Table of Contents
Previous Chapter
Table of Contents
Previous Chapter
Chapter 1
Programming languages have traditionally divided the world into two parts--data and operations on data. Data is static and immutable, except as the operations may change it. The procedures and functions that operate on data have no lasting state of their own; they're useful only in their ability to affect data.
This division is, of course, grounded in the way computers work, so it's not one that you can easily ignore or push aside. Like the equally pervasive distinctions between matter and energy and between nouns and verbs, it forms the background against which we work. At some point, all programmers--even object-oriented programmers--must lay out the data structures that their programs will use and define the functions that will act on the data.
With a procedural programming language like C, that's about all there is to it. The language may offer various kinds of support for organizing data and functions, but it won't divide the world any differently. Functions and data structures are the basic elements of design.
Object-oriented programming doesn't so much dispute this view of the world as restructure it at a higher level. It groups operations and data into modular units called objects and lets you combine objects into structured networks to form a complete program. In an object-oriented programming language, objects and object interactions are the basic elements of design.
Every object has both state (data) and behavior (operations on data). In that, they're not much different from ordinary physical objects. It's easy to see how a mechanical device, such as a pocket watch or a piano, embodies both state and behavior. But almost anything that's designed to do a job does too. Even simple things with no moving parts such as an ordinary bottle combine state (how full the bottle is, whether or not it's open, how warm its contents are) with behavior (the ability to dispense its contents at various flow rates, to be opened or closed, to withstand high or low temperatures).
It's this resemblance to real things that gives objects much of their power and appeal. They can not only model components of real systems, but equally as well fulfill assigned roles as components in software systems.
To invent programs, you need to be able to capture the same kinds of abstractions and express them in the program design.
It's the job of a programming language to help you do this. The language should facilitate the process of invention and design by letting you encode abstractions that reveal the way things work. It should let you make your ideas concrete in the code you write. Surface details shouldn't obscure the architecture of your program.
All programming languages provide devices that help express abstractions. In essence, these devices are ways of grouping implementation details, hiding them, and giving them, at least to some extent, a common interface--much as a mechanical object separates its interface from its implementation.
Looking at such a unit from the inside, as the implementor, you'd be concerned with what it's composed of and how it works. Looking at it from the outside, as the user, you're concerned only with what it is and what it does. You can look past the details and think solely in terms of the role that the unit plays at a higher level.
The principal units of abstraction in the C language are structures and functions. Both, in different ways, hide elements of the implementation:
Object-oriented programming languages don't lose any of the virtues of structures and functions. But they go a step further and add a unit capable of abstraction at a higher level, a unit that hides the interaction between a function and its data.
Suppose, for example, that you have a group of functions that all act on a particular data structure. You want to make those functions easier to use by, as far as possible, taking the structure out of the interface. So you supply a few additional functions to manage the data. All the work of manipulating the data structure--allocating it, initializing it, getting information from it, modifying values within it, keeping it up to date, and freeing it--is done through the functions. All the user does is call the functions and pass the structure to them.
With these changes, the structure has become an opaque token that other programmers never need to look inside. They can concentrate on what the functions do, not how the data is organized. You've taken the first step toward creating an object.
The next step is to give this idea support in the programming language and completely hide the data structure so that it doesn't even have to be passed between the functions. The data becomes an internal implementation detail; all that's exported to users is a functional interface. Because objects completely encapsulate their data (hide it), users can think of them solely in terms of their behavior.
With this step, the interface to the functions has become much simpler. Callers don't need to know how they're implemented (what data they use). It's fair now to call this an ``object.''
The hidden data structure unites all of the functions that share it. So an object is more than a collection of random functions; it's a bundle of related behaviors that are supported by shared data. To use a function that belongs to an object, you first create the object (thus giving it its internal data structure), then tell the object which function it should invoke. You begin to think in terms of what the object does, rather than in terms of the individual functions.
This progression from thinking about functions and data structures to thinking about object behaviors is the essence of object-oriented programming. It may seem unfamiliar at first, but as you gain experience with object-oriented programming, you'll find it's a more natural way to think about things. Everyday programming terminology is replete with analogies to real-world objects of various kinds--lists, containers, tables, controllers, even managers. Implementing such things as programming objects merely extends the analogy in a natural way.
