Use RSPM with (macOS) fallback to CRAN

[salim-b]

RStudio Public Package Manager (RSPM) is an awesome service to the community, no doubt. It's main drawback compared to CRAN: No binaries for macOS.

renv seems to have RSPM integration to some degree, which is great. But I can't figure out a sane way to use the RSPM CRAN mirror URL (https://packagemanager.rstudio.com/cran/latest) for Linux and Windows, and at the same time the regular CRAN URL (https://cloud.r-project.org) for macOS (or the RSPM mirror only for Linux and regular CRAN also for Windows, doesn't really matter I guess).

I'm using the explicit snapshot type for my project (with a DESCRIPTION file listing immediate dependencies) and tried various different configurations to achieve the above goal. The nearest I could get is

1. setting regular CRAN, i.e.

   "R": {
     "Repositories": [
       {
         "Name": "CRAN",
         "URL": "https://cloud.r-project.org"
       }
     ]
   }

   in the renv.lock file;

2. conditionally overwriting the repos option in the project's .Rprofile after activating renv when on Linux, i.e.

   source("renv/activate.R")
   
   if (Sys.info()["sysname"] == "Linux") {
     options(repos =
               getOption("repos") |>
               magrittr::inset("CRAN", "https://packagemanager.rstudio.com/cran/latest") |>
               renv:::renv_rspm_transform())
   }

---

With the above config on Linux (Ubuntu 22.04 in my case), the weird thing is that when I e.g. do

renv::purge("sp", prompt = FALSE)

Output

The following packages will be purged from the cache:

	sp 1.5-0 [Hash: 2a118bdd2db0a301711e0a9e3e3206d5]

* Removed 1 package from the cache.

and then do

renv::repair()

Output

The following package(s) have broken symlinks into the cache:

	sp

Use `renv::repair()` to try and reinstall these packages.

* Querying repositories for available source packages ... Done!
Retrieving 'https://packagemanager.rstudio.com/cran/__linux__/jammy/latest/src/contrib/sp_1.5-0.tar.gz' ...
	OK [file is up to date]
Installing sp [1.5-0] ...
	OK [installed binary]
Moving sp [1.5-0] into the cache ...
	OK [moved to cache in 0.31 milliseconds]

the package sp is fetched from RSPM as can be seen in the output above (fine!).

But when I instead run

renv::restore(clean = TRUE,
              prompt = FALSE)

Output

The following package(s) have broken symlinks into the cache:

	sp

Use `renv::repair()` to try and reinstall these packages.

The following package(s) will be updated:

# CRAN ===============================
- sp   [* -> 1.5-0]

* Querying repositories for available source packages ... Done!
Retrieving 'https://cloud.r-project.org/src/contrib/sp_1.5-0.tar.gz' ...
	OK [downloaded 1009.7 Kb in 0.2 secs]
Installing sp [1.5-0] ...
	OK [built from source]
Moving sp [1.5-0] into the cache ...
	OK [moved to cache in 0.34 milliseconds]

after the purge, the package sp is still fetched from regular CRAN, as can again be seen in the output above (this sucks...).

---

The entry for package sp in the project's renv.lock file looks as follows:

"sp": {
  "Package": "sp",
  "Version": "1.5-0",
  "Source": "Repository",
  "Repository": "CRAN",
  "Hash": "2a118bdd2db0a301711e0a9e3e3206d5",
  "Requirements": [
    "lattice"
  ]
}

---

Is the differing behavior between renv::repair() and renv::restore() above the intended behaviour? If so, why? And would it still be possible to fetch packages from RSPM on some platforms and from CRAN on others within a single renv project?

[kevinushey]

The main difference between repair() and restore() is that repair() uses install(), and there are some subtle differences in how repositories are handled. In particular:

• install() uses, in order of precedence, [repos argument > RENV_CONFIG_REPOS_OVERRIDE > getOption("repos")]
• restore() uses, in order of precedence, [repos argument > RENV_CONFIG_REPOS_OVERRIDE > The repositories defined in the lockfile > getOption("repos")].

In other words, install() doesn't consider the repositories defined in the lockfile, while restore() does. So I think the problem here is that your attempt to change the repos R option is effectively ignored by renv here, and so restore() still uses the lockfile repositories.

As a workaround, I believe you could set your repositories as e.g.

# set up the default repos for your platform
options(repos = < ... >)

# tell renv to use that during restore by default
options(renv.config.repos.override = getOption("repos"))

since renv doesn't have the notion of platform-specific repositories (or other platform-specific lockfile entries).

The main reason why we consider the lockfile repositories directly in restore() is that some users might want to call renv::restore() without explicitly activating a project, and so in that case the user intention is likely to use the repositories as defined in the lockfile.

[salim-b]

Thank you very much for the clear explanation @kevinushey! Helped a lot to better understand what's going on under the hood. And I can properly work around the issue now 😄

since renv doesn't have the notion of platform-specific repositories (or other platform-specific lockfile entries).

Do you plan on supporting such a notion? Most straight-forward, I guess, would be to add support for an optional platforms key for R.Repositories in renv.lock. So we could configure repos for example as:

"R": {
  "Repositories": [
    {
      "Name": "CRAN",
      "URL": "https://packagemanager.rstudio.com/cran/latest",
      "platforms": ["Windows", "Linux"]
    },
    {
      "Name": "CRAN",
      "URL": "https://cloud.r-project.org",
      "platforms": "Darwin"
    }
  ]
}

[angerhang]

Thank you very much for the clear explanation @kevinushey! Helped a lot to better understand what's going on under the hood. And I can properly work around the issue now 😄

since renv doesn't have the notion of platform-specific repositories (or other platform-specific lockfile entries).

Do you plan on supporting such a notion? Most straight-forward, I guess, would be to add support for an optional platforms key for R.Repositories in renv.lock. So we could configure repos for example as:

"R": {
  "Repositories": [
    {
      "Name": "CRAN",
      "URL": "https://packagemanager.rstudio.com/cran/latest",
      "platforms": ["Windows", "Linux"]
    },
    {
      "Name": "CRAN",
      "URL": "https://cloud.r-project.org",
      "platforms": "Darwin"
    }
  ]
}

Something like this will be very helpful

[hadley]

I think this is basically a duplicate of #1052