Using 'get' Inside Nested Functions on Local Variables

Question

I've never quite gotten my head around nesting functions and passing arguments by reference. My strategy is typically to do something like get('variabletopassbyreference') inside the child function to accomplish this.

Until now, I have been passing global variables to the function and this worked fine. Today I tried to create local variables inside a function and then pass those to a nested function within that function and it failed. I'm unable to get get to work. I also tried tinkering the pos and inherits but to no avail.

I cannot find an exact answer on the net. If I could get this construct to work then that's my preference because I have a bunch of other functions that I've coded up in similar fashion. If I shouldn't be doing this at all and should be doing something else, then that information would be appreciated as well.

An example is below -

test1 <- function(a1,b1) {

  # cat(ls()) # a1 b1
  # cat(ls(pos = 1)) # c test1 test2

  testvalue <- get('c') * get(a1, inherits = TRUE) * get(b1)

  testvalue

}

test2 <- function() {

  a = 1
  b <- 2
  # cat(ls()) # a b
  test1('a','b')

}

c = 3
test2()

I get the following error -

Error in get(a1, inherits = TRUE) : object 'a' not found 

More generic example -

a = 0

test1 <- function(a1,b1) {

  # cat(ls()) # a1 b1
  # cat(ls(pos = 1)) # c test1 test2

  testvalue <- get('c') * a1 * b1

  assign(x = 'a', value = 2.5)
  assign(x = 'a', value = 3.5, envir = parent.frame())
  assign(x = 'a', value = 4.5, envir = .GlobalEnv)
  cat(a)
  cat(' - value of a local within test1\n')
  testvalue

} 

test2 <- function() {

  a = 1
  b <- 2
  # cat(ls()) # a b

  cat(a)
  cat(' - value of a local within test2 before test1 called\n')
  test1(a1 = a, b1 = b)
  cat(a)
  cat(' - value of a local within test2 after test1 called\n')

}
cat(a)
cat(' - value of a global before test 2 \n')
c = 3
test2()

cat(a)
cat(' - value of a global after test 2 \n')

Answer 1

Also, pass the environment that the variables are located in. Note that parent.frame() refers to the environment in the currently running instance of the caller.

test1 <- function(a1, b1, env = parent.frame()) {

  a <- get(a1, env)
  b <- get(b1, env)
  c <- get('c', env)

  testvalue <- c * a * b

  testvalue

}

c <- 3
test2() # test2 as in question
## 6

Here a and b are in env c is not in env but it is in an ancestor of env and get looks through ancenstors as well.

ADDED Note that R formulas can be used to pass variable names with environments:

test1a <- function(formula) {
    v <- all.vars(formula)
    values <- sapply(v, get, environment(formula))
    prod(values)
}

test2a <- function() {
    a <- 1
    b <- 2
    test1a(~ a + b + c)
}

c <- 3
test2a()
## 6

REVISION: Corrected. Added comment. Added info on formulas.

Answer 2

Since you are asking, this definitely looks like a bad design to me. The recommended approach is to stick to R's way of pass-by-value. And as much as possible, make every function take everything it uses as arguments:

test1 <- function(a1, b1, c1 = 1) {
   testvalue <- c1 * a1 * b1   
   testvalue
}

test2 <- function(cc = 1) {
   a <- 1
   b <- 2
   test1(a1 = a, b1 = b, c1 = cc)
}

cc <- 3
test2(cc = cc)

(I replaced c with cc since it is the name of a function, hence a bad idea to use as variable name.)

A less acceptable but maybe closer approach to what you have is to not pass all arguments to your functions and let R look for them in the calling stack:

test1 <- function(a1, b1) {
   testvalue <- cc * a1 * b1   
   testvalue
}

test2 <- function() {
   a <- 1
   b <- 2
   test1(a, b)
}

cc <- 3
test2()

If for some reason the first approach does not work for you, please explain why so I get a chance to maybe convince you otherwise. It is the recommended way of programming in R.

---

Following on the discussion and your edit, I'll recommend you look at the proto package as an alternative to get and assign. Essentially, proto objects are environments so it's nothing you can't do with base R but it helps make things a bit cleaner:

test1 <- function(x) {
   testvalue <- x$c * x$a * x$b
   x$a <- 3.5
   testvalue
}

test2 <- function(x) {
   x$a <- 1
   x$b <- 2
   cat(x$a, '\n')
   test1(x)
   cat(x$a, '\n')
}

library(proto)
x <- proto(c = 3)
test2(x)

From a programming point of view, test1 and test2 are functions with side-effects (they modify the object x). Beware that its a risky practice.

Or maybe a better approach is to make test1 and test2 be methods of a class, then it is acceptable if they modify the instance they are running on:

x <- proto() # defines a class

x$test1 <- function(.) {
   testvalue <- .$c * .$a * .$b
   .$a <- 3.5
   testvalue
}

x$test2 <- function(.) {
   .$a <- 1
   .$b <- 2
   cat(.$a, '\n')
   .$test1()
   cat(.$a, '\n')
}

library(proto)
y <- x$proto(c = 3)  # an instance of the class
y$test2()

If you are not interested in using a third-party package (proto), then look at R's support for building classes (setClass, setRefClass). I do believe using an object-oriented design is the right approach given your specs.