Foundations

Last month, we took a long look at strong typing. We saw that while strong typing offers many benefits in a language such as Java, trying to attach strong typing to ColdFusion produces really difficult problems. And last month, due to my "in" connections, we were even able to briefly interview the Java compiler.

Some might have thought that I was arguing the case against strong/static typing, as many developers have recently. They have argued that static typing is an emperor without clothes. Bruce Eckel, of Thinking in Java fame, has called strong typing "a Faustian bargain." Guido van Rossum, creator of Python, has said, "Strong typing catches many bugs, but it also makes you focus too much on getting the types right and not enough on getting the rest of the program right." Developer Ned Batchelder has summed up the case against strong typing: Static typing prevents certain kinds of failures. Unfortunately, it also prevents certain kinds of successes."

My own view of strong/static typing is different: I find it immensely helpful in a language like Java. I think that a good deal of the problems developers have with strong typing in that language would be greatly ameliorated by designing with interfaces. And I appear to be in good company. The "Gang of Four" authors who wrote the seminal work, Design Patterns, offered readers the excellent advice: "Design to interfaces, not to implementations."

Clearly, the issue of strong versus weak / static versus dynamic typing is one on which many smart and experienced developers disagree. In this article, I want to make a case that, whatever your view of strong/static typing is, in regards to ColdFusion, it is a moot point simply because ColdFusion does not support strong typing.

But is this a defect in ColdFusion or merely a difference? Certainly languages such as Smalltalk, Python, and Ruby seem to do fine without strong typing. To that illustrious list, I would add ColdFusion.

In place of the debate over whether strong or weak typing is better, perhaps a better question is this: How can we benefit from the dynamic typing ColdFusion offers? First, I think we would do well to adopt what Dave Thomas (The Pragmatic Programmers) whimsically calls duck typing - as in, "if it walks like a duck and quacks like a duck, it must be a duck!"

With duck typing, the idea of a type becomes more a matter of "Can this object respond to this message?" rather than "Does this object fit into a type hierarchy?". This is particularly important with ColdFusion since the only type hierarchy available is one of inheritance, which commonly causes fragile code.

Let's look at an example of duck typing. Having just returned from teaching two classes in Las Vegas, I'll use an example from the poker craze that has turned ordinarily calm souls into wild men (and women!) pushing their entire stack of chips into the pot while calling "All in!" Let's suppose that we've been asked to come up with a domain model for a Las Vegas poker room.

We might begin with a Player CFC that has the following methods:

• bet (accepts a numeric value)
• check
• fold
• raise (accepts a numeric value)

We probably also want a Dealer CFC with these methods:

• dealHoleCards
• promptPlayer
• acceptBet (accepts a numeric value)
• dealFlop
• dealTurn
• dealRiver
• declareWinner

Perhaps, we'll have an Employee class that Dealer inherits. No doubt an Employee will have many methods but the one closest to our hearts is this:

• acceptPay (accepts a numeric value)

Now, a little known fact is that many Las Vegas poker rooms employ proposition player (or simply props). Prop players are paid by the house to stimulate action, but play with their own money. Since prop players are paid by the house, our Prop class might rightfully extend Employee.

But, let's suppose that we need to model a tournament. Our Tournament CFC could have methods like:

• register (accepts a Player and a numeric amount)
• payWinners

We run into a problem here. It may be that in order to fill a tournament, the house asks a prop to enter a tournament. But the Tournament's register method accepts only Player objects. In a strongly typed system, we really want our Prop players to be of both Employee and Player types.

Java solves this problem through interfaces. A class can extend only one superclass, but it can implement as many interfaces as needed. Interfaces define a type and the methods that can be called on that type, but provide no method bodies - that is, no implementations of these methods. Assuming we want Prop to extend Employee (and that ColdFusion had interfaces), we could provide a Player interface that would be functionally similar to this Java version:

public interface Player{
    public void bet(int bet);
    public void check();
    public void fold();
    public void raise(int bet);
}

Interfaces provide type inheritance, but not code inheritance: