IRC channel logs

2021-04-26.log

back to list of logs

<zimoun`>rekado_: cool!
<rekado_>there seems to be a regression in nyacc’s ffi helper, so guile-drmaa fails to record all #define’d symbols
<rekado_>I’m using this to run slurm: https://elephly.net/paste/slurm.sh.html
<rekado>it is concerning that merely computing what needs to be built takes a very long time over inferiors.
<rekado>this significantly delays the appearance of the build plan, so it just sits there for half a minute or longer, silently busy.
<rekado>so far I had only focused on performance when *nothing* needs to be done, but it’s more problematic when something does need to be done and no build plan appears.
<efraim>how much of it is compiling the past guix checkout?
<efraim>I image that would take a while
<rekado>none of it
<rekado>I open an inferior to the “invoking Guix”, i.e. the one that was used to run “guix workflow”
<rekado>looks like GWL submitting jobs to a cluster via DRMAA works as it should. Very straight-forward.
<civodul>rekado: thing is inferior-eval-with-store throws aways its cache every time
<civodul>so the inferior ends up recompiling intermediate derivations for every package
<rekado>civodul: yeah, I tried to let inferior-eval-with-store keep some state, but wasn’t immediately successful
<civodul>you could try saving/restoring the object cache
<civodul>on master there are additional caches, like (@@ (guix packages) %derivation-cache)
<civodul>that one should actually persist across inferior-eval-with-store
<zimoun>hi!
<efraim>hi!
<civodul>o/
<ndeck>hi everyone !
<civodul>hello ndeck!
<zimoun>Have someone used the API token provided by SWH?
<civodul>there's a token mechanism to avoid the rate limit?
*civodul didn't know
<zimoun>Well, it can help using a specific account; instead of based on IPs.
<zimoun>zack’s answer. ;-)
<civodul>sure, that's what GitHub & co. are doing (we have that for 'guix refresh -t github')
<civodul>how does one get a token?
<zimoun>create an account <https://archive.softwareheritage.org/oidc/login/>
<zimoun>thanks for the Github info, I will give a look.
<efraim>is it a problem if we embed it in our sources? Like other distros did with google API keys and chromium?
<civodul>i think it'd be a problem
<civodul>the account is meant to be per-person i guess
<civodul>otherwise that defeats the whole thing
<rekado>when starting rstudio from Guix I see locale warnings and .libPaths() does not include packages from my Guix profile.
<rekado>it looks like rserver purges the needed environment variables
<PurpleSym>rekado: Yes, it does start with a fresh environment. I have a patch here: https://github.com/leibniz-psychology/guix-zpid/blob/master/rstudio-server-1.4-propagate-env.patch
<rekado>I see.
<PurpleSym>(To be specific, this applies to the spawned rsession.)
<rekado>I wonder if there’s a built-in mechanism to affect the environment.
<rekado>does it read any other site files? Or user files, or something like that…?
<rekado>I’d prefer not to let the full environment leak into rsession processes
<PurpleSym>Afaik there’s a rstudio-specific R site file.
<PurpleSym>But I never tried to use it, since each user starts his own rserver anyway for us and thus environment leaking is actually required.