Design Patterns


Design Patterns: Elements of Reusable Object-Oriented Software is a software engineering book describing software design patterns. The book was written by Erich Gamma, Richard Helm, Ralph Johnson, and John Vlissides, with a foreword by Grady Booch. The book is divided into two parts, with the first two chapters exploring the capabilities and pitfalls of object-oriented programming, and the remaining chapters describing 23 classic software design patterns. The book includes examples in C++ and Smalltalk.
It has been influential to the field of software engineering and is regarded as an important source for object-oriented design theory and practice. More than 500,000 copies have been sold in English and in 13 other languages. The authors are often referred to as the Gang of Four.

History

The book started at a birds of a feather session at OOPSLA '90, "Towards an Architecture Handbook", run by Bruce Anderson, where Erich Gamma and Richard Helm met and discovered their common interest. They were later joined by Ralph Johnson and John Vlissides. The original publication date of the book was October 21, 1994 with a 1995 copyright, hence it is often cited with a 1995-year, despite being published in 1994. The book was first made available to the public at the OOPSLA meeting held in Portland, Oregon, in October 1994.
In 2005 the ACM SIGPLAN awarded that year's Programming Languages Achievement Award to the authors, in recognition of the impact of their work "on programming practice and programming language design". As of March 2012, the book was in its 40th printing.

Introduction, Chapter 1

Chapter 1 is a discussion of object-oriented design techniques, based on the authors' experience, which they believe would lead to good object-oriented software design, including:
The authors claim the following as advantages of interfaces over implementation:
Use of an interface also leads to dynamic binding and polymorphism, which are central features of object-oriented programming.
The authors refer to inheritance as white-box reuse, with
white-box referring to visibility, because the internals of parent classes are often visible to subclasses. In contrast, the authors refer to object composition as black-box reuse because no internal details of composed objects need be visible in the code using them.
The authors discuss the tension between inheritance and encapsulation at length and state that in their experience, designers overuse inheritance. The danger is stated as follows:
They warn that the implementation of a subclass can become so bound up with the implementation of its parent class that any change in the parent's implementation will force the subclass to change. Furthermore, they claim that a way to avoid this is to inherit only from abstract classes—but then, they point out that there is minimal code reuse.
Using inheritance is recommended mainly when adding to the functionality of existing components, reusing most of the old code and adding relatively small amounts of new code.
To the authors, 'delegation' is an extreme form of object composition that can always be used to replace inheritance. Delegation involves two objects: a 'sender' passes itself to a 'delegate' to let the delegate refer to the sender. Thus the link between two parts of a system are established only at runtime, not at compile-time. The Callback article has more information about delegation.
The authors also discuss so-called parameterized types, which are also known as generics or templates. These allow any type to be defined without specifying all the other types it uses—the unspecified types are supplied as 'parameters' at the point of use.
The authors admit that delegation and parameterization are very powerful but add a warning:
The authors further distinguish between 'Aggregation', where one object 'has' or 'is part of' another object and acquaintance, where one object merely 'knows of' another object. Sometimes acquaintance is called 'association' or the 'using' relationship. Acquaintance objects may request operations of each other, but they aren't responsible for each other. Acquaintance is a weaker relationship than aggregation and suggests much looser coupling between objects, which can often be desirable for maximum maintainability in a design.
The authors employ the term 'toolkit' where others might today use 'class library', as in C# or Java. In their parlance, toolkits are the object-oriented equivalent of subroutine libraries, whereas a 'framework' is a set of cooperating classes that make up a reusable design for a specific class of software. They state that applications are hard to design, toolkits are harder, and frameworks are the hardest to design.

Case study, Chapter 2

Chapter 2 is a step-by-step case study on "the design of a 'What-You-See-Is-What-You-Get' document editor called Lexi."
The chapter goes through seven problems that must be addressed in order to properly design Lexi, including any constraints that must be followed. Each problem is analyzed in depth, and solutions are proposed. Each solution is explained in full, including pseudo-code and a slightly modified version of Object Modeling Technique where appropriate.
Finally, each solution is associated directly with one or more design patterns. It is shown how the solution is a direct implementation of that design pattern.
The seven problems and their solutions, are as follows:

Document Structure

The document is "an arrangement of basic graphical elements" such as characters, lines, other shapes, etc., that "capture the total information content of the document". The structure of the document contains a collection of these elements, and each element can in turn be a substructure of other elements.
Problems and Constraints
  1. Text and graphics should be treated the same way
  2. The implementation should treat complex and simple structures the same way. It should not have to know the difference between the two.
  3. Specific derivatives of abstract elements should have specialized analytical elements.
Solution and Pattern
A recursive composition is a hierarchical structure of elements, that builds "increasingly complex elements out of simpler ones". Each node in the structure knows of its own children and its parent. If an operation is to be performed on the whole structure, each node calls the operation on its children.
This is an implementation of the composite pattern, which is a collection of nodes. The node is an abstract base class, and derivatives can either be leaves, or collections of other nodes. When an operation is performed on the parent, that operation is recursively passed down the hierarchy.

