Constructing function from a closure with Magrittr is resulting in an error upon function evaluation

Question

When I use magrittr to pass a value for x to the function below, it results in an unusable function. Why is this happening? I have version magrittr_1.5.

library(magrittr)
f <- function(x) { function(y) { x + y } }

# Case 1 (works)
f.5 <- f(5)
f.5(6) # returns 11

# Case 2 (works)
f.5 <- f(5) 
6 %>% f.5 # returns 11

# Case 3 (fails)
f.5 <- 5 %>% f
6 %>% f.5 # Error in x + y (from 1) : 
          # non-numeric argument to binary operator 

# Case 4 (fails)
f.5 <- 5 %>% f
f.5(6); # Same error as case 3

Answer 1

Here is a partial answer. First off, it is easy to make it work if you evaluate x in the calling environment of the inner anonymous function,

library(magrittr)
f <- function(x) { x; function(y) x + y }
f.5 <- 5 %>% f
6 %>% f.5
# [1] 11

To investigate what is going wrong, see what x is evaluating to,

f <- function(x) {
    function(y) {
        print(sprintf("x: %s.", toString(x)))
        x + y
    }
}
## the failing case
fails <- 5 %>% f
6 %>% fails
# [1] "x: function (y) \n{\n    print(sprintf(\"x: %s\", toString(x)))\n    x + y\n}, TRUE"

It is pointing the result of calling withVisible on f, which happens when a function, freduce, is called (code). In the body of the pipe function, there is a line,

body(magrittr:::pipe)[[2]][[3]][[10]]
# env[["freduce"]] <- freduce

where freduce is made available (look at magrittr:::pipe for full context).

If this line is modified to simply be the actual code in the function freduce (ie. copying magrittr:::freduce), it seems to work,

## Make a modified %>% operator
mypipe <- magrittr:::pipe

## Change that line
body(mypipe)[[2]][[3]][[10]] <- quote(
    env[["freduce"]] <-
        function(value, function_list) {
            k <- length(function_list)
            if (k == 1L) {
                result <- withVisible(function_list[[1L]](value))
                if (result[["visible"]])
                    result[["value"]]
                else
                    invisible(result[["value"]])
            } else {
                Recall(function_list[[1L]](value), function_list[-1L])
            }
        }
)

## Redefine
`%>%` <- mypipe()

## Test (now it works?!)
fails <- 5 %>% f
6 %>% fails
# [1] "x: 5"
# [1] 11

So, this is a partial solution because I can't figure out why redefining freduce in this way makes it work.