Cannot override field with method in subclass

[hriebl]

I just figured out that it's not possible to override a field with a method in a subclass. Paradoxically, the opposite direction works. Maybe this is the desired behavior or I overlooked this aspect in the package documentation/the other GitHub issues. If so, please let me know.

field → field works:

parent <- R6Class("parent", list(
  x = 1
))

child <- R6Class("child", inherit = parent, list(
  x = 2
))

parent$new()
# <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: 1
child$new()
# <child>
#   Inherits from: <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: 2

field → method doesn't work:

parent <- R6Class("parent", list(
  x = 1
))

child <- R6Class("child", inherit = parent, list(
  x = function() 2
))

parent$new()
# <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: 1
child$new()
# <child>
#   Inherits from: <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: 1

method → method works:

parent <- R6Class("parent", list(
  x = function() 1
))

child <- R6Class("child", inherit = parent, list(
  x = function() 2
))

parent$new()
# <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: function ()
child$new()
# <child>
#   Inherits from: <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: function ()

parent$new()$x()
# [1] 1
child$new()$x()
# [1] 2

method → field works:

parent <- R6Class("parent", list(
  x = function() 1
))

child <- R6Class("child", inherit = parent, list(
  x = 2
))

parent$new()
# <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: function ()
child$new()
# <child>
#   Inherits from: <parent>
#   Public:
#     clone: function (deep = FALSE)
#     x: 2

parent$new()$x()
# [1] 1

[wch]

This is a bug. I think that it makes sense to disallow overriding in either direction, from method->field or field->method, because then the subclass will behave differently from the superclass.

[hriebl]

Actually, I was a bit surprised it didn't work, because there is usually no type checking in R. I was trying to write a "template class" with a lot of NA fields, which the subclasses would override with useful methods. But probably you're right and it would be more sane to prohibit this behavior altogether.

[wch]

I was wondering about the use case you had in mind. We've been thinking about interfaces for R6, but haven't yet decided on a course of action (#163).

[hriebl]

Interfaces seem like the proper way to implement what I was trying to achieve. Anyway, I think it would be a good idea to make the "field → method", "method → field" behavior consistent. :)

[jayqi]

I was trying to write a "template class" with a lot of NA fields, which the subclasses would override with useful methods. But probably you're right and it would be more sane to prohibit this behavior altogether.

@hriebl A different idea that works today is to define methods that don't do anything in your template class, or to be a little fancier, define methods that error out if called.

[hriebl]

Thank you, @jayqi, I ended up doing something like that. The whole bug is not a big deal, you can definitely work around it, but it's a bit weird...

[krlmlr]

Is this related to the inability to access public fields via super$?

library(R6)

MySuper <- R6Class(
  "MySuper",
  public = list(
    public_member = NULL,
    initialize = function(...) {
      self$public_member <- 1
    },
    get_super_public_member = function() {
      self$public_member
    }
  )
)

MySub <- R6Class(
  "MySub",
  inherit = MySuper,
  public = list(
    get_public_member = function() {
      list(
        self = self$public_member,
        super = super$public_member
      )
    }
  )
)

my_sub <- MySub$new()
my_sub$get_super_public_member()
#> [1] 1
my_sub$get_public_member()
#> $self
#> [1] 1
#> 
#> $super
#> NULL

^{Created on 2020-12-11 by the reprex package (v0.3.0)}

[MSHelm]

I also just stumbled over this issue. I initially had defined the method as a field = NULL in the super class and defined the according method in the subclass. Would be great if this behavior could be made consistent or described in the documentation.