A programming language can be judged by the kinds of abstractions that it enables you to encode. You shouldn't be distracted by extraneous matters or forced to express yourself using a vocabulary that doesn't match the reality you're trying to capture.
If, for example, you must always tend to the business of keeping the right data matched with the right procedure, you're forced at all times to be aware of the entire program at a low level of implementation. While you might still invent programs at a high level of abstraction, the path from imagination to implementation can become quite tenuous--and more and more difficult as programs become bigger and more complicated.
By providing another, higher level of abstraction, object-oriented programming languages give you a larger vocabulary and a richer model to program in.
Likely, if you've ever tackled any kind of difficult programming problem, your design has included groups of functions that work on a particular kind of data--implicit ``objects'' without the language support. Object-oriented programming makes these function groups explicit and permits you to think in terms of the group, rather than its components. The only way to an object's data, the only interface, is through its methods.
By combining both state and behavior in a single unit, an object becomes more than either alone; the whole really is greater than the sum of its parts. An object is a kind of self-sufficient ``subprogram'' with jurisdiction over a specific functional area. It can play a full-fledged modular role within a larger program design.
Clearly, a programmatic Faucet can be smarter than a real one (it's analogous to a mechanical faucet with lots of gauges and instruments attached). But even a real faucet, like any system component, exhibits both state and behavior. To effectively model a system, you need programming units, like objects, that also combine state and behavior.
A program consists of a network of interconnected objects that call upon each other to solve a part of the puzzle. Each object has a specific role to play in the overall design of the program and is able to communicate with other objects. Objects communicate through messages, requests to perform a method.
The objects in the network won't all be the same. For example, in addition to Faucets, the program that models water usage might also have WaterPipe objects that can deliver water to the Faucets and Valve objects to regulate the flow among WaterPipes. There could be a Building object to coordinate a set of WaterPipes, Valves, and Faucets, some Appliance objects--corresponding to dishwashers, toilets, and washing machines--that can turn Valves on and off, and maybe some Users to work the Appliances and Faucets. When a Building object is asked how much water is being used, it might call upon each Faucet and Valve to report its current state. When a User starts up an Appliance, the Appliance will need to turn on a Valve to get the water it requires.
There's a tendency, for example, to think of objects as ``actors'' and to endow them with human-like intentions and abilities. It's tempting sometimes to talk about an object ``deciding'' what to do about a situation, ``asking'' other objects for information, ``introspecting'' about itself to get requested information, ``delegating'' responsibility to another object, or ``managing'' a process.
Rather than think in terms of functions or methods doing the work, as you would in a procedural programming language, this metaphor asks you to think of objects as ``performing'' their methods. Objects are not passive containers for state and behavior, but are said to be the agents of the program's activity.
This is actually a useful metaphor. An object is like an actor in a couple of respects: It has a particular role to play within the overall design of the program, and within that role it can act fairly independently of the other parts of the program. It interacts with other objects as they play their own roles, but is self-contained and to a certain extent can act on its own. Like an actor on stage, it can't stray from the script, but the role it plays it can be multi-faceted and quite complex.
The idea of objects as actors fits nicely with the principal metaphor of object-oriented programming--the idea that objects communicate through ``messages.'' Instead of calling a method as you would a function, you send a message to an object requesting it to perform one of its methods.
Although it can take some getting used to, this metaphor leads to a useful way of looking at methods and objects. It abstracts methods away from the particular data they act on and concentrates on behavior instead. For example, in an object-oriented programming interface, a start method might initiate an operation, an archive method might archive information, and a draw method might produce an image. Exactly which operation is initiated, which information is archived, and which image is drawn isn't revealed by the method name. Different objects might perform these methods in different ways.
Thus, methods are a vocabulary of abstract behaviors. To invoke one of those behaviors, you have to make it concrete by associating the method with an object. This is done by naming the object as the ``receiver'' of a message. The object you choose as receiver will determine the exact operation that's initiated, the data that's archived, or the image that's drawn.
Since methods belong to objects, they can be invoked only through a particular receiver (the owner of the method and of the data structure the method will act on). Different receivers can have different implementations of the same method, so different receivers can do different things in response to the same message. The result of a message can't be calculated from the message or method name alone; it also depends on the object that receives the message.
