pak_solver: Find the ideal set of packages and versions to install

In r-lib/pak: Another Approach to Package Installation

Find the ideal set of packages and versions to install

Description

pak contains a package dependency solver, that makes sure that the package source and version requirements of all packages are satisfied, before starting an installation. For CRAN and BioC packages this is usually automatic, because these repositories are generally in a consistent state. If packages depend on other other package sources, however, this is not the case.

Details

Here is an example of a conflict detected:

> pak::pkg_install(c("r-lib/pkgcache@conflict", "r-lib/cli@message"))
Error: Cannot install packages:
  * Cannot install `r-lib/pkgcache@conflict`.
    - Cannot install dependency r-lib/cli@main
  * Cannot install `r-lib/cli@main`.
- Conflicts r-lib/cli@message

r-lib/pkgcache@conflict depends on the main branch of r-lib/cli, whereas, we explicitly requested the message branch. Since it cannot install both versions into a single library, pak quits.

When pak considers a package for installation, and the package is given with its name only, (e.g. as a dependency of another package), then the package may have any package source. This is necessary, because one R package library may contain only at most one version of a package with a given name.

pak's behavior is best explained via an example. Assume that you are installing a local package (see below), e.g. local::., and the local package depends on pkgA and user/pkgB, the latter being a package from GitHub (see below), and that pkgA also depends on pkgB. Now pak must install pkgB and user/pkgB. In this case pak interprets pkgB as a package from any package source, instead of a standard package, so installing user/pkgB satisfies both requirements.

Note that that cran::pkgB and user/pkgB requirements result a conflict that pak cannot resolve. This is because the first one must be a CRAN package, and the second one must be a GitHub package, and two different packages with the same cannot be installed into an R package library.

r-lib/pak documentation built on Nov. 9, 2024, 2:34 a.m.