Always list base packages in workspace_dependencies()

[DavisVaughan]

I noticed that if you open purrr to https://github.com/tidyverse/purrr/blob/481e829f297fd4315b386518215157f361475ad0/R/adverb-safely.R#L42 and just try to goto definition on force(), then nothing happens.

That's because force() is a base function, but base isn't listed in Imports or Depends so it doesn't get picked up in workspace_dependencies() and we never try and set its package R/ sources.

Similar story if you open an arbitrary workspace of scripts where you use any package from the default R search() path without first calling library() or :: on that package somewhere in the file.

I think the right thing to do is to just ensure that the default search() packages are always listed in workspace_dependencies(). This is definitely right for workspace scripts. For workspace packages technically an individual package file would only need {base}, but if you add any extra scripts into the package, like inst/data.R, then we'd require the rest of the base packages anyways, so I think it is fine to just default to all of the default search() packages for workspace packages too. It doesn't affect diagnostics there, just the set of packages we try and download sources for, so over approximation is pretty harmless.

[lionel-]

For workspace packages technically an individual package file would only need {base}, but if you add any extra scripts into the package, like inst/data.R, then we'd require the rest of the base packages anyways, so I think it is fine to just default to all of the default search() packages for workspace packages too. It doesn't affect diagnostics there, just the set of packages we try and download sources for, so over approximation is pretty harmless.

For scripts in packages, I agree. For files in R/, we should only provide base by default. Is that still the case (or was the case to begin with? I don't remember)?

This will affect diagnostics later on so I would rather get this right now.