Keeping R packages needed for the extension in separate library

[gaborcsardi]

I wonder if there is a good way to keep the R packages needed for the extension in a separate package library. I.e. {languageserver}, {sess} and (maybe) {processx} and their dependencies.

The motivation is that I am experimenting with implementing true virtual environments for R (as in Python and uv). In an R project with a virtual environment all packages (beyond base and recommended) are stored in ~/.rvenv/lib. The library path only has the project library and the system library:

> .libPaths()
[1] "/private/tmp/test/.rvenv/lib"                       
[2] "/Users/gaborcsardi/.local/share/rig/r/4.6.1/library"

This does not work great when the extension tries to install {languageserver} and {sess}. I can set an extra library in libPaths, and install {languageserver} there manually.

But {sess} must be loaded into the interactive R sessions, so currently that and its dependencies (processx, ps, R6, later, Rcpp, rlang, jsonlite, rstudioapi) have to live in a library that is on the library path.

I wonder if anyone has ideas to improve this.

Maybe vscode-R could install all these packages into a separate library, and put that library on the library path as well, before the system library? Would you be willing to have a feature like this? It could be a single check box called "Keep vscode-R R packages in separate library". If this checkbox is set then 1) when {languageserver} or {sess} are missing, vscode-R would install them into an R version and platform dependent library, kept in the machine cache directory, and 2) would put this library on the .libPaths(), before the system library.

Of course I am also open to other suggestions.

[eitsupi]

Putting sess aside for the moment, languageserver itself depends on quite a few R packages. I think this makes dependency conflicts fairly likely when vscode-R is used together with an isolated project environment such as renv.

So I expect we will eventually need something along the followings:

1. Development/tooling dependencies such as languageserver should try to minimize their R package dependencies as much as reasonably possible. See jgd and v3.0.0 release status (roadmap) #1727 (comment) for related discussion.
2. These development/tooling dependencies should be declared in a manifest file, similar to Cargo.toml or pixi.toml, and installed into a dedicated library managed as a single environment.

[gaborcsardi]

I am doing this now in rig: https://rig.r-lib.org/reference/proj.html#the-tools-package-library

I.e. there is a separate user level library called tools, its packages are not managed by the project, that is automatically added to the library path in projects. Then I can manually install pkgload, roxygen2, languageserver, etc. there and all projects pick it up.

I haven't used this for long, but it seems to work pretty well. Of course I still have a problem with 'sess' because I cannot install it myself, I cannot tell vscode-R to install it into the tools library, and I also cannot add it to the manifest.