Related to #1944.

Proposes switching from {rmarkdown} to {markdown} for the R package's vignettes.

Benefits of this Change

Reduces the set of dependencies that have to build successfully to run R CMD check --as-cran against {lightgbm}, leading to:

• faster CI builds
• lower risk of CI failures
• lower risk of failed builds and disruptions on CRAN
• slightly reduced reverse-dependency burden for CRAN and maintainers of {rmarkdown}

CRAN_DB <- tools::CRAN_package_db()

.get_strong_deps <- function(packages) {
    ret <- tools::package_dependencies(
        packages = packages
        , db = CRAN_DB
        , which = c("Depends", "Imports", "LinkingTo")
        , recursive = TRUE
    )
    ret_vec <- c(unlist(ret), packages)
    return(sort(unique(ret_vec)))
}

# combination of the following:
#   * LightGBM's strong dependencies (recursive)
#   * LightGBM's "Suggests" dependencies
#   * strong dependencies of those "Suggests" dependencies (recursive)
#
# excluding {rmarkdown} / {markdown}
lgb_r_cmd_check_deps <- setdiff(
    .get_strong_deps(c("lightgbm", "knitr", "RhpcBLASctl", "testthat"))
    , "lightgbm"
)
length(lgb_r_cmd_check_deps)
# 47

strong_deps_for_rmarkdown <- .get_strong_deps("rmarkdown")
strong_deps_for_markdown <- .get_strong_deps("markdown")

# additional packages added by {rmarkdown}
setdiff(strong_deps_for_rmarkdown, lgb_r_cmd_check_deps)

# [1] "base64enc"   "bslib"       "cachem"      "ellipsis"   
# [5] "fastmap"     "fontawesome" "htmltools"   "jquerylib"  
# [9] "memoise"     "mime"        "rappdirs"    "rmarkdown"  
# [13] "sass"        "stringi"     "stringr"     "tinytex"

# additional packages added by {markdown}
setdiff(strong_deps_for_markdown, lgb_r_cmd_check_deps)
# [1] "commonmark" "markdown"

But what are we giving up?

My understanding is the main benefits of {rmarkdown} over {markdown} are things that are not relevant to the simple documentation {lightgbm} produces. Things like:

• support for many more outputs formats, such as Microsoft Word documents, interactive notebooks, and 5+ formats of slide show (see https://rmarkdown.rstudio.com/lesson-9.html)
• interactivity within the RStudio IDE (see https://rmarkdown.rstudio.com/lesson-14.html)

{markdown}, as stated in its README (https://github.com/rstudio/markdown), aims to be simpler and lighter-weight:

It is a deliberate design choice to keep this markdown package lightweight...
... this package is intended for minimalists

How I tested this

Enabled this branch on readthedocs and built the docs there.

The build took 807 seconds (build link), roughly the same as other builds took over the last 7 days (build history).

Confirmed that the rendered vignette looks correct at the documentation site: https://lightgbm.readthedocs.io/en/docs-lighter-weight-vignettes/R/articles/basic_walkthrough.html.

Many of this project's R CI jobs build the vignettes and test them with R CMD check --as-cran, so I'm confident this change wouldn't cause issues for CRAN.

Confirmed all references to {rmarkdown} have been replaced by {markdown}:

git grep -i rmarkdown

Notes for Reviewers

This PR inspired by @jangorecki's post on the r-pkg-devel mailing list: https://stat.ethz.ch/pipermail/r-package-devel/2024q1/010280.html. {data.table} recently made a similar change in pursuit of reducing the weight of their dependencies.

Related:

• Simple markdown::knit() function rstudio/markdown#109
• trim suggested dependencies Rdatatable/data.table#5745
• vignette render with markdown rather than rmarkdown Rdatatable/data.table#5773