By separating the message (the requested behavior) from the receiver (the owner of a method that can respond to the request), the messaging metaphor perfectly captures the idea that behaviors can be abstracted away from their particular implementations.
In this, objects are similar to C structures. Declaring a structure defines a type. For example, this declaration
struct key {
char *word;
int count;
};
defines the struct key type. Once defined, the structure name can be used to produce any number of instances of the type:struct key a, b, c, d;The declaration is a template for a kind of structure, but it doesn't create a structure that the program can use. It takes another step to allocate memory for an actual structure of that type, a step that can be repeated any number of times.
struct key *p = malloc(sizeof(struct key) * MAXITEMS);
Similarly, defining an object creates a template for a kind of object. It defines a class of objects. The template can be used to produce any number of similar objects--instances of the class. For example, there would be a single definition of the Faucet class. Using this definition, a program could allocate as many Faucet instances as it needed.
A class definition is like a structure definition in that it lays out an arrangement of data elements (instance variables) that become part of every instance. Each instance has memory allocated for its own set of instance variables, which store values peculiar to the instance.
However, a class definition differs from a structure declaration in that it also includes methods that specify the behavior of class members. Every instance is characterized by its access to the methods defined for the class. Two objects with equivalent data structures but different methods would not belong to the same class.
This kind of module is a unit defined by the file system. It's a container for source code, not a logical unit of the language. What goes into the container is up to each programmer. You can use them to group logically related parts of the code, but you don't have to. Files are like the drawers of a dresser; you can put your socks in one drawer, underwear in another, and so on, or you can use another organizing scheme or simply choose to mix everything up.
In Objective-C, for example, it would be possible to define the part of the Valve class that interacts with WaterPipes in the same file that defines the WaterPipe class, thus creating a container module for WaterPipe-related code and splitting Valve class into more than one file. The Valve class definition would still act as a modular unit within the construction of the program--it would still be a logical module--no matter how many files the source code was located in.
The mechanisms that make class definitions logical units of the language are discussed in some detail under ``MECHANISMS OF ABSTRACTION'' below.
Reusability is influenced by a variety of different factors, including:
Nevertheless, a general comparison would show that class definitions lend themselves to reusable code in ways that functions do not. There are various things you can do to make functions more reusable--passing data as arguments rather than assuming specifically-named global variables, for example. Even so, it turns out that only a small subset of functions can be generalized beyond the applications they were originally designed for. Their reusability is inherently limited in at least three ways:
Moreover, because an object's data is hidden, a class can be reimplemented to use a different data structure without affecting its interface. All programs that use the class can pick up the new version without changing any source code; no reprogramming is required.
Encapsulation keeps the implementation of an object out of its interface, and polymorphism results from giving each class its own name space. The following sections discuss each of these mechanisms in turn.
In C, a function is clearly encapsulated; its implementation is inaccessible to other parts of the program and protected from whatever actions might be taken outside the body of the function. Method implementations are similarly encapsulated, but, more importantly, so are an object's instance variables. They're hidden inside the object and invisible outside it. The encapsulation of instance variables is sometimes also called information hiding.
It might seem, at first, that hiding the information in instance variables would constrain your freedom as a programmer. Actually, it gives you more room to act and frees you from constraints that might otherwise be imposed. If any part of an object's implementation could leak out and become accessible or a concern to other parts of the program, it would tie the hands both of the object's implementor and of those who would use the object. Neither could make modifications without first checking with the other.
Suppose, for example, that you're interested in the Faucet object being developed for the program that models water use and you want to incorporate it in another program you're writing. Once the interface to the object is decided, you don't have to be concerned as others work on it, fix bugs, and find better ways to implement it. You'll get the benefit of these improvements, but none of them will affect what you do in your program. Because you're depending solely on the interface, nothing they do can break your code. Your program is insulated from the object's implementation.
Moreover, although those implementing the Faucet object would be interested in how you're using the class and might try to make sure that it meet your needs, they don't have to be concerned with the way you're writing your code. Nothing you do can touch the implementation of the object or limit their freedom to make changes in future releases. The implementation is insulated from anything that you or other users of the object might do.
