kcf-jackson / declare

Optional type annotation in R

R package "declare"

Allows user to add optional type annotation to a function via comments. The package uses a simple mechanism, and it is designed such that the annotation may be "switched on and off" as one pleases. This is an experiment of programming in R with stronger type.

Usage

examples.R

# Add one to a number
#$ add_one <- declare::type(add_one, x = "numeric")  # names on LHS and RHS must match
add_one <- function(x) { x + 1 }


# Column-bind two rows together
#$ cbind_rows <- declare::type(cbind_rows, r1 = "matrix", r2 = "matrix")
cbind_rows <- function(r1, r2) { cbind(r1, r2) }


# Multiply the first 3 rows of an object by a constant
#$ f <- declare::type(f, y = "numeric")  # partial specification
f <- function(x, y) { x[1:3, ] * y }

main.R

declare::source_with_type_annotation("examples.R")
add_one(10)          # Runs fine
add_one(nrow(1:10))  # Expect type error since nrow of a vector returns NULL.


X <- matrix(1:9, 3, 3)
Y <- matrix(1:3, ncol = 3)
cbind_rows(Y, Y)     # Runs fine
cbind_rows(X[1,], Y) # Expect type error since X[1,] is converted to vector implicitly.


x <- matrix(1:9, 3, 3)
x2 <- as.data.frame(x)
k <- 5
f(x, k)     # runs fine
f(x2, k)    # runs fine
f(x2, "A")  # type error

Notes

• An addin named "Source file with type annotation" is provided; it lets you source the active file in RStudio with type annotation. One may consider binding that to the hotkey 'ctrl-t' or 'cmd-t'.

• declare::type is just an usual R function call which takes a function and a type specification, and returns a procedure that calls a checker before the function.

• source_with_type_annotation is also a R function call (really, what else can it be ;p) that parses the file and makes sure the type call is evaluated immediately after the definition. (This means if you combine examples.R and main.R above into one file, it will still run fine.)