It is possible to overwrite methods in R6 objects

[richfitz]

Once an R6 object has been constructed, it is possible to overwrite its methods, which seems undesirable:

Thing <- R6::R6Class("Thing",
                     public=list(
                       a=1L,
                       f=function() a))
obj <- Thing$new()
obj$f() # 1L
obj$f <- NULL # is allowed
obj$f() # Error: attempt to apply non-function

And of course, putting the method "back" doesn't work because the environment of that function is wrong:

obj$f <- function() a
obj$f() # Error: object 'a' not found

Is there any way of easily preventing this happen without sacrificing the speed goals of R6?

[richfitz]

Sorry - should have said: this affects the current CRAN version (1.0.1), as well as development version a1177fc49282676e4a0ab00ab7c629d783159443

[wch]

Yes, it probably makes sense to lock the bindings for methods when lock=TRUE. As it is, that option stops you from adding new variables, but doesn't prevent modifying existing ones.

[wch]

I decided that method bindings should always be locked; if users really want to modify them, they can always unlock the bindings themselves.

[mmuurr]

I realized this was closed 6(!) years ago ... but I'm wondering how one would "unlock the bindings themselves". I suddenly find myself in a position where I'd like to update a method after an object has been created (i.e. an update on the instance). I tried creating the generator with lock_objects = FALSE, but I'm still getting the cannot change value of locked binding for 'foo' error. Is this doable with current R6 implementation?

[gaborcsardi]

unlockBinding() unlocks a binding.

[mmuurr]

@gaborcsardi ahhhh... I hadn't thought of that base function. Thanks!