Q7 type system provides an infrastructure to create objects in R; It is more advanced than native R classes, and is free from the grand narrative of hereditary OOP.
See other vignettes for :
R6
..my
Q7 employs conventional OOP terms concepts, with some slight variations:
object - a unit of program and data, may refer to type or instance, or both
type - blueprint for an object
instance - an embodiment of a type
member - things bound to an object; some members are functions
method - a function that is bound and (usually) domestic to an object
type()
.my
, which refers to the instance itselffeature()
implement()
Make a type:
Everything defined within the function’s closure become members of the object. The function’s arguments are accessible by bound functions of the object, but not become members themselves.
The following symbols are reserved by the Q7 type system and shall not be re-bound by the user.
Environments:
.my
: an object’s public environment, which the user and other parts of the program interact with.private
: an object’s private environment, which is parent to the .my
environmentBinding Modifiers:
private
: designates a binding in the private environmentpublic
: designates a binding in the public environment (default)final
: designates an immutable binding in the public environmentprivate_final
: designates an immutable binding in the private environmentactive
: designates an active binding in the public environmentactive_private
: designates an active binding in the private environmentFunctions:
initialize
: runs at the instantiation of an objectfinalize
: runs when an object is destroyed by the garbage collectorThere are two main strategies of extending an object: inheritance and composition. Q7 employs composition, and the benefit is obvious.
When you code with inheritance, your mind must navigate from sub- to super- classes from the inside out; Composition, on the other hand, is the linear addition to existing code, which is simpler for the mind to follow.
Types and instances can both be extended in the same manner. The concatenative nature of Q7 makes different objects truly independent from each other.
To extend an object, use implement()
. If the object is a type, the resulting type must to be bound to a name; if the object is an instance, it is modified in place (see below). Modifying a type will not impact instances already created by the same type.
Code can also be packaged with feature()
for later use.
hasFeatureTwo <- feature({
n <- 100 # Overwrites n from TypeTwo
x <- 10 # Overwrites x from hasFeatureOne
private[x_plus_n.old] <- x_plus_n
# Rename to preserve the old x_plus_n()
# Mark private, because it is only going to be used by the new x_plus_n()
x_plus_n <- function(){
cat(sprintf("adding x (%i) to n (%i)...\n", x, n)) # do some extra thing
x_plus_n.old() # call the old function
}
})
type_two_with_features <- TypeTwo() %>%
hasFeatureOne() %>%
hasFeatureTwo()
type_two_with_features$x_plus_n()
#> adding x (10) to n (100)...
#> [1] 110
Any domestic function of an object can read from and write to the private environment. Remember to use the double arrow - <<-
- because you want the assignment to pierce the function’s closure and reach the object itself.
Use caution: if the symbol you’re assigning to with <<-
does not exist in either public or private environments of the object, it will end up somewhere outside the object, sometimes in the global environment.
Counter <- type(function(){
private[count] <- 0
add_one <- function(){
count <<- count + 1
# Your IDE's syntax checker may alert you that
# `count` is not found in scope.
# You can safely ignore this.
}
get_count <- function(){
count
}
})
counter <- Counter()
ls(counter) # `count` can't be seen from the out side
#> [1] "add_one" "get_count"
counter$get_count() # but count can be read by domestic function
#> [1] 0
counter$add_one() # ... and be written to
counter$add_one()
counter$get_count() # when we read it again the number changes
#> [1] 2
As stated above, the private environment (.private
) is parent of the public environment (.my
). Parameters supplied to the arguments of the constructor function are implicitly private. When two members in private and public environments have the same name, they may co-exist. However, only the one in .my
will win; the one in .private
must be explicitly qualified.
The following code allows direct outside access to the count
object.
exposePrivate <- feature({
.my$pvt_env <- .private$.private # `.private` contains a reference of itself with the same name, assigns it to `.my`
#pvt_env <- .private # also works
})
counter %>% exposePrivate()
# .private reference appears in the object
ls(counter, all.names = TRUE)
#> [1] "add_one" "get_count" "pvt_env"
counter$.private
#> NULL
counter$pvt_env$count # It is now possible to directly access any variable in the private environment
#> [1] 2