Polymorphism results from the fact that every class lives in its own name space. The names assigned within a class definition won't conflict with names assigned anywhere outside it. This is true both of the instance variables in an object's data structure and of the object's methods:
The main benefit of polymorphism is that it simplifies the programming interface. It permits conventions to be established that can be reused in class after class. Instead of inventing a new name for each new function you add to a program, the same names can be reused. The programming interface can be described as a set of abstract behaviors, quite apart from the classes that implement them.
Polymorphism also permits code to be isolated in the methods of different objects rather than be gathered in a single function that enumerates all the possible cases. This makes the code you write more extensible and reusable. When a new case comes along, you don't have to reimplement existing code, but only add a new class with a new method, leaving the code that's already written alone.
For example, suppose you have code that sends a draw message to an object. Depending on the receiver, the message might produce one of two possible images. When you want to add a third case, you don't have to change the message or alter existing code, but merely allow another object to be assigned as the message receiver.
The same is true if want to define a new kind of object; the description is simpler if it can start from the definition of an existing object.
With this in mind, object-oriented programming languages permit you to base a new class definition on a class already defined. The base class is called a superclass; the new class is its subclass. The subclass definition specifies only how it differs from the superclass; everything else is taken to be the same.
Nothing is copied from superclass to subclass. Instead, the two classes are connected so that the subclass inherits all the methods and instance variables of its superclass, much as you want your listener's understanding of ``schooner'' to inherit what they already know about sailboats. If the subclass definition were empty (if it didn't define any instance variables or methods of its own), the two classes would be identical (except for their names) and share the same definition. It would be like explaining what a ``fiddle'' is by saying that it's exactly the same as a ``violin.'' However, the reason for declaring a subclass isn't to generate synonyms, but to create something at least a little different from its superclass. You'd want to let the fiddle play bluegrass in addition to classical music.
As the above figure shows, every inheritance hierarchy begins with a root class that has no superclass. From the root class, the hierarchy branches downward. Each class inherits from its superclass, and through its superclass, from all the classes above it in the hierarchy. Every class inherits from the root class.
Each new class is the accumulation of all the class definitions in its inheritance chain. In the example above, class D inherits both from C, its superclass, and the root class. Members of the D class will have methods and instance variables defined in all three classes--D, C, and root.
Typically, every class has just one superclass and can have an unlimited number of subclasses. However, in some object-oriented programming languages (though not in Objective-C), a class can have more than one superclass; it can inherit through multiple sources. Instead of a single hierarchy that branches downward as shown in the above figure, multiple inheritance lets some branches of the hierarchy (or of different hierarchies) merge.
There's rarely a reason to program a taxonomy like this, but the analogy is a good one. Subclasses tend to specialize a superclass or adapt it to a special purpose, much as a species specializes a genus.
Here are some typical uses of inheritance:
In hindsight, it's evident what a serious constraint this was. It limited not only how programs were constructed, but what you could imagine a program doing. It constrained design, not just programming technique. Functions (like malloc()) that dynamically allocate memory as a program runs opened possibilities that didn't exist before.
Compile-time and link-time constraints are limiting because they force issues to be decided from information found in the programmer's source code, rather than from information obtained from the user as the program runs.
Although dynamic allocation removes one such constraint, many others, equally as limiting as static memory allocation, remain. For example, the elements that make up an application must be matched to data types at compile time. And the boundaries of an application are typically set at link time. Every part of the application must be united in a single executable file. New modules and new types can't be introduced as the program runs.
Object-oriented programming seeks to overcome these limitations and to make programs as dynamic and fluid as possible. It shifts much of the burden of decision making from compile time and link time to run time. The goal is to let program users decide what will happen, rather than constrain their actions artificially by the demands of the language and the needs of the compiler and linker.
Three kinds of dynamism are especially important for object-oriented design:
incompatible types in assignmentType checking is useful, but there are times when it can interfere with the benefits you get from polymorphism, especially if the type of every object must be known to the compiler.
assignment of integer from pointer lacks a cast
Suppose, for example, that you want to send an object a message to perform the start method. Like other data elements, the object is represented by a variable. If the variable's type (its class) must be known at compile time, it would be impossible to let run-time factors influence the decision about what kind of object should be assigned to the variable. If the class of the variable is fixed in source code, so is the version of start that the message invokes.
