[R-package] allow use of custom R executable when building CRAN package

[jameslamb]

There are several CI jobs in this project which rely on non-standard builds of R:

• valgrind:

  LightGBM/.ci/test_r_package_valgrind.sh

  Line 5 in d574fcb

  RDvalgrind CMD INSTALL --preclean --install-tests lightgbm_*.tar.gz || exit -1

• sanitizers:

  LightGBM/.github/workflows/r_package.yml

  Line 197 in d574fcb

  RDscript${{ matrix.r_customization }} -e "install.packages(c('R6', 'data.table', 'jsonlite', 'Matrix', 'testthat'), repos = 'https://cran.r-project.org', Ncpus = parallel::detectCores())"

Those custom builds of R also have their own libraries of installed R packages. Consider the following:

docker run \
  --rm \
  -it \
  wch1/r-debug:latest \
  /bin/bash

# confirm R and RD don't have {data.table}
# result --> Error in library(data.table) : there is no package called ‘data.table’
Rscript -e "library(data.table)"
RDscript -e "library(data.table)"

# install {data.table} for R only
Rscript -e "install.packages('data.table', repos = 'https://cran.r-project.org')"

# check again...{data.table} can be library()'d with R/Rscript, but not RD/RDscript
Rscript -e "library(data.table)"
RDscript -e "library(data.table)"

This project's CI jobs using custom R builds all rely on the script build-cran-package.sh to build the package. That script has its R executable hard-coded to R.

LightGBM/build-cran-package.sh

Line 143 in d574fcb

R CMD build \

Until now, this hasn't caused any issues. Since this project doesn't use vignettes, R CMD build is just lightweight code that moves files around and creates a .tar.gz with those files in it.

However, as of #3946, R CMD build will actually install {lightgbm}, so that it can build vignette outputs (as noted in #4752). Once that happens, it will be problematic to mix builds of R, e.g. to use RDsanscript -e "install.packages(...)" to install dependencies and then R CMD build to build {lightgbm}.

Noticed this when the sanitizer builds failed on #3946 . e.g. https://github.com/microsoft/LightGBM/runs/4053638309?check_suite_focus=true failed with the following error

Error in loadVignetteBuilder(pkgdir, TRUE) :
vignette builder 'knitr' not found
Execution halted

What this PR changes

This PR proposes allowing customization of the R executable used by build-cran-package.sh. It also changes existing CI jobs using custom R builds to take advantage of that feature.

Pushing similar changes to #3946 fixed the issues I mentioned above, so I'm confident this will work.

Why not also allow this customization for CMake-based builds?

I decided not to propose a similar change to build_r.R (the build script for CMake-based builds), just to avoid this PR getting larger and to avoid further complicating that already-complex script. If someone requests such a change to build_r.R I'd support it, but I don't think we should add it until we receive such a request.

[StrikerRUS]

Wow, nicely done! Thank you very much for finding non-invasive solution to alter the R executable in our CI. Just one very minor suggestion.

[github-actions]

This pull request has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.