Install a paper's declared dependencies into a throwaway library

Installs each non-base dependency into lib_dir (a temp library, never the user's), so a run does not mutate the user's installed packages. CRAN packages come from install.packages(); Bioconductor packages (source == "bioc", see repro_dependencies()/.repro_bioc_packages()) from BiocManager::install(), installing BiocManager itself first if absent; GitHub/URL sources from the ref the code named (remotes::install_github() / install.packages(url)), honouring a pinned @ref when present. Every install's outcome is recorded; installing runs package build/configure scripts, so this is part of the gated execute phase.

Usage

repro_install_deps(
  install_deps,
  lib_dir,
  cran_to_main_lib = FALSE,
  deadline = NULL
)

Arguments

install_deps

the module's install_deps frame (package, source, ref, and optionally version, from .repro_recorded_plan()), base packages already excluded

lib_dir

the throwaway library path (created if absent); ignored for CRAN-source packages when cran_to_main_lib = TRUE (see below)

cran_to_main_lib

if TRUE, CRAN-source packages are installed into your DEFAULT library (no lib argument — same as calling install.packages(pkg) yourself) instead of the throwaway lib_dir, and so PERSIST after the run rather than being deleted with the sandbox. This trades a one-time, permanent change to your R library for install work that is shared across every later call needing the same package (useful when reproducibility_check() runs across many papers in one session/script: a CRAN package installed for an early paper is already present, and skipped as already-installed, for every later one). GitHub/URL sources are deliberately excluded from this and always go into the throwaway lib_dir regardless of this argument — those are arbitrary, unreviewed code you likely do not want permanently in your main library. Default FALSE (everything throwaway, the original behaviour).

A CRAN-source package that install.packages() cannot find is not necessarily a broken dependency: CRAN removes (archives) packages that stop passing its checks, but every version ever published stays available as a source tarball under the CRAN Archive (https://cran.r-project.org/src/contrib/Archive/<pkg>/). When the live-CRAN install fails, this retries once from the Archive's most recent version before giving up — a package the paper's authors could install at the time (and that later dropped off live CRAN) still installs, instead of the run recording an "errored" script for a cause that has nothing to do with the paper's code. The retry is recorded in via_archive so the report can say the run used a stale-but-real version rather than the one the authors had.

deadline

the time (Sys.time() plus the time for installing) after which no further package is started; NULL (the default) sets no limit

Value

a data frame with package, source, installed (logical), message (error text on failure, else ""), via_archive (logical, TRUE when installed from the CRAN Archive after a live-CRAN failure), category (see .repro_classify_install_message(); NA for a successful install), version_recorded (the version asked for, or NA), version_installed (the version now in the libraries, or NA) and version_note (why the recorded version was not installed, else "").

Details

A row with a version is installed at that version with remotes::install_version(), unless the libraries already hold it; packages it depends on that are not there yet get their current version. When that version cannot be installed, the current version is installed instead and version_note says why. Such a row always goes into the throwaway lib_dir, so an old version can never replace one in your own library. A Bioconductor package always gets its current version, since Bioconductor versions follow the R version.

Examples

if (FALSE) { # \dontrun{
# installs packages into a throwaway library
deps <- repro_dependencies("library(pwr)")
repro_install_deps(deps, file.path(tempdir(), "lib"))
} # }