IRC channel logs


back to list of logs

<drakonis>hey there
<civodul>interesting slide deck on R & reproducibility:
<civodul>rekado, zimoun: ↑
<civodul>it does mention Guix, just to say it's "overkill for reproducible research" :-)
<civodul>like, "it must be useful, but only for reproducibility professionals"?
<zimoun>civodul: considering only R packages, indeed Guix helps for more reproducibility but it is not a game changer because the R community is already doing a huge work of curation and maintainance.
<zimoun>although it is hard to re-set the same R environment as in 2020; when it is trivial with guix time-machine. ;-)
<zimoun>Have a nice week-end
<civodul>but i guess R doesn't have a tool that takes a list of package/versions as input and reinstantiates it?
<drakonis>require, maybe?
<drakonis>there's no straightforward way to do versioning
<drakonis>using devtools is the closest way to achieving it
<rekado>civodul: heh, that’s cute.
<rekado>civodul: the R people that I know use packrat:
<rekado>or at least they point to it when asked about how they would reproduce their R environment.
<civodul>oh so you can give it some sort of a manifest?
<rekado>(pointing to something is much more common than actual use, no matter what aspect of reproducibility)
<rekado>you can take a snapshot of your R environment, bundle it up for sharing, and then later unbundle it again.
<civodul>sounds pretty cool
<rekado>but this fails when you use things that aren’t 100% R
<civodul>how frequent is that, though?
<civodul>R seems to be well isolated from the rest of the world
<rekado>there are lots of packages near the root of the graph that are bindings to C libraries
<rekado>I maintain a little VM-on-demand thing that lets you create a Guix environment, spawn an AWS EC2 VM with RStudio, where the environment can be used.
<rekado>I need to allow users to use install.packages or devtools
<rekado>so the environment needs to have all sorts of “common” libraries and toolchains
<rekado>things like cairo, curl, gcc-toolchain, etc
<rekado>or else you just can’t build any of the popular R packages
<rekado>gzip, cairo, libxt, openssl, curl, zlib, etc
<rekado>that said, packrat can be “good enough”
<rekado>certainly better than just sessionInfo()
<civodul>i see
<civodul>yeah folks take those libraries for granted
<civodul>as if it were hardware
<civodul>maybe that's often "good enough"
<civodul>well, gcc-toolchain is more of a problem than say zlib