Trade-offs

Learning objectives:

• Decide whether to use S3 or S4 (or maybe S7) for a project.
• Decide whether to use R6 or S3 (or maybe S7) for a project.

Default to S3*

“I recommend that you default to S3.” - Hadley Wickham

• Simple & widely used
• R6 leads to non-idiomatic R code
  • More like other languages, so tempting for newcomers

* Or maybe S7?

• “Functional OOP” aka “Generic function OOP” aka “Multimethod OOP”
  • S3, S4, S7
• “Encapsulated OOP”
  • R6

Use S4 for large team projects*

• S4 requires more up-front design than S3
• S4 is more formal, strict, and verbose
• Structure from S4 reduces risk from new team members
• Examples:
  • Bioconductor (~3-4k packages)
  • {Matrix} (108 classes, 23 generic functions, and 1780 methods)
• S4 lacks great docs (needs a book)

* Maybe use S7 instead?

Use S3 for standard, shared methods

• S3 generics are global functions
  • Standard generics: print(), summary(), plot(), predict(), etc.
  • Easy to create your own class & methods
• R6 methods are local to the object
  • Can have different args than standard method
  • Can mean something else entirely vs standard
• R6 objects are S3 objects, too!
  • Can define S3 methods for R6 classes

Use R6 for stateful objects

• R6 methods can modify object state
  • No need to thread state through functions
• Chain R6 methods with $

# Stack is an `R6::R6Class()` with `push()`, `pop()`, and `length()` methods
s <- Stack$new()
s$
  push(10)$
  push(20)$
  pop()
#> [1] 20

OOP system comparison

	S3	S4	S7	R6
Package	base R	base R	S7	R6
OOP type	Functional	Functional	Functional	Encapulated
Complexity	Low	High	Medium	High
Payoff	Low	High	High	High
Team size	Small	Large	Large	Small-large
Namespace	Global	Global	Global	Local
Modify in place	No	No	No	Yes
Method chaining	|>	|>	|>	$
Get/set	$	@	@	$
Create class	function()	setClass()	new_class()	R6Class()
Create validator	function()	setValidity() | setClass(validator = f())	new_class(validator = f())	$validate()
Create generic	UseMethod()	setGeneric()	new_generic()	NA
Create method	g.y <- function()	setMethod()	method()	R6Class()
Create object	class(x) <- "y"	new()	x <- class_y()	$new()
Additional components	attributes	slots	properties	fields & methods