R: How to overload + operator to call a method which adds a function to an R6 class?

Question

My aim is to overload the + operator in order to call a method on an R6 object. Here is a toy example with an R6 object of class aclass that has a method add_function(foo) to store some arbitrary function in itself.

I want to substitute the call to add_function(foo) by + foo in the style of ggplot2's +.gg or +(e1,e2). I checked the ggplot2 github code but could not figure this out.

Here is an R6 object to which we can add arbitrary functions

library(R6)
# The class (called aclass) looks like so:
Aclass <- R6Class("aclass",
  public = list(
    fun = NULL,
    x = NULL,
    initialize = function(x) {
      self$x <- x
      return(invisible(self))
    },
    # this is the function adding part
    add_fun = function(fun) {
      self$fun <- substitute(fun)
    },
    # and here we execute the functions
    do_fun = function() {
      return(eval(self$fun)(self$x))
}))

# Say, we want to add this function to an aclass object
foo <- function(x) mean(x, na.rm=TRUE)

This works

# Instantiate class
my_class <- Aclass$new(c(0,1,NA))
# Add the function
my_class$add_fun(function(x) mean(x, na.rm=TRUE))
# my_class$add_fun(foo) # this also works
# Execute the function - beautiful - returns 0.5
my_class$do_fun()

This fails

My goal is to do the same as above, i.e. adding foo to the object, but now by overloading the + operator to do the job in a nicer way

# Let's try to overload the + operator
`+.aclass` <- function(e1, e2) {
  return(e1$add_fun(e2))
}

# option 1
my_class <- Aclass$new(c(0,1,NA)) + foo
# fails
# Looking at my class we get
my_class
# > my_class                                       
# e2

# option 2 <--- this is my end goal
my_class <- Aclass$new(c(0,1,NA)) + mean(x, na.rm=TRUE)
# also fails because then R tries to execute the function
# > Error in mean(x, na.rm = TRUE) : Object 'x' not found

# Note: in this case, the custom +.aclass is not even called,
# but R calls the normal + method I think

I am not sure how to solve this or if it is actually possible...

Thanks for any help!

Billy34 · Accepted Answer · 2019-12-05T14:55:28.067

by removing the substitute in your add_fun function and returning self (because you are affecting it to a new object)

    # this is the function adding part
    add_fun = function(fun) {
      self$fun <- fun
      self
    },

it will work with

my_class <- Aclass$new(c(0,1,NA)) + foo
and also with
my_class <- Aclass$new(c(0,1,NA)) + function(x) mean(x,na.rm=TRUE)

for your goal case it would be harder

+ mean(x, na.rm=TRUE)

even ggplot2 functions are evaluated (eg labs(x="title") will evaluates to an object of class label that being added to an ggplot will finally call function ggplot_add.labels)

By slightly tweaking your goal case using formula notation it can be solved

+ ~mean(x, na.rm=TRUE)

using this code

`+.aclass` <- function(e1, e2) {
  se2 <- substitute(e2)

  # detect formula
  if(is.call(se2) && as.character(se2)=="~") {
    # build template function
    e3 <- substitute(function(x) y)
    # replace y in template by formula body
    e3[[3]] <- se2[[2]]
    # use our templated formula function
    e2 <- e3
  }
  e1$add_fun(e2)
  e1
}

Finally using higher level functions available in rlang we can simplify it to

`+.aclass` <- function(e1, e2) {
  e1$add_fun(rlang::as_function(e2))
  e1
}

Note that the formula like function will need to have .x instead of x

R: How to overload + operator to call a method which adds a function to an R6 class?

Here is an R6 object to which we can add arbitrary functions

This works

This fails

1 Answers1