Prototype-based programming
Prototype-based programming is a style of
object-oriented programming in which
classes are not present, and behaviour reuse (known as
inheritance in class-based languages) is accomplished through a process of
cloning existing
objects which serve as
prototypes. This model is also known as
class-less,
prototype-oriented, or
instance-based programming.
The original (and most canonical) example of a prototype-based language is the programming language
Self developed by David Ungar and Randall Smith. However, the class-less programming style has recently grown increasingly popular, and has been adopted for the
programming languages JavaScript,
Squeak when using the Viewer framework to manipulate
Morphic components,
Cecil,
NewtonScript,
Io,
MOO,
REBOL,
Kevo, and several others.
With class-based languages,
objects come in two general types.
Classes define the basic layout and functionality of objects, and
instances are "usable" objects based on the patterns of a particular class. In this model,
classes act as collections of behaviour (
methods) and structure which are the same for all instances, whereas
instances carry the objects' data. The role distinction is thus primarily based on a distinction between structure and behaviour on the one hand, and
state on the other.
Advocates of prototype-based programming often argue that class-based languages encourage a model of development which focus first on the taxonomy and relationships between classes. In contrast, prototype-based programming is seen as encouraging the programmer to focus on the behaviour of some set of examples and only later worry about classifying these objects into archetypal objects which are later used in a fashion similar to classes. As such, many prototype-based systems encourage the alteration of prototypes during
runtime, whereas only a very few class-based object-oriented systems (such as the first dynamic object-oriented system,
Smalltalk) allow classes to be altered during the execution of a program.
Prototype-based programming is often associated with particular schools of thought in
cognitive psychology which emphasize
prototypes or
exemplars as key attributes of the learning process.
While the vast majority of prototype-based systems are based around interpreted and dynamically typed programming languages, it is important to point out that statically typed systems based around prototypes are technically feasible. The
Omega programming language discussed in
Prototype-Based Programming [Section 2.8 (pg.177). Günther Blaschek, Omega: Statically Typed Prototypes] is an example of such a system.
In class-based languages a new instance is constructed through the class's
constructor and an optional set of constructor
arguments. The resulting instance is modeled on the layout and behaviour dictated by the chosen class.
In prototype-based systems there are two methods of constructing new objects, through
cloning of an existing object, and through
ex nihilo ("from scratch") object creation. While most systems support a variety of cloning,
ex nihilo object creation is not as prominent.
[Section 1.2 (pg.17). Chistophe Dony, Jacques Malenfan, Daniel Bardou, Classifying Prototype-based Programming Languages]Systems which support
ex nihilo object creation allow new objects to be created from scratch without cloning from an existing prototype. Such systems provide a special syntax for specifying the properties and behaviours of new objects without referencing existing objects. In many prototype languages, there is often a basic
Object prototype which carries commonly needed methods and is used as a master prototype for all other objects. One useful aspect of
ex nihilo object creation is to ensure that a new object's slot names do not have
namespace collisions with the top-level
Object object. In the Mozilla
JavaScript implementation, one can accomplish this by setting an object's
__proto__ property to null.
Cloning refers to a process whereby a new object is constructed by copying the behaviour of an existing object (its prototype). The new object then carries all the qualities of the original. From this point on the new object can be modified. In some systems the resulting child object maintains an explicit link (via
delegation or
resemblance) to its prototype, and changes in the prototype cause corresponding changes to be apparent in its clone. Other systems, such as the
Forth-like programming language
Kevo, do not propagate change from the prototype in this fashion, and instead follow a more
concatenative model where changes in cloned objects do not automatically propagate across descendants.
[Section 1.1 (pg.14). Antero Taivalsaari, Classes vs. Prototypes: Some Philosophical and Historical Observations]In prototype-based languages which use
delegation the language runtime is capable of
dispatching to the correct method or finding the right piece of data simply by following a series of delegation pointers (from object to its prototype) until a match is found. All that is required to establish this behaviour-sharing behaviour between objects is thus the delegation pointer. Unlike the relationship between class and instance in class-based object-oriented languages, the relationship between the prototype and its offshoots does not require that the child object have a memory or structural similarity to the prototype beyond this link. As such, the child object can continue to be modified and amended over time without re-arranging the structure of its associated class as in class-based systems. It is also important to note that not only data but also methods can be added or changed. For this reason most prototype-based languages refer to both data and methods as "slots".