If, on the other hand, it's possible to wait until run time to discover the class of the variable, any kind of object could be assigned to it. Depending on the class of the receiver, the start message might invoke different versions of the method and produce very different results.
Dynamic typing thus gives substance to dynamic binding (discussed next). But it does more than that. It permits associations between objects to be determined at run time, rather than forcing them to be encoded in a static design. For example, a message could pass an object as an argument without declaring exactly what kind of object it is--that is, without declaring its class. The message receiver might then send its own messages to the object, again without ever caring about what kind of object it is. Because the receiver uses the object it's passed to do some of its work, it is in a sense customized by an object of indeterminate type (indeterminate in source code, that is, not at run time).
int strcmp(const char *, const char *); /* case sensitive */and declare a pointer to a function that has the same return and argument types:
int strcasecmp(const char *, const char *); /* case insensitive*/
int (* compare)(const char *, const char *);You can then wait until run time to determine which function to assign to the pointer,
if ( **argv == 'i' )and call the function through the pointer:
compare = strcasecmp;
else
compare = strcmp;
if ( compare(s1, s2) )This is akin to what in object-oriented programming is called dynamic binding, delaying the decision of exactly which method to perform until the program is running.
. . .
Although not all object-oriented languages support it, dynamic binding can be routinely and transparently accomplished through messaging. You don't have to go through the indirection of declaring a pointer and assigning values to it as shown in the example above. You also don't have to assign each alternative procedure a different name.
Messages invoke methods indirectly. Every message expression must find a method implementation to ``call.'' To find that method, the messaging machinery must check the class of the receiver and locate its implementation of the method named in the message. When this is done at run time, the method is dynamically bound to the message. When it's done by the compiler, the method is statically bound.
However, if the class of the receiver is dynamically typed, there's no way for the compiler to determine which method to invoke. The method can be found only after the class of the receiver is resolved at run time. Dynamic typing thus entails dynamic binding.
Dynamic typing also makes dynamic binding interesting, for it opens the possibility that a message might have very different results depending on the class of the receiver. Run-time factors can influence the choice of receiver and the outcome of the message.
Dynamic typing and binding also open the possibility that the code you write can send messages to objects not yet invented. If object types don't have to be decided until run time, you can give others the freedom to design their own classes and name their own data types, and still have your code send messages to their objects. All you need to agree on are the messages, not the data types.
Note: Dynamic binding is routine in Objective-C. You don't need to arrange for it specially, so your design never needs to bother with what's being done when.
Some object-oriented programming environments overcome this constraint and allow different parts of an executable program to be kept in different files. The program can be launched in bits and pieces as they're needed. Each piece is dynamically loaded and linked with the rest of program as it's launched. User actions can determine which parts of the program are in memory and which aren't.
Only the core of a large program needs to be loaded at the start. Other modules can be added as the user requests their services. Modules the user doesn't request make no memory demands on the system.
Dynamic loading raises interesting possibilities. For example, an entire program wouldn't have to be developed at once. You could deliver your software in pieces and update one part of it at a time. You could devise a program that groups many different tools under a single interface, and load just the tools the user wants. The program could even offer sets of alternative tools to do the same job. The user would select one tool from the set and only that tool would be loaded. It's not hard to imagine the possibilities. But because dynamic loading is relatively new, it's harder to predict its eventual benefits.
Perhaps the most important current benefit of dynamic loading is that it makes applications extensible. You can allow others to add to and customize a program you've designed. All your program needs to do is provide a framework that others can fill in, then at run time find the pieces that they've implemented and load them dynamically.
For example, in the OPENSTEP environment, Interface Builder dynamically loads custom palettes and inspectors, and the Workspace Manager(tm) dynamically loads inspectors for particular file formats. Anyone can design their own custom palettes and inspectors that these applications will load and incorporate into themselves.
The main challenge that dynamic loading faces is getting a newly loaded part of a program to work with parts already running, especially when the different parts were written by different people. However, much of this problem disappears in an object-oriented environment because code is organized into logical modules with a clear division between implementation and interface. When classes are dynamically loaded, nothing in the newly loaded code can clash with the code already in place. Each class encapsulates its implementation and has an independent name space.
