pak
pak copied to clipboard
r-lib
Install older version if current CRAN version requests a too recent version of R
DiagrammeR requires R 3.5 as of version 1.0.7: https://github.com/rich-iannone/DiagrammeR/commit/6812dee9ba3b803a20832a0a72e0a74924f0d2fc#diff-9cc358405149db607ff830a16f0b4b21f7366e3c99ec00d52800acebe21b231cR20 . As a consequence, workflows on R 3.4 that try to install DiagrammeR now fail: https://github.com/r-lib/pillar/runs/4892581824?check_suite_focus=true#step:4:1044
Could pak install DiagrammeR 1.0.6 in this situation?
I searched the issues for "older", sorry if I missed a duplicate.
Unfortunately this is not feasible to do automatically, because there is no metadata available for older packages. So we would need to download (potentially a lot of) packages and extract them to see what their dependency requirements are.
OTOH you can solve this manually, by specifying the version you want in extra-packages, either using a URL remote or using the github.com/cran mirror:
- uses: r-lib/actions/setup-r-dependencies@v2
with:
extra-packages: cran/[email protected], ...
This is not R version dependent of course, so if you only want it to apply to R 3.4, then you'll need two r-lib/actions/setup-dependencies steps, one for R 3.4 and one for the other R versions.
Btw. if DiagrammeR is not a hard dependency, you can also ignore it (on R 3.4), like this: https://github.com/r-lib/actions/tree/v2-branch/setup-r-dependencies#ignoring-optional-dependencies-that-need-a-newer-r-version
Thanks. I can work with the ?ignore-before-r= syntax. Would you consider an optional "best effort mode" for suggested dependencies? In this mode the failure would be turned to a message but the installation would succeed (minus DiagrammeR).
In general I don't like the idea of doing this automatically, because you just don't know what exactly is going to be installed. This is basically what install.packages() does, and we didn't have a great experience with it.
Does pak maintain its own metadata cache? Is that cache cached on GHA?
Determining the maximum supported version for a particular CRAN package would mean to download sequentially, but only once. This would greatly increase the experience on older R installations.
No, there is no such metadata stored anywhere.
Well, except for https://crandb.r-pkg.org/ of course. In theory we could use that, but it is far from trivial, because it is too big to just put it in the pak dependency solver.
There isn't anything actionable about this, so I am going to close it now.