Formatting

Formatting differs from structure. Formatting is a method of constructing a particular instance of the document's physical structure. This includes breaking text into lines, using hyphens, adjusting for margin widths, etc.
Problems and Constraints
  1. Balance between quality, speed and storage space
  2. Keep formatting independent from the document structure.
Solution and Pattern
A Compositor class will encapsulate the algorithm used to format a composition. Compositor is a subclass of the primitive object of the document's structure. A Compositor has an associated instance of a Composition object. When a Compositor runs its Compose, it iterates through each element of its associated Composition, and rearranges the structure by inserting Row and Column objects as needed.
The Compositor itself is an abstract class, allowing for derivative classes to use different formatting algorithms
The Strategy Pattern is used to accomplish this goal. A Strategy is a method of encapsulating multiple algorithms to be used based on a changing context. In this case, formatting should be different, depending on whether text, graphics, simple elements, etc., are being formatted.

Embellishing the User Interface

The ability to change the graphical interface that the user uses to interact with the document.
Problems and Constraints
  1. Demarcate a page of text with a border around the editing area
  2. Scroll bars that let the user view different parts of the page
  3. User interface objects should not know about the embellishments
  4. Avoid an "explosion of classes" that would be caused by subclassing for "every possible combination of embellishments" and elements
Solution and Pattern
The use of a transparent enclosure allows elements that augment the behaviour of composition to be added to a composition. These elements, such as Border and Scroller, are special subclasses of the singular element itself. This allows the composition to be augmented, effectively adding state-like elements. Since these augmentations are part of the structure, their appropriate Operation will be called when the structure's Operation is called. This means that the client does not need any special knowledge or interface with the structure in order to use the embellishments.
This is a Decorator pattern, one that adds responsibilities to an object without modifying the object itself.

Supporting Multiple Look-And-Feel Standards

refers to platform-specific UI standards. These standards "define guidelines for how applications appear and react to the user".
Problems and Constraints
  1. The editor must implement standards of multiple platforms so that it is portable
  2. Easily adapt to new and emergent standards
  3. Allow for run-time changing of look-and-feel
  4. Have a set of abstract elemental subclasses for each category of elements
  5. Have a set of concrete subclasses for each abstract subclass that can have a different look-and-feel standard.
Solution and Pattern
Since object creation of different concrete objects cannot be done at runtime, the object creation process must be abstracted. This is done with an abstract guiFactory, which takes on the responsibility of creating UI elements. The abstract guiFactory has concrete implementations, such as MotifFactory, which creates concrete elements of the appropriate type. In this way, the program need only ask for a ScrollBar and, at run-time, it will be given the correct concrete element.
This is an Abstract Factory. A regular factory creates concrete objects of one type. An abstract factory creates concrete objects of varying types, depending on the concrete implementation of the factory itself. Its ability to focus on not just concrete objects, but entire families of concrete objects "distinguishes it from other creational patterns, which involve only one kind of product object".

Supporting Multiple Window Systems

Just as look-and-feel is different across platforms, so is the method of handling windows. Each platform displays, lays out, handles input to and output from, and layers windows differently.
Problems and Constraints
  1. The document editor must run on many of the "important and largely incompatible window systems" that exist
  2. An Abstract Factory cannot be used. Due to differing standards, there will not be a common abstract class for each type of widget.
  3. Do not create a new, nonstandard windowing system
Solution and Pattern
It is possible to develop "our own abstract and concrete product classes", because "all window systems do generally the same thing". Each window system provides operations for drawing primitive shapes, iconifying/de-iconifying, resizing, and refreshing window contents.
An abstract base Window class can be derived to the different types of existing windows, such as application, iconified, dialog. These classes will contain operations that are associated with windows, such as reshaping, graphically refreshing, etc. Each window contains elements, whose Draw functions are called upon by the Window's own draw-related functions.
In order to avoid having to create platform-specific Window subclasses for every possible platform, an interface will be used. The Window class will implement a Window implementation abstract class. This class will then in turn be derived into multiple platform-specific implementations, each with platform-specific operations. Hence, only one set of Window classes are needed for each type of Window, and only one set of WindowImp classes are needed for each platform. In addition, adding a new window type does not require any modification of platform implementation, or vice versa.
This is a Bridge pattern. Window and WindowImp are different, but related. Window deals with windowing in the program, and WindowImp deals with windowing on a platform. One of them can change without ever having to modify the other. The Bridge pattern allows these two "separate class hierarchies to work together even as they evolve independently".

