Install Python dependencies inside a throwaway Docker container

The Docker-backed twin of repro_install_deps_py(): same return contract, same throwaway-library semantics, but the actual pip install runs INSIDE a container (default networking, since PyPI must be reachable), writing into lib_dir via a bind mount — so the run phase (repro_run_scripts_py_docker()), which mounts the same lib_dir with the network off, finds the packages already there.

Usage

repro_install_deps_py_docker(
  install_deps,
  lib_dir,
  image = getOption("metacheck.docker_image_py", .repro_docker_default_image_py),
  timeout = 600
)

Arguments

install_deps

a data frame as produced inside reproducibility_check() (package, source, ref, base) — the non-base rows of repro_dependencies_py()'s output

lib_dir

throwaway pip install --target directory ON THE HOST (created if absent); bind-mounted into the container at /pylib

image

the Docker image to install into (see .repro_docker_default_image_py)

timeout

timeout in seconds for the whole install container run (default 600)

Value

a data frame with package, source, installed (logical), message (pip's stderr on failure, else ""), via_archive (always FALSE — pip has no CRAN-Archive-style fallback), and category (always NA) — same columns as repro_install_deps_py() so callers do not need to branch on which backend ran.

Examples

if (FALSE) { # \dontrun{
# needs Docker
deps <- repro_dependencies_py("import numpy as np")
repro_install_deps_py_docker(deps, file.path(tempdir(), "pylib"))
} # }