R/appDependencies.R

In rsconnect: Deploy Docs, Apps, and APIs to 'Posit Connect', 'shinyapps.io', and 'RPubs'

#' Detect application dependencies
#'
#' @description
#' `appDependencies()` recursively detects all R package dependencies for an
#' application by parsing all `.R` and `.Rmd` files and looking for calls
#' to `library()`, `require()`, `requireNamespace()`, `::`, and so on.
#' It then adds implicit dependencies (i.e. an `.Rmd` requires Rmarkdown)
#' and adds all recursive dependencies to create a complete manifest of
#' package packages need to be installed to run the app.
#'
#' # Dependency discovery
#'
#' rsconnect use one of three mechanisms to find which packages your application
#' uses:
#'
#' 1. If `renv.lock` is present, it will use the versions and sources defined in
#'    that file. If you're using the lockfile for some other purpose and
#'    don't want it to affect deployment, add `renv.lock` to `.rscignore`.
#'
#' 2. Otherwise, rsconnect will call `renv::snapshot()` to find all packages
#'    used by your code. If you'd instead prefer to only use the packages
#'    declared in a `DESCRIPTION` file, run
#'    `renv::settings$snapshot.type("explicit")` to activate renv's "explicit"
#'    mode.
#'
#' 3. Dependency resolution using renv is a new feature in rsconnect 1.0.0, and
#'    while we have done our best to test it, it still might fail for your app.
#'    If this happens, please [file an issue](https://github.com/rstudio/rsconnect/issues)
#'    then set `options(rsconnect.packrat = TRUE)` to revert to the old
#'    dependency discovery mechanism.
#'
#' # Remote installation
#'
#' When deployed, the app must first install all of these packages, and
#' rsconnect ensures the versions used on the server will match the versions
#' you used locally. It knows how to install packages from the following
#' sources:
#'
#' * CRAN and BioConductor (`Source: CRAN` or `Source: Bioconductor`). The
#'   remote server will ignore the specific CRAN or Bioconductor mirror that
#'   you use locally, always using the CRAN/BioC mirror that has been configured
#'   on the server.
#'
#' * Other CRAN like and CRAN-like repositories. These packages will have
#'   a `Source` determined by the value of `getOptions("repos")`. For example,
#'   if you've set the following options:
#'
#'   ```R
#'   options(
#'      repos = c(
#'        CRAN = "https://cran.rstudio.com/",
#'        CORPORATE = "https://corporate-packages.development.company.com"
#'      )
#'   )
#'   ```
#'
#'   Then packages installed from your corporate package repository will
#'   have source `CORPORATE`. Posit Connect
#'   [can be configured](https://docs.posit.co/connect/admin/appendix/configuration/#RPackageRepository)
#'   to override their repository url so that (e.g.) you can use different
#'   packages versions on staging and production servers.
#'
#' * Packages installed from GitHub, GitLab, or BitBucket, have `Source`
#'   `github`, `gitlab`, and `bitbucket` respectively. When deployed, the
#'   bundle contains the additional metadata needed to precisely recreated
#'   the installed version.
#'
#' It's not possible to recreate the packages that you have built and installed
#' from a directory on your local computer. This will have `Source: NA` and
#' will cause the deployment to error. To resolve this issue, you'll need to
#' install from one of the known sources described above.
#'
#' # Suggested packages
#'
#' The `Suggests` field is not included when determining recursive dependencies,
#' so it's possible that not every package required to run your application will
#' be detected.
#'
#' For example, ggplot2's `geom_hex()` requires the hexbin package to be
#' installed, but it is only suggested by ggplot2. So if your app uses
#' `geom_hex()` it will fail, reporting that the hexbin package is not
#' installed.
#'
#' You can overcome this problem with (e.g.) `requireNamespace(hexbin)`.
#' This will tell rsconnect that your app needs the hexbin package, without
#' otherwise affecting your code.
#'
#' @inheritParams deployApp
#' @returns A data frame with one row for each dependency (direct, indirect,
#'   and inferred), and 4 columns:
#'
#'   * `Package`: package name.
#'   * `Version`: local version.
#'   * `Source`: a short string describing the source of the package install,
#'      as described above.
#'   * `Repository`: for CRAN and CRAN-like repositories, the URL to the
#'      repository. This will be ignored by the server if it has been configured
#'      with its own repository name -> repository URL mapping.
#' @examples
#' \dontrun{
#'
#' # dependencies for the app in the current working dir
#' appDependencies()
#'
#' # dependencies for an app in another directory
#' appDependencies("~/projects/shiny/app1")
#' }
#' @seealso [rsconnectPackages](Using Packages with rsconnect)
#' @export
appDependencies <- function(appDir = getwd(),
                            appFiles = NULL,
                            appFileManifest = NULL,
                            appMode = NULL) {
  appFiles <- listDeploymentFiles(appDir, appFiles, appFileManifest)
  appMetadata <- appMetadata(appDir, appFiles = appFiles, appMode = appMode)
  if (!needsR(appMetadata)) {
    return(data.frame(
      Package = character(),
      Version = character(),
      Source = character(),
      Repository = character(),
      stringsAsFactors = FALSE
    ))
  }

  bundleDir <- bundleAppDir(
    appDir = appDir,
    appFiles = appFiles,
    appMode = appMetadata$appMode
  )
  defer(unlink(bundleDir, recursive = TRUE))

  extraPackages <- inferRPackageDependencies(appMetadata)
  deps <- computePackageDependencies(bundleDir, extraPackages, quiet = TRUE)
  deps[c("Package", "Version", "Source", "Repository")]
}

needsR <- function(appMetadata) {
  if (appMetadata$appMode %in% c("static", "tensorflow-saved-model")) {
    return(FALSE)
  }

  # All non-Quarto content currently uses R by default.
  # Currently R is only supported by the "knitr" engine, not "jupyter" or
  # "markdown"
  is.null(appMetadata$quartoInfo) ||
    "knitr" %in% appMetadata$quartoInfo[["engines"]]
}

inferRPackageDependencies <- function(appMetadata) {
  deps <- switch(appMetadata$appMode,
    "rmd-static" = c("rmarkdown", if (appMetadata$hasParameters) "shiny"),
    "quarto-static" = "rmarkdown",
    "quarto-shiny" = c("rmarkdown", "shiny"),
    "rmd-shiny" = c("rmarkdown", "shiny"),
    "shiny" = "shiny",
    "api" = "plumber"
  )
  if (appMetadata$documentsHavePython) {
    deps <- c(deps, "reticulate")
  }
  deps
}