In addition, dynamic typing and dynamic binding let classes designed by others fit effortlessly into the program you've designed. Once a class is dynamically loaded, it's treated no differently than any other class. Your code can send messages to their objects and theirs to yours. Neither of you has to know what classes the other has implemented. You need only agree on a communications protocol.
Some connections can be entirely transitory. A message might contain an argument identifying an object, perhaps the sender of the message, that the receiver can communicate with. As it responds to the message, the receiver can send messages to that object, perhaps identifying itself or still another object that that object can in turn communicate with. Such connections are fleeting; they last only as long as the chain of messages.
But not all connections between objects can be handled on the fly. Some need to be recorded in program data structures. There are various ways to do this. A table might be kept of object connections, or there might be a service that identifies objects by name. However, the simplest way is for each object to have instance variables that keep track of the other objects it must communicate with. These instance variables--termed outlets because they record the outlets for messages--define the principal connections between objects in the program network.
Although the names of outlet instance variables are arbitrary, they generally reflect the roles that outlet objects play. The figure below illustrates an object with four outlets--an ``agent,'' a ``friend,'' a ``neighbor,'' and a ``boss.'' The objects that play these parts may change every now and then, but the roles remain the same.
Some outlets are set when the object is first initialized and may never change. Others might be set automatically as the consequence of other actions. Still other can be set freely, using methods provided just for that purpose.
However they're set, outlet instance variables reveal the structure of the application. They link objects into a communicating network, much as the components of a water system are linked by their physical connections or as individuals are linked by their patterns of social relations.
Sometimes one object should be seen as being part of another. For example, a Faucet might use a Meter object to measure the amount of water being released. The Meter would serve no other object and would act only under orders from the Faucet. It would be an intrinsic part of the Faucet, in contrast to an Appliance's extrinsic connection to a Valve.
Similarly, an object that oversees other objects might keep a list of its charges. A Building object, for example, might have a list of all the WaterPipes in the program. The WaterPipes would be considered an intrinsic part of the Building and belong to it. WaterPipes, on the other hand, would maintain extrinsic connections to each other.
Intrinsic outlets behave differently than extrinsic ones. When an object is freed or archived in a file on disk, the objects that its intrinsic outlets point to must be freed or archived with it. For example, when a Faucet is freed, its Meter is rendered useless and therefore should be freed as well. A Faucet that was archived without its Meter would be of little use when it was unarchived again (unless it could create a new Meter for itself).
Extrinsic outlets, on the other hand, capture the organization of the program at a higher level. They record connections between relatively independent program subcomponents. When an Appliance is freed, the Valve it was connected to still is of use and remains in place. When an Appliance is unarchived, it can be connected to another Valve and resume playing the same sort of role it played before.
Object-oriented programs often are activated by a flow of events, reports of external activity of some sort. Applications that display a user interface are driven by events from the keyboard and mouse. Every touch of a key or click of the mouse generates events that the application receives and responds to. An object-oriented program structure (a network of objects that's prepared to respond to an external stimulus) is ideally suited for this kind of user-driven application.
Nevertheless, the question often arises of whether to add more functionality to a class or to factor out the additional functionality and put it in an separate class definition. For example, a Faucet needs to keep track of how much water is being used over time. To do that, you could either implement the necessary methods in the Faucet class, or you could devise a generic Meter object to do the job, as suggested earlier. Each Faucet would have an outlet connecting it to a Meter, and the Meter would not interact with any object but the Faucet.
The choice often depends on your design goals. If the Meter object could be used in more than one situation, perhaps in another project entirely, it would increase the reusability of your code to factor the metering task into a separate class. If you have reason to make Faucet objects as self-contained as possible, the metering functionality could be added to the Faucet class.
It's generally better to try to for reusable code and avoid having large classes that do so many things that they can't be adapted to other situations. When objects are designed as components, they become that much more reusable. What works in one system or configuration might well work in another.
Dividing functionality between different classes doesn't necessarily complicate the programming interface. If the Faucet class keeps the Meter object private, the Meter interface wouldn't have to be published for users of the Faucet class; the object would be as hidden as any other intrinsic Faucet instance variable.
When you design an object-oriented program, you are, in effect, putting together a computer simulation of how something works. Object networks look and behave like models of real systems. An object-oriented program can be thought of as a model, even if there's no actual counterpart to it in the real world.
