In RobinHankin/Oarray: Arrays with Arbitrary Offsets

options(rmarkdown.html_vignette.check_title = FALSE) 
knitr::opts_chunk$set(echo = TRUE)
library("Oarray")
set.seed(0)

This short document gives near-minimal examples showing the problems of evaluation within accessor methods in the Oarray package. Here we see an extractor method for the "foo" class, which uses match.call() in the same way as [.Oarray() of the Oarray package.

`[.foo` <- function(x, ...){
  mc <- match.call()
  fi <- sys.call()

  cat("mc gives: ")
  print(mc)   # NB: cannot cat(mc) as cat() cannot handle language types.

  for(i in 1:6){
    cat(paste("mc[[",i, "]] gives: |",sep=""))
    cat(mc[[i]])
    cat("|\n")
  }
  for(i in 1:6){
    cat(paste("fi[[",i, "]] gives: |",sep=""))
    cat(fi[[i]])
    cat("|\n")
  }

  cat("In the above, the output of print() is given between the two vertical lines.\n")

  for(i in 0:9){
    cat(noquote(paste("eval(mc[[5]],",i,") evaluates to " , eval(mc[[5]],i),"\n",sep="")))
  }

  cat("----\n")

  for(i in 0:9){
    cat(noquote(paste("eval(fi[[5]],",i,") evaluates to " , eval(fi[[5]],i),"\n",sep="")))
  }

  print(noquote(paste("mc[[5]]  evaluates to " , eval(mc[[5]],i),sep="")))

  cat("\n\n\n\n\n")
  return(0)
}

Above, in function [.foo(), mc is match.call() and we are interested in mc[[5]] because this is p in the calls below (such as "a[,,p=2,]").

a <- 0
class(a) <- "foo"

p <- 1
a[,,p=2,]

Above, the first element of mc is the function call, the second is a, the third and fourth are <missing>, the fifth is just 2. Now we will try using the value of variable p:

p <- 1
a[,,p,]

Above, we use p as a variable (that happens to have the value 1). So both these examples work as intended. But, I will give some examples that do not work as intended, specifically some R idiom in which symbol p exists in multiple environments. We will create a function f() and access [.foo() from within f(). First, we will pass the value of p to f():

p <- 1
f <- function(a,p){
    print(p)
    a[,,p,]
}
f(a,p)

Above we see that f() works as intended: it is picking up the value of p from the global environment (viz, 1). Now we will try passing a named argument to f():

p <- 1
f(a,p=3)

Above we want p to have the value 3, but all the evaluations of mc[[5]] (which is he symbol p) evaluate to 1, its value in the global environment. This is not what we want: it is surprising and misleading. Now an unnamed argument:

p <- 1
f(a,3)

Above we see another surprising result: even though f()'s scope has a value for p of 3, mc[[5]] nevertheless returns the global value of 1. Final example (which gives an error) shows that p might not exist in all environments that might be accessed: