1.2. Art or Science?
There is no doubt that patterns are a thriving meme, and one with great utility. Entire academic conferences are now dedicated to patterns, Ackerman and Gonzalez’s patterns-based engineering is becoming a defined discipline in its own right , and industry consultants are now expected to have them under their belt and be able to whip out Unified Modeling Language (UML) diagrams of them on the spot. Tools exist to produce, display, generate, and extract patterns. Patterns, as a collective whole, are an assumed component of the software engineering landscape. We’re just not quite sure how they fit into that landscape or how they fit with each other. Two issues prevent a more comprehensive approach to patterns, and unfortunately they are ubiquitous in the industry. The first is treating patterns as frozen elements to be blindly copied, the second is confusing language-specific pattern implementations with variants of the patterns themselves.
1.2.1. Viewing Patterns as Rote
Ask a dozen developers to define design patterns, and you’ll likely get a dozen answers. Among the more traditional “a solution to a recurring problem within a particular context” answers, you’re also likely to hear phrases such as “a recipe” or “an example structure” or “some sample code,” betraying a rather narrow view of what patterns provide. Patterns are intended to be mutated, to be warped and molded, to meet the needs of the particular forces at play in the context of the problem, but all too often, a developer simply copies and pastes the sample code from a patterns text or site and declares the result a successful application of the pattern. This is usually a recipe for failure instead of a recipe for producing a good design.
Pure rote copying of the structure of the pattern “because this authority says so” is a reversion to Alexander’s concept of unselfconscious design. We undermine the entire purpose of design patterns when we do that. We need to be able to describe the whys behind a pattern as well as the hows. Without the understanding of the reasons that led to the description of that pattern, rote application often results in misapplication. At best, the result is a broken pattern that simply does not match the intended outcome. At worst, it injects an iatrogenic pattern into the system—one that is intended and thought to be of benefit but instead produces a malignant result that may take years to uncover. It doesn’t just fail to provide the expected enhancement, it actively creates a new problem that may be worse than the original one. This is patterns as tribal mythology—action without understanding.
The traditional design pattern form, as defined in Design Patterns , explains the whys behind a pattern—motivations, applicability, and consequences—but it is up to the reader to tease out the underlying concepts that form a pattern. To some degree, these subconcepts are described in the Participants (what are the pieces) and Collaborations (how do they relate) sections for each patterns, but again, these are frequently treated by developers as checklists of pieces of the solution for rote implementation instead of as a description of the underlying concepts and abstractions that comprise a solution.
1.2.2. Language-Dependent Views
Ask a developer how important patterns are to his or her work, and frequently the answer will be based on the implementation language the developer is using. This isn’t surprising. Different languages offer different strengths centered around the concepts they support and how they express them. How those concepts happen to be expressed is more often the start of flame wars between language fans, but ignoring the underlying concepts leads to much argument over nothing of particular consequence in most cases. Whether blocks are delineated by curly braces, as in the C family, or by whitespace, as with Python, isn’t nearly as important as having the concept of blocks in the first place.
What this means is simply that some patterns are easier to implement in some languages than in others. In fact, some languages can make the concepts behind certain patterns so simple to implement that they’re known as language features. The Visitor pattern is a good example.1 Visitor’s Implementation section [21, pg. 338] says, “Visitor achieves [its goal] by using a technique called double-dispatch. It’s a well-known technique. In fact, some programming languages support it directly (CLOS, for example).” What does this mean? It means that mentioning the Visitor design pattern to CLOS (Common LISP Object System) developers will leave them scratching their heads. “A pattern? For a language feature? Why?” In CLOS, Visitor is essentially built in. You don’t need a pattern to tell you how to best express the concept—it’s already there in the language as a basic feature. In most other languages, however, Visitor provides a clean way of expressing the same programming concept of double dispatch.
This illustrates an important point. If you mention double dispatch instead of the Visitor pattern to the same CLOS developers, they would know what you mean, how to use double dispatch, and when not to use it. Terminology, particularly shared common terminology, matters a great deal.
This is true for all languages and all patterns: some languages make certain patterns easier or trivial to implement and other patterns more difficult to realize. No language can really be considered superior to another in this case, however. One common myth is that design patterns make up for flaws in programming languages, but that isn’t the case. Design patterns describe useful concepts, regardless of the language used to implement them. Whether a specific concept is baked into the feature set of a language or must be implemented from scratch is irrelevant. The concept is still being expressed in the implementation, and that is the critical observation that lets us talk about software design independently of software implementation. Design is concepts; how those concepts are made concrete in a given language is implementation.
When you get down to it, there’s no reason you couldn’t implement every pattern in the GoF text in plain C—but it would be extremely tedious. You’d have to build up best practices for binding data and functions into meaningful semantic units, encapsulating that data, ensuring that data is ready to use at first accessibility, and so on. This sounds like a lot of work, but these were concepts considered so important that they launched a revolution in language features to make them easier to work with. That revolution was object-oriented programming.
In object-oriented languages, those concepts are included as primary language features called classes, visibility, and constructors. Again, we can refer to the GoF: “If we assumed procedural languages, we might have included design patterns called ‘Inheritance,’ ‘Encapsulation,’ and ‘Polymorphism.’” The authors felt that this statement was important enough that it appears in Section 1.1 in the Introduction. And yet again, this is a fundamental point that seems lost on most developers, so let me restate it.
Patterns are language-independent concepts; they take form and become concrete solutions only when you implement them within a particular language with a given set of language features and constructs.
This means that it is a bit strange to talk about “Java design patterns,” “C++ design patterns,” “WebSphere design patterns,” and so on, even though we all do it. It’s a mildly lazy form of shorthand for what we really mean, or should mean: design patterns as implemented in Java, C++, WebSphere, and so on, regardless of language or API.2
Unfortunately, if you’re like many developers who have encountered one of the multitude of books on design patterns, you may have been trained, or at least have been erroneously led to believe, that there is some ephemeral yet fundamental difference between patterns as expressed in Java and those expressed in another language such as Smalltalk. There really isn’t. The concepts are the same; only the manner in which they are expressed and the ease with which a programmer can implement them in that specific language differ.
We need to focus on these when investigating design patterns, and these abstractions must be the crux of understanding patterns. Unless we make the effort to look at patterns as language-independent concepts, we are merely learning rote recipes again and losing much of what makes them so useful.
1.2.3. From Myth to Science
The issues described previously belie an underlying problem with design patterns as they are often conveyed, used, and understood today. All too often, we still don’t know why we do what we do, even when we use design patterns in our code. By using design patterns so inflexibly, we’ve simply better documented a body of unselfconscious snippets without the comprehension that comes from a methodical analysis of the snippets.
We have an art. What we need is a science. After all, we throw around the terms computer science and software engineering with abandon. Treating patterns as sample code misses the point of design patterns. Design patterns enable us as an industry to experiment with those concepts and share, discuss, and refine our findings.
Patterns as rote recipes are tribal mythology.
Patterns as concepts are the foundations of a science.
Elemental Design Patterns are the building blocks of that science.