Each component of the model--each kind of object--is described in terms of its behavior and responsibilities and its interactions with other components. Because an object's interface lies in its methods, not its data, you can begin the design process by thinking about what a system component will do, not how it's represented in data. Once the behavior of an object is decided, the appropriate data structure can be chosen, but this is a matter of implementation, not the initial design.
For example, in the water-use program, you wouldn't begin by deciding what the Faucet data structure looked like, but what you wanted a Faucet to do--make a connection to a WaterPipe, be turned on and off, adjust the rate of flow, and so on. The design is therefore not bound from the outset by data choices. You can decide on the behavior first, and implement the data afterwards. Your choice of data structures can change over time without affecting the design.
Designing an object-oriented program doesn't necessarily entail writing great amounts of code. The reusability of class definitions means that the opportunity is great for building a program largely out of classes devised by others. It might even be possible to construct interesting programs entirely out of classes someone else defined. As the suite of class definitions grows, you have more and more reusable parts to choose from.
Reusable classes come from many sources. Development projects often yield reusable class definitions, and some enterprising developers have begun marketing them. Object-oriented programming environments typically come with class libraries. There are well over a hundred classes in the OPENSTEP libraries. Some of these classes offer basic services (hashing, data storage, remote messaging). Others are more specific (user interface devices, video displays, a sound editor).
Typically, a group of library classes work together to define a partial program structure. These classes constitute a software framework (or kit) that can be used to build a variety of different kinds of applications. When you use a framework, you accept the program model it provides and adapt your design to it. You use the framework by:
The framework, in essence, sets up part of a object network for your program and provides part of its class hierarchy. Your own code completes the program model started by the framework.
As software tries to do more and more, and programs become bigger and more complicated, the problem of managing the task also grows. There are more pieces to fit together and more people working together to build them. The object-oriented approach offers ways of dealing with this complexity, not just in design, but also in the organization of the work.
The sheer size of the effort and the number of people working on the same project at the same time in the same place can get in the way of the group's ability to work cooperatively towards a common goal. In addition, collaboration is often impeded by barriers of time, space, and organization.
That's where object-oriented programming techniques can help. For example, the reusability of object-oriented code means that programmers can collaborate effectively even when they work on different projects at different times or are in different organizations, just by sharing their code in libraries. This kind of collaboration holds a great deal of promise, for it can conceivably lighten difficult tasks and bring impossible projects into the realm of possibility.
With an object-oriented design, it's easier to keep common goals in sight, instead of losing them in the implementation, and easier for everyone to see how the piece they're working on fits into the whole. Their collaborative efforts are therefore more likely to be on target.
During implementation, only this interface needs to be coordinated, and most of that falls naturally out of the design. Since each class encapsulates its implementation and has its own name space, there's no need to coordinate implementation details. Collaboration is simpler when there are fewer coordination requirements. The difficulties that are avoided are the easiest ones to manage.
This has benefits, not just for organizing the implementation, but for fixing problems later. Since implementations are contained within class boundaries, problems that come up are also likely to be isolated. It's easier to track down bugs when they're located in a well-defined part of the program.
Separating responsibilities by class also means that each part can be worked on by specialists. Classes can be updated periodically to optimize their performance and make the best use of new technologies. These updates don't have to be coordinated with other parts of the program. As long as the interface to an object doesn't change, improvements to its implementation can be scheduled at any time.
Inheritance also increases the reusability and reliability of code. The code placed in a superclass is tested by all its subclasses. The generic class you find in a library will have been tested by other subclasses written by other developers for other applications.
Classes and frameworks from an object-oriented library can make substantial contributions to your program. When you program with the OPENSTEP software frameworks, for example, you're effectively collaborating with the programmers at NeXT; you're contracting a part of your program, often a substantial part, to them. You can concentrate on what you do best and leave other tasks to the library developer. Your projects can be prototyped faster, completed faster, with less of a collaborative challenge at your own site.
The increased reusability of object-oriented code also increases its reliability. A class taken from a library is likely to have found its way into a variety of different applications and situations. The more the code has been used, the more likely it is that problems will have been encountered and fixed. Bugs that would have seemed strange and hard to find in your program might already have been tracked down and eliminated.
Table of Contents
Next Chapter
