In lorenzwalthert/precommit: Pre-Commit Hooks

knitr::opts_chunk$set(
  collapse = TRUE,
  comment = "#>"
)

pre-commit >= 2.11.0 supports R as a language and each hook repo has its own virtual environment with package versions locked. For that reason, we should also test with exactly these versions.

This package has five testing workflows:

hook testing. Tests the hooks (as well as hook script helpers?) in the renv in which they'll be shipped. No R CMD Check or pre-commit executable needed. All platforms. To avoid convolution of the testing environment (that contains testthat and other packages) and the hook environment (specified in renv.lock), we must only activate the hook environment right when the script is called, but run_test() must run in the testing environment. Since --vanilla is inherited in the child process initiated from run_test(), the only way to do this is to set an env variable when running run_test() and check in the user R profile if it is set, and then activate the renv. This is done with R_PRECOMMIT_HOOK_ENV.
complete testing: Tests hooks as well as usethis like access functionality with latest CRAN packages, on all platforms, with all installation methods.
CRAN testing. On CRAN, complete testing is ran, tests that check pre-commit executable access are disabled.
end-2-end testing: This simulates closely what happens when you run git commit with pre-commit activated: We update the .pre-commit-config.yaml to the pushed SHA and run pre-commit run, covering all hooks, to ensure in particular file linking, file permissions etc. in .pre-commit-hooks.yaml are correct for exported hooks.