Skip to content


Subversion checkout URL

You can clone with
Download ZIP
Prototype Programming with Contexts and Forms
branch: master

Fetching latest commit…

Cannot retrieve the latest commit at this time

Failed to load latest commit information.

Contextual prototype programming in R.

Prototype programming is an object oriented (OO) programming paradigm that uses objects rather than classes to represent behavior and structure of objects. Inheritance is also classless, that is object inherit from other objects.

Class-based object orientation (Java, C++, Ruby etc.) separates the instances (objects) from their abstract representation (classes). Thus, a developer has first to declare classes and then instantiate objects. This has an advantage of keeping declarations separately from dynamically created objects, but makes designing process more tedious. Prototype-based OO completely drops classes and thus allows for a faster development cycle.


ProtoClasses package is designed for complex applications with very pronounced hierarchical organization of the functionality, code and data. It has similarities with proto package and Reference Classes in R and it seamlessly blends with R's class system.

There are two fundamental objects, protoCells and protoContext, which both inherit (in S4 sense) from common evnProtoClass. The latest is only of the very little importance and won't be discussed much in what follows.

ProtoCells are the elementary objects that can instantiate other objects. They encapsulate methods, fields and forms. Methods and fields are pretty much like methods and fields in other OO systems (like Reference Classes in R). Forms are specific to protoClasses package and are essentially unevaluated expressions that proto objects inherit from each other. Forms are essentially an R-ish way to implement proper dynamic dispatch. More on this later.

ProtoCells are a very handy tool to hierarchically build functionality. Often it is very convenient to "store" this functionality in some common place instead of letting it "float" around in the form of individual objects. This is the task of protoContexts.

ProtoContexts are also proto objects that have methods, fields and forms, but in addition they can also store protoCells. In this sense, protoContexts are like containers for protoCells. The interpretation of protoContexts is application specific but it is always a context of a complex task. It can be, for example, a model containing data, analysis code, results, simulation, etc. It can be, a workplace, holding all your code and data relevant to some task or application (just like R's global environment). It could be a plot, holding your data, code and all the layers necessary for the plot.

Proto Cells

All proto objects are implemented as classed environments, that is they inherit from S4 class environment. See [implementation].

Each proto object has a short type (object name), which default to "--":

x <- protoCell()
## [1] "--"

All protoCells inherit from the root cell with the name "*". The long type (full name) consists of the concatenation of all the short types of its direct and indirect prototypes:

x <- protoCell(type = "x")
y <- x$new(type = "y")
z <- y$new("funny_type")
## [1] "funny_type"
## [1] "funny_type.y.x.*"

In the above example x is a prototype of y, and y is a prototype of z and it means in technical terms that x is parent environment of y, and y is parent environment of z. Thus, all objects defined in x are visible in z unless they are redefined in y or z.

The hierarchy always starts with the root (*) type that holds all the builtin functionality necessary for the proto-system to work. The enclosure of the root proto object is by default global environment.


Fields represent the abstract idea of a state of the objects. In protoClasses fields are implemented by means of accessor functions. You can initialize fields in two ways. Simple way, by supplying the initial value, and advanced by supplying the accessor function that will take care of the storing and retrieval of the field.

Simple interface builds the default accessor function that simply assigns and retrieves the value in the proto object with the appropriate conversion as given by the class of initial value supplied. For example:

x$initFields(int = 12L, str = "sfdsfdf")
## [1] 12
x$int <- 13.3
## [1] 13
x$str <- 13.3
## [1] "13.3"

In the above example int fields assign and retrieves an integer value from the object; str is always a string.

Children of x automatically inherit all the fields (also methods and forms) from their parents:

## [1] 13
## [1] 13
y$int <- -100
## [1] -100

Fields can be accessed by means of the built in field fields:

## [1] 13

Default fields assign the object in the proto environment with the same names as supplied during the initialization:

## [1] "e"   "int" "str"

For a more complex fields you can supply accessors directly:

liquid <- protoCell("liquid")

liquid$initFields(kelvin = 0, celsius = protoField(function(value) {
    if (missing(value)) {
        kelvin + 273.15
    } else {
        assign("kelvin", value - 273.15, envir = .self)

## [1] 273.1
liquid$kelvin <- 100
## [1] 373.1
liquid$celsius <- 100
## [1] -173.1

The convention for field accessor is pretty simple. It is a function one argument and it is called with no argument if it is used as a getter (liquid$celsius) and it is called with one argument when it is used as a setter (liquid$celsius <- value).

Note that each field accessor can access the calling object with .self. In the above example celsius accessor assigns variable kelvin new value in environment .self. Moreover, during the call to celsius field, the accessor enclosure is dynamically set to be x (same as .self).

Often it is useful to access other fields from withing field accessor. This could be done by means of object .self or by using .fields container from within accessor function. The following are equivalent in terms of behavior:

fahrenheit1 <- function(value) {
    if (missing(value)) 
        .fields$celsius * 9/5 + 32 else .fields$celsius <- (value - 32) * 5/9

fahrenheit2 <- function(value) {
    if (missing(value)) 
        .self$fields$celsius * 9/5 + 32 else .self$fields$celsius <- (value - 32) * 5/9

liquid$initFields(fahr1 = protoField(fahrenheit1), fahr2 = protoField(fahrenheit2))

The first version of fields uses the internal container .field that stores all the accessors. Same holds for methods, forms and cells, which are stored in specialized containers. Containers will be rigorously defined latter.


Methods are just functions that can see all the objects from withing the proto object. In contrast to R standard functions, which have fixed enclosure (lexical scope), the enclosure of methods is the proto object itself. This is an instance of dynamic scope.

For example:

liquid$initMethods(freeze = function() .self$celsius <- freezing_temp, boil = function() .self$celsius <- boiling_temp, 
    warm_up = function(by = 10) assign("kelvin", kelvin + by, .self))

liquid$celsius <- 100
## [1] 133

The freeze and boil methods will through an error because freezing_temp and boiling_temp are undefined.

watter <- liquid$new("watter", expr = expression(freezing_temp = 0, boiling_temp = 100))

watter$celsius <- 17
## [1] 30
## Error: object 'freezing_temp' not found
## [1] 30

The above example ilustriate the crucial distinction between OO and functional programming - dinamic scoping. The scope of freeze and boil methods is being dynamically changed depending on the context. Even though these methods were defined in object liquid, when called by watter they operate on objects defined in watter and not in liquid. This mecanism is very powerful but it is also complex and often very dificult to analize.


Proto forms are specially managed expressions. They allow selective inheritance of the code. As you will see proto forms inheritance is the ultimate example of code reuse.

x$initForms(A = form(B1 = form({
    a <- 10
}), B2 = form(C = form({
    a * 10

## e(A.B1) (from <text>#2)
##  {
##      a <- 10
##  }
## e(A.B2) .. e(A.B2.C) (from <text>#4)
##     {
##         a * 10
##     }

Let's write a method to test this out.

x$initMethods(test = function() e(A))  ## e function evaluates forms in .self environent
## [1] 100

Now replace A.B1 form:

y$setForms(A.B1 = form(a <- -10))
## [1] -100

Proto Context


Application Development

S4 Inheritance


Other Programming Paradigms




Something went wrong with that request. Please try again.