3 An R Package Engineering Workflow

Good Software Engineering Practice for R Packages

Zhenglin Ruan

July 31, 2025

Motivation

From an idea to a production-grade R package

Example scenario: in your daily work, you notice that you need certain one-off scripts again and again.

The idea of creating an R package was born because you understood that “copy and paste” R scripts is inefficient and on top of that, you want to share your helpful R functions with colleagues and the world…

Professional Workflow

Photo CC0 by ELEVATE on pexels.com

Typical work steps

  1. Idea
  2. Concept creation
  3. Validation planning
  4. Specification:
    1. User Requirements Spec (URS),
    2. Functional Spec (FS), and
    3. Software Design Spec (SDS)
  1. R package programming
  2. Documented verification
  3. Completion of formal validation
  4. R package release
  5. Use in production
  6. Maintenance

Workflow in Practice

Photo CC0 by Chevanon Photography on pexels.com

Frequently Used Workflow in Practice

  1. Idea
  2. R package programming
  3. Use in production
  4. Bug fixing
  5. Use in production
  1. Bug fixing + Documentation
  2. Use in production
  3. Bug fixing + Further development
  4. Use in production
  5. Bug fixing + …

Bad practice!

Why?

Why practice good engineering?

Cost distribution among software process activities

doi:10.14569/IJACSA.2020.0110375

Why practice good engineering?

Origin of errors in system development

Boehm, B. (1981). Software Engineering Economics. Prentice Hall.

Why practice good engineering?

  • Don’t waste time on maintenance
  • Be faster with release on CRAN
  • Don’t waste time with inefficient and buggy further development
  • Fulfill regulatory requirements1
  • Save refactoring time when the PoC becomes the release version
  • You don’t have to be shy any longer about inviting other developers to contribute to the package on GitHub

Why practice good engineering?

Invest time in

  • requirements analysis,
  • software design, and
  • architecture…

… but in many cases the workflow must be workable for a single developer or a small team.

Workable Workflow

Photo CC0 by Kateryna Babaieva on pexels.com

Suggestion for a Workable Workflow

  1. Idea
  2. Design docs
  3. R package programming
  4. Quality check (see Ensuring Quality by Chunyan)
  5. Use in production

Example - Step 1: Idea

Let’s assume that you used some lines of code to create simulated data in multiple projects:

dat <- data.frame(
    group = c(rep(1, 50), rep(2, 50)),
    values = c(
        rnorm(n = 50, mean = 8, sd = 12),
        rnorm(n = 50, mean = 14, sd = 11)
    )
)

Idea: put the code into a package

Example - Step 2: Design docs

  1. Describe the purpose and scope of the package
  2. Analyse and describe the requirements in clear and simple terms (“prose”)
Obligation level Key word1 Description
Duty shall “must have”
Desire should “nice to have”
Intention will “optional”

Example - Step 2: Design docs

Purpose and Scope

The R package simulatr shall enable the creation of reproducible fake data.

Package Requirements

simulatr shall provide a function to generate normal distributed random data for two independent groups. The function shall allow flexible definition of sample size per group, mean per group, standard deviation per group. The reproducibility of the simulated data shall be ensured via an optional seed It should be possible to print the function result. A graphical presentation of the simulated data will also be possible.

Example - Step 2: Design docs

Useful formats / tools for design docs:

UML Diagram

Example - Step 3: Packaging

R package programming

  1. Create basic package project (see R Packages by Shuang)
  2. C&P existing R scripts (one-off scripts, prototype functions) and refactor1 it if necessary
  3. Create R generic functions
  4. Document all functions

Example - Step 3: Packaging

One-off script as starting point:

sim.data <- function(n1, n2, m1, m2, s1, s2) {
    data.frame(
        group = c(rep(1, n1), rep(2, n2)),
        values = c(
            rnorm(n = n1, mean = m1, sd = s1),
            rnorm(n = n2, mean = m2, sd = s2)
        )
    )
}

Example - Step 3: Packaging

Refactored script:

getSimulatedTwoArmMeans <- function(n1, n2, mean1, mean2, sd1, sd2) {
    data.frame(
        group = c(rep(1, n1), rep(2, n2)),
        values = c(
            rnorm(n = n1, mean = mean1, sd = sd1),
            rnorm(n = n2, mean = mean2, sd = sd2)
        )
    )
}

Almost all functions, arguments, and objects should be self-explanatory due to their names.