User Operations

All actions the user can take with the document, ranging from entering text, changing formatting, quitting, saving, etc.
Problems and Constraints
  1. Operations must be accessed through different inputs, such as a menu option and a keyboard shortcut for the same command
  2. Each option has an interface, which should be modifiable
  3. Operations are implemented in several different classes
  4. In order to avoid coupling, there must not be a lot of dependencies between implementation and user interface classes.
  5. Undo and redo commands must be supported on most document changing operations, with no arbitrary limit on the number of levels of undo
  6. Functions are not viable, since they don't undo/redo easily, are not easily associated with a state, and are hard to extend or reuse.
  7. Menus should be treated like hierarchical composite structures. Hence, a menu is a menu item that contains menu items which may contain other menu items, etc.
Solution and Pattern
Each menu item, rather than being instantiated with a list of parameters, is instead done with a Command object.
Command is an abstract object that only has a single abstract Execute method. Derivative objects extend the Execute method appropriately. These objects can be used by widgets or buttons just as easily as they can be used by menu items.
To support undo and redo, Command is also given Unexecute and Reversible. In derivative classes, the former contains code that will undo that command, and the latter returns a boolean value that defines if the command is undoable. Reversible allows some commands to be non-undoable, such as a Save command.
All executed Commands are kept in a list with a method of keeping a "present" marker directly after the most recently executed command. A request to undo will call the Command.Unexecute directly before "present", then move "present" back one command. Conversely, a Redo request will call Command.Execute after "present", and move "present" forward one.
This Command approach is an implementation of the Command pattern. It encapsulates requests in objects, and uses a common interface to access those requests. Thus, the client can handle different requests, and commands can be scattered throughout the application.

Spell Check and Hyphenation

This is the document editor's ability to textually analyze the contents of a document. Although there are many analyses that can be performed, spell check and hyphenation-formatting are the focus.
Problems and Constraints
  1. Allow for multiple ways to check spelling and identify places for hyphenation
  2. Allow for expansion for future analysis
  3. Be able to iterate through a text's contents without access to the text's actual structure
  4. Allow for any manner of traversal of document
Solution and Pattern
Removing the integer-based index from the basic element allows for a different iteration interface to be implemented. This will require extra methods for traversal and object retrieval. These methods are put into an abstract Iterator interface. Each element then implements a derivation of the Iterator, depending on how that element keeps its list.
Functions for traversal and retrieval are put into the abstract Iterator interface. Future Iterators can be derived based on the type of list they will be iterating through, such as Arrays or Linked Lists. Thus, no matter what type of indexing method any implementation of the element uses, it will have the appropriate Iterator.
This is an implementation of the Iterator pattern. It allows the client to traverse through any object collection, without needing to access the contents of the collection directly, or be concerned about the type of list the collection's structure uses.
Now that traversal has been handled, it is possible to analyze the elements of a structure. It is not feasible to build each type of analysis into the element structure themselves; every element would need to be coded, and much of the code would be the same for similar elements.
Instead, a generic CheckMe method is built into the element's abstract class. Each Iterator is given a reference to a specific algorithm. When that Iterator iterates through its collection, it calls each element's CheckMe, passing the specified algorithm. CheckMe then passes a reference to its element back to said algorithm for analysis.
Thus, to perform a spell check, a front-to-end iterator would be given a reference to a SpellCheck object. The iterator would then access each element, executing its CheckMe method with the SpellCheck parameter. Each CheckMe would then call the SpellCheck, passing a reference to the appropriate element.
In this manner, any algorithm can be used with any traversal method, without hard-code coupling one with the other. For example, Find can be used as "find next" or "find previous", depending on if a "forward" iterator was used, or a "backwards" iterator.
In addition, the algorithms themselves can be responsible for dealing with different elements. For example, a SpellCheck algorithm would ignore a Graphic element, rather than having to program every Graphic-derived element to not send themselves to a SpellCheck.

Patterns by type

Creational

s are ones that create objects, rather than having to instantiate objects directly. This gives the program more flexibility in deciding which objects need to be created for a given case.
These concern class and object composition. They use inheritance to compose interfaces and define ways to compose objects to obtain new functionality.
Most of these design patterns are specifically concerned with communication between objects.
Criticism has been directed at the concept of software design patterns generally, and at Design Patterns specifically. A primary criticism of Design Patterns is that its patterns are simply workarounds for missing features in C++, replacing elegant abstract features with lengthy concrete patterns, essentially becoming a "human compiler" or "generating by hand the expansions of some macro". Peter Norvig demonstrates that 16 out of the 23 patterns in Design Patterns are simplified or eliminated in Lisp or Dylan. Related observations were made by Hannemann and Kiczales who implemented several of the 23 design patterns using an aspect-oriented programming language and showed that code-level dependencies were removed from the implementations of 17 of the 23 design patterns and that aspect-oriented programming could simplify the implementations of design patterns.
Paul Graham wrote:
There has also been humorous criticism, such as a show trial at OOPSLA '99 on 3 November 1999, and a parody of the format, by Jim Coplien, entitled "".
In an interview with InformIT in 2009, Erich Gamma stated that the book authors had a discussion in 2005 on how they would have refactored the book and concluded that they would have recategorized some patterns and added a few additional ones. Gamma wanted to remove the Singleton pattern, but there was no consensus among the authors to do so.