[R-pkg-devel] R scripts evaluated in a package

Dear R-devel,

I recently volunteered to help maintain an orphaned package that supplies some shiny widgets. When looking at the reverse dependencies I see that quite a few packages supply a shiny app via R scripts in the inst-directory. In the /R directory there is a script with a function like:

myapp <- function()
{
        run_app_R_script <- '
    appDir <- system.file("shinyapp", package = "apackage")

    if (appDir == "") {
      stop("Could not find the shinyapp directory. Try re-installing
        the `apackage` package.", call. = FALSE
      )
    }

    shiny::runApp(appDir, launch.browser = TRUE, display.mode = "normal")'

  eval(parse(text = run_app_R_script))
}

This is so common it seems to be a design pattern, but it is absolutely not necessary. Shiny applications can be run just as easily from functions in the R directory of a package and do not need external scripts or directories.

The app scripts in the inst/shinyapp directory typically contain a number of `library` statements. The thus loaded packages may or may not be listed in the imports or suggests. Sometimes it seems like a means to get around the allowed maximum number of packages to import note. One can easily imagine more serious violations (I have not seen any).

A problem that results is that potential breaking changes in the package I help maintain, e.g. removing a function, do not show up in the revdepcheck, since R files that live outside the R directory are not included in Rcmd check. Yet such changes will leave `apackage` completely broken (often these packages contain only one function).

I cannot find anything in WRE directly forbidding this practice but it seems to at least go against the intent of WRE.

My question: is this allowed? If not, could this be added to Rcmd check or should it be reported in some way?

Kind regards,
Jesse







[https://mailing.cito.nl/handtekeningen/citolab.png]