[Rd] Reference Class error message: may be caused by lazy evaluation?

John Chambers jmc at r-project.org
Thu Jun 9 19:11:21 CEST 2011


Good catch.

Here's the problem.  To save space and time, reference methods are not 
all copied to every object in the class.  Instead, the methods are 
copied in when first accessed.   Methods are functions which use the 
object as their environment.  So that is the sense in which "lazy 
evaluation" is involved.

If a method calls another method (add() calling addOne() in your 
example), then the method for the `$` operator knows to copy over that 
method (addOne).  (The second of my examples below shows this.)  But if 
the method _refers_ to another method without calling it, the code 
analysis does not currently catch the reference.

We can fix that, although it's a little subtle. Meanwhile, your 
showself() is a good workaround.

For anyone interested, the code below illustrates.

One point of style.  I would suggest saving the generator object and 
calling its $new() method, as below, rather than treating the reference 
class as an S4 class.  The result is identical AFAIK, but the style is 
more typical of such OOP languages.

John

---------------------------
 > tc <-  setRefClass("testclass", fields = list(a = "numeric"),
+             methods = list(
....
+               ))
 > t1 <- tc$new(a=1)
 > ss = t1$show
 > ss
Class method definition for method show()
function ()
{
     print(addOne)
}
<environment: 0x10188d5f8>
 > ev = environment(ss)
 > ev
<environment: 0x10188d5f8>
 > t1
An object of class "testclass"
<environment: 0x10188d5f8>  # <<<< same environment
 > objects(ev)
[1] "a"    "show"  #<<<< not addOne, though
 > t1$addOne
Class method definition for method addOne()
function ()
{
     a <<- a + 1
     print(a)
}
<environment: 0x10188d5f8>
 > objects(ev)
[1] "a"      "addOne" "show"  # <<<< now addOne is there


On 6/8/11 4:38 PM, Tengfei Yin wrote:
> Dear All,
>
> I came across an error message recently when constructing a reference class,
> an example is attached below, it looks like only if I call a specific method
> in advance, otherwise it cannot be found in defined method without using
> .self, this make it difficulty that sometimes in my initialize method, I
> need to call other method defined in the same reference class, the
> workaround for this is add .sef to it.
>
>
> ###############  example begin ############################
> setRefClass("testclass", fields = list(a = "numeric"),
>              methods = list(
>                addOne = function(){
>                  a<<- a+1
>                  print(a)
>                },
>                add = function(){
>                  addOne()
>                },
>                show = function(){
>                  print(addOne)
>                },
>                showself = function(){
>                  print(.self$addOne)
>                }
>                ))
>
> obj<- new("testclass", a = 1)
> obj$show()                                                              #
> Error in print(addOne) : object 'addOne' not found
> obj$addOne()                                                           #
> return 2, works
> obj$show()                                                              #
> after calling addOne(), show() works
>
> ## works if use .self$...
> obj2<- new("testclass", a = 1)
> obj2$showself()
>
> ## works when call the method directly within another method
> obj3<- new("testclass", a = 1)
> obj3$add()
>
> ################ end ##########################################
>
> I am still learning this new technique, if I made any mistake I didn't
> notice before, please let me know, I will really appreciate that.
>
> Thanks a lot!
>
> Tengfei
>



More information about the R-devel mailing list