Example - Step 3: Packaging

Define that the result is a list1 which is defined as class2:

getSimulatedTwoArmMeans <- function(n1, n2, mean1, mean2, sd1, sd2) {
    result <- list(n1 = n1, n2 = n2, 
         mean1 = mean1, mean2 = mean2, sd1 = sd1, sd2 = sd2)
    result$data <- data.frame(
        group = c(rep(1, n1), rep(2, n2)),
        values = c(
            rnorm(n = n1, mean = mean1, sd = sd1),
            rnorm(n = n2, mean = mean2, sd = sd2)
        )
    )
    # set the class attribute
    result <- structure(result, class = "SimulationResult")
    return(result)
}

Example - Step 3: Packaging

The output is impractical, e.g., we need to scroll down:

x <- getSimulatedTwoArmMeans(n1 = 50, n2 = 50, mean1 = 5, mean2 = 7, sd1 = 3, sd2 = 4)
x
$n1
[1] 50

$n2
[1] 50

$mean1
[1] 5

$mean2
[1] 7

$sd1
[1] 3

$sd2
[1] 4

$data
    group     values
1       1  7.9913524
2       1  2.4433099
3       1  4.8248363
4       1  4.3003821
5       1  5.1070738
6       1  5.6821700
7       1  6.9767241
8       1  3.5878576
9       1  7.3006997
10      1  7.3315869
11      1  0.3552069
12      1  5.7443710
13      1  4.5599642
14      1  6.6151279
15      1  7.8710057
16      1 -0.6704843
17      1  5.0435107
18      1  4.3993856
19      1  1.3970932
20      1  7.1525637
21      1  8.0186925
22      1  5.6764564
23      1  3.3590526
24      1  5.8097090
25      1  6.0043209
26      1  2.9758671
27      1  9.1309671
28      1  6.1951981
29      1  9.5007511
30      1  5.3578726
31      1  3.5172697
32      1  5.6763330
33      1  8.7859767
34      1  5.1906407
35      1  6.6614119
36      1  5.5707242
37      1  2.1098313
38      1  3.4000577
39      1  5.8239890
40      1  6.6442456
41      1  7.0296165
42      1  9.8081476
43      1  6.5420327
44      1  4.0359417
45      1  8.1592040
46      1  4.2568957
47      1  5.9482276
48      1  9.3280873
49      1  2.8239449
50      1  8.4520435
51      2  6.2519630
52      2 10.6679844
53      2  6.5189243
54      2  4.3832382
55      2  9.5586722
56      2  6.1415727
57      2  6.4627028
58      2  5.2091507
59      2  9.2602250
60      2  7.4173883
61      2  8.1497441
62      2 13.4838278
63      2  8.3845988
64      2  1.9356994
65      2  6.1407534
66      2  8.0005619
67      2 11.7882685
68      2  5.2035962
69      2  9.3936924
70      2  5.0722108
71      2  6.7987329
72      2 11.1883661
73      2  7.1990932
74      2 10.0564793
75      2 14.0917182
76      2  0.2258156
77      2  1.9302200
78      2  1.7673350
79      2  8.1984231
80      2  6.6291737
81      2  7.4035736
82      2 14.3373141
83      2  9.9027858
84      2  5.4479080
85      2  7.1740732
86      2  7.9115043
87      2  1.6422530
88      2  2.8073269
89      2  3.6387938
90      2 12.3151242
91      2  9.9603405
92      2  7.7684747
93      2  2.6857278
94      2 13.6305369
95      2 14.2229172
96      2  7.9807683
97      2  9.9216758
98      2  8.0806395
99      2  4.3491484
100     2 10.4791571

attr(,"class")
[1] "SimulationResult"

Solution: implement generic function print

Example - Step 3: Packaging

Generic function print:

print.SimulationResult <- function(x, ...) {
    args <- list(n1 = x$n1, n2 = x$n2, 
        mean1 = x$mean1, mean2 = x$mean2, sd1 = x$sd1, sd2 = x$sd2)
    
    print(list(
        args = format(args), 
        data = dplyr::tibble(x$data)
    ), ...)
}
x
#' @title
#' Print Simulation Result
#'
#' @description
#' Generic function to print a `SimulationResult` object.
#'
#' @param x a \code{SimulationResult} object to print.
#' @param ... further arguments passed to or from other methods.
#' 
#' @examples
#' x <- getSimulatedTwoArmMeans(n1 = 50, n2 = 50, mean1 = 5, 
#'      mean2 = 7, sd1 = 3, sd2 = 4, seed = 123)
#' print(x)
#'
#' @export
$args
   n1    n2 mean1 mean2   sd1   sd2 
 "50"  "50"   "5"   "7"   "3"   "4" 

$data
# A tibble: 100 × 2
   group values
   <dbl>  <dbl>
 1     1   7.99
 2     1   2.44
 3     1   4.82
 4     1   4.30
 5     1   5.11
 6     1   5.68
 7     1   6.98
 8     1   3.59
 9     1   7.30
10     1   7.33
# ℹ 90 more rows

Website with pkgdown

Setup of pkgdown

  • pkgdown makes it quick and easy to build a website for your package
  • After installing pkgdown, just use usethis::use_pkgdown() to get started
  • Main configuration happens in _pkgdown.yml file
  • Many customizations can be applied, but main work during development is to keep the reference section updated with names of .Rd files

Example _pkgdown.yml file

---
url: https://openpharma.github.io/mmrm

template:
  bootstrap: 5
  params:
    ganalytics: UA-125641273-1

navbar:
  right:
    - icon: fa-github
      href: https://github.com/openpharma/mmrm

reference:
  - title: Package
    contents:
      - mmrm-package
  - title: Functions
    contents:
      - mmrm
      - fit_mmrm
      - mmrm_control
      - fit_single_optimizer
      - refit_multiple_optimizers
      - df_1d
      - df_md
      - component

Publication as GitHub Page

  • It is helpful for users to read the website online
  • GitHub is very helpful here because it allows
    • A separate branch gh-pages that stores the rendered website
    • GitHub actions automatically render the website when the main branch is updated
  • To get started, use usethis::use_pkgdown_github_pages()
    • Or, manually deploy site with pkgdown::deploy_to_branch()

Exercise

Photo CC0 by Pixabay on pexels.com

Preparation

  1. Download the unfinished R package simulatr
  2. Extract the package zip file
  3. Open the project with RStudio
  4. Complete the tasks below

Tasks

Add assertions to improve the usability and user experience

Tip on assertions

Use the package checkmate to validate input arguments.

Example:

playWithAssertions <- function(n1) {
  checkmate::assertInt(n1, lower = 1)
}
playWithAssertions(-1)

Error in playWithAssertions(-1) : Assertion on ‘n1’ failed: Element 1 is not >= 1.

Add three additional results:

  1. n total,
  2. creation time, and
  3. allocation ratio

Tip on creation time

Sys.time(), format(Sys.time(), '%B %d, %Y'), Sys.Date()

Add an additional result: t.test result

Add an optional alternative argument and pass it through t.test:

alternative = c("two.sided", "less", "greater")

Implement the generic functions print and plot.

Tip on print

Use the plot example function from above and extend it.

Tip on plot

Use R base plot or ggplot2 to create a grouped boxplot of the fake data.

Optional extra tasks:

  • Implement the generic functions summary and cat

  • Implement the function kable known from the package knitr as generic. Tip: use

    kable <- function(x) UseMethod("kable")

    to define kable as generic

Optional extra task1:

Document your functions with Roxygen2

  1. If you are already familiar with Roxygen2

References

  • Gillespie, C., & Lovelace, R. (2017). Efficient R Programming: A Practical Guide to Smarter Programming. O’Reilly UK Ltd. [Book | Online]
  • Grolemund, G. (2014). Hands-On Programming with R: Write Your Own Functions and Simulations (1. Aufl.).
    O’Reilly and Associates. [Book | Online]
  • Rupp, C., & SOPHISTen, die. (2009). Requirements-Engineering und -Management: Professionelle, iterative Anforderungsanalyse für die Praxis (5. Ed.). Carl Hanser Verlag GmbH & Co. KG. [Book]
  • Wickham, H. (2015). R Packages: Organize, Test, Document, and Share Your Code (1. Aufl.). O’Reilly and Associates. [Book | Online]
  • Wickham, H. (2019). Advanced R, Second Edition.
    Taylor & Francis Ltd. [Book | Online]

License information