2 R packages

openstatsware Workshop: Good Software Engineering Practice for R Packages

Phil

October 16, 2023

Introduction

What You Know Already

  • Packages provide a mechanism for loading optional code, data, and documentation
  • A library is a directory into which packages are installed
  • install.packages() is used to install packages into the library
  • library() is used to load and attach packages from the library
    • “Attach” means that the package is put in your search list — objects in the package can be used directly
  • Remember that package \(\neq\) library

What We Want to Talk About Now

  • How to write, build, test, and check your own package 😊
  • How to do this in a methodical and sustainable way
  • Give tips and tricks based on practical experience

Contents of a Package

How is a Package Structured?

Package source = directory with files and subdirectories

  • Mandatory:
    • DESCRIPTION
    • NAMESPACE
    • R
    • man
  • Typically also includes:
    • data
    • inst
    • src
    • tests
    • vignettes
    • NEWS

How to Get Started Quickly

Once upon a time, developers would set up this structure manually 🥱

Nowadays, it’s super fast with:

  • usethis::create_package()
  • RStudio > File > New Project > New Directory > R Package

DESCRIPTION File

  • Package: Choose the name of your package
    • Not unimportant!
    • Check CRAN to see if your name is available
  • Title: Add a Title for Your Package (Title Case)
  • Version: Start with a low package version
    • Major.Minor.Patch syntax
  • Authors@R: Add authors and maintainer
  • Description: Like an abstract, including references

DESCRIPTION File (cont’d)

  • License: Important for open sourcing
    • Consider permissive licenses such as Apache and MIT
  • Depends:
    • Which R version users need to have at a minimum
    • Ideally don’t put any package here
    • Packages will be loaded and attached upon library your package
  • Imports: Packages which you import functions, methods, classes from
  • Suggests: Packages for documentation processing (roxygen2), running examples, tests (testthat), vignettes

R Folder

  • Only contains R code files (recommended to use .R suffix)
    • Can create a file with usethis::use_r("filename")
  • Assigns R objects, i.e. mostly functions, but could also be constant variables, data sets, etc.
  • Should not have any side effects, i.e. avoid require(), options() etc.
  • If certain code needs to be sourced first, use on top of file (which will update the Collate field of DESCRIPTION automatically)
#' @include dependency.R
NULL

NAMESPACE File

  • Defines the namespace of the package, to work with R’s namespace management system
  • Namespace directives in this file allow to specify:
    • Which objects are exported to users and other packages
    • Which are imported from other packages

NAMESPACE File (cont’d)

  • Controls the search strategy for variables:
    1. Local (in the function body etc.)
    2. Package namespace
    3. Imports
    4. Base namespace
    5. Normal search() path

man Folder

  • Contains documentation files for the objects in the package in the .Rd format
    • The syntax is a bit similar to LaTeX
  • All user level objects should be documented
  • Internal objects don’t need to be documented — but I recommend it!
  • Once upon a time, developers would set up these .Rd files and the NAMESPACE manually 🥱
  • Fortunately, nowadays we have roxygen2! 🚀

roxygen2 to the Rescue!

  • We can include the documentation source directly in the R script on top of the objects we are documenting
  • Syntax is composed of special comments #' and special macros preceded with @
  • In RStudio, running Build > More > Document will render the .Rd files and the NAMESPACE file for you
  • Get started with usethis::use_roxygen_md()
  • Placing your cursor inside a function in RStudio, create a roxygen2 skeleton with Code > Insert Roxygen Skeleton

Setting up roxygen2 in your project

roxygen2 Source

R/my_sum.R:

#' My Summation Function
#'
#' This is my first function and it sums two numbers.
#'
#' @param x first summand.
#' @param y second summand.
#'
#' @return The sum of `x` and `y`.
#' @export
#' 
#' @note This function is a bit boring but that is ok.
#' @seealso [Arithmetic] for an easier way.
#'
#' @examples
#' my_sum(1, 2)
my_sum <- function(x, y) {
  x + y
}

roxygen2 Output

man/my_sum.Rd:

% Generated by roxygen2: do not edit by hand
% Please edit documentation in R/bla.R
\name{my_sum}
\alias{my_sum}
\title{My Summation Function}
\usage{
my_sum(x, y)
}
\arguments{
\item{x}{first summand.}

\item{y}{second summand.}
}
\value{
The sum of \code{x} and \code{y}.
}
\description{
This is my first function and it sums two numbers.
}
\note{
This function is a bit boring but that is ok.
}
\examples{
my_sum(1, 2)
}
\seealso{
\link{Arithmetic} for an easier way.
}

roxygen2 Output (cont’d)

NAMESPACE:

# Generated by roxygen2: do not edit by hand

export(my_sum)

tests Folder

  • Where store the unit tests covering the functionality of the package
  • Get started with usethis::use_testthat() and usethis::use_test() and populate tests/testthat folder with unit tests
  • Rarely, tests cannot be run within testthat framework, then these can go into R scripts directly in tests directory
  • We will look at unit tests in detail later

data Folder

  • For (example) data that you ship in your package to the user
    • Get started with usethis::use_data()
    • Note: Usually we use lazy data loading, therefore no data() call needed before using the data
  • If you generate the example data, save the R script, too
    • Put that into data-raw folder, start with usethis::use_data_raw()

inst Folder

  • Contents will be copied recursively to installation directory
    • Be careful not to interfere with standard folder names
  • For data that is used by functions in the package itself
    • Those would typically go into inst/extdata folder
    • Load with system.file("path/file", package = "mypackage")
  • CITATION: For custom citation() output
    • Create it with usethis::use_citation()
  • inst/doc can contain documentation files (typically pdf)

src Folder

  • Contains sources and headers for any code that needs compilation
  • Should only contain a single language here
    • Because R uses it, mixing C, C++ and Fortran usually works with OS native compilers
  • Much more complex to write and maintain than an R only package
  • Typically only makes sense for
    • Wrapping existing libraries for use in R
    • Speeding up complex computations — starting point: Rcpp::Rcpp.package.skeleton()

vignettes Folder

  • Special case of documentation files (pdf or html) created by compiling source files
  • Package users don’t need to recompile the vignettes - they are delivered with the package
  • Start a new vignette with usethis::use_vignette()
    • Starts an Rmd vignette, compiled with knitr
  • Important for the user to understand the high-level ideas
  • Complements function-level documentation from our roxygen2 chunks

NEWS File

  • Lists user-visible changes worth mentioning
  • In each new release, add items at the top under the version they refer to
  • Don’t discard old items: leave them in the file after the newer items
  • Start one with usethis::use_news_md()

Building the Package

Documenting the Package

  • The first step is to produce the documentation files and NAMESPACE
  • In RStudio: Build > More > Document
  • In the console: devtools::document()

Checking the Package

  • R comes with pre-defined check command for packages: “the R package checker” aka R CMD check
  • About 22 checks are run (so quite a lot), including things like:
    • Can the package be installed?
    • Is the code syntax ok?
    • Is the documentation complete?
    • Do tests run successfully?
    • Do examples run successfully?
  • In RStudio: Build > Check
  • In the console: devtools::check()

Building the Package

  • The R package folder can be compressed into a single package file
  • Typically we manually only build “source” package
    • In RStudio: Build > More > Build Source Package
    • In the console: devtools::build()
  • Makes it easy to share the package with others and submit to CRAN

Installing the Package

  • R comes with pre-defined install command for packages: R CMD INSTALL
  • In RStudio: Build > Install
  • In the console: devtools::install()
  • Note: During development it’s usually sufficient to use Build > More > Load All
    • Runs devtools::load_all()
    • Roughly simulates what happens when package would be installed and loaded
    • Unexported objects and helpers under tests will also be available
    • Key: much faster!

Exercise

Let’s try this out now 😊

  1. Set up a new R package with a fancy name
  2. Fill out the DESCRIPTION file
  3. Include a new function
  4. Add roxygen2 documentation
  5. Export the function to the namespace
  6. Produce the package documentation
  7. Run checks
  8. Build the package

References

License Information

Good Software Engineering Practice for R Packages - Home | Slides | License