodul>is it plausible that many packages are this big still?
<civodul>note that instead of removing them, we could arrange so CI does not build them
<civodul>i don’t think there’s a way to do that tho :-)
<civodul>unless we write a manifest