Under pure prototypingâ€"which is also referred to as
concatenative prototypes and is exemplified in the
Kevo programming languageâ€"there are no visible pointers or links to the original prototype from which an object is cloned. The prototype object is copied exactly, but given a different name (or reference). This process can be compared in some ways to biological
mitosis. Methods and attributes are simply duplicated as-is.
Advantages to this approach include the fact that object authors can alter the copy without worrying about side-effects across other children of the parent. A further advantage is that the computational cost of method lookup during dispatch is drastically reduced when compared to delegation, where an exhaustive search must be made of the entire delegation chain before failure to find a method or slot can be admitted.
Disadvantages to the concatenative approach include the organizational difficulty of propagating changes through the system; if a change occurs in a prototype, it is not immediately or automatically available on its clones. However, Kevo does provide additional primitives for publishing changes across sets of objects based on their similarity (so-called
family resemblances) rather than through taxonomic origin, as is typical in the delegation model.
Another disadvantage is that, in the most naive implementations of this model, additional memory is wasted (versus the delegation model) on each clone for the parts that have stayed the same between prototype and clone. However, it is possible to provide concatenative behaviour to the programming while sharing implementation and data behind-the-scenes; such an approach is indeed followed by
Kevo.
[Antero Taivalsaari. Kevo, a prototype-based object-oriented programming lnaguage based on concatenation and module operations. Technical Report Report LACIR 92-02, University of Victoria, 1992.].
An alternative quasi-solution to the problem of clones interfering with the behaviour of the parent is to provide a means whereby the potential parent is flagged as being clonable or not. In
MOO, this is achieved with the "f" flag. Only objects with the "f" flag can be cloned. In practice, this leads to certain objects serving as surrogate classes; their properties are kept constant to serve as initial values for their children. These children then tend to have the "f" flag not set.
Advocates of class-based object models who criticize prototype-based systems often have concerns that could be seen as similar to those concerns that proponents of static type systems for programming languages have of dynamic type systems (see
Datatype). In particular, those concerns revolve around:
correctness,
safety,
predictability and
efficiency.
In regard to the first three points, classes are often seen as analogous to types (and indeed, in most statically typed object-oriented languages they serve that role) and are proposed to provide contractual guarantees to their instances and users of their instances that they will behave in a given fashion.
On the last point, efficiency, the declaration of classes simplifies many
compiler optimizations that allow the development of efficient method and instance variable lookup. In the case of the
Self programming language, much development time has been spent on the development of compilation and interpretation techniques to improve the performance of prototype-based systems versus their class-based competitors. For example, Lisaac compiler produces code almost as fast as C. Tests have been run with a MPEG-2 codec written in Lisaac, copied from a C version. These tests show the Lisaac version is 1.9% slower than the C version with 37% fewer lines of code.
Finally, perhaps the most common criticism leveled against prototype-based languages is that the community of
software developers is not familiar with them, despite the popularity and market permeation of
JavaScript, (and in point of fact, the current work on a fourth edition of the
ECMAScript standard is seeking to turn JavaScript into a class-based language). Further, as prototype-based systems are relatively novel, and few and far between,
best practices for software development using them have not become widespread.
*Actor-Based Concurrent Language, ABCL:
ABCL/1,
ABCL/R,
ABCL/R2,
ABCL/c+*
Agora*
Cecil*
Cel*
ColdC*
ECMAScript a.k.a.
ActionScript,
DMDScript,
JavaScript (first named
Mocha, then
LiveScript),
JScript*
Factor*
Io*
Javascript*
Lisaac*
Logtalk*
Lua*
LPC*
MOO*
NewtonScript*
Obliq*
OpenLaszlo*
REBOL*
Self*
Slate*
Squeak when using the Viewer framework to manipulate
Morphic components
*
TADS*
*
Class-based programming (contrast)
*
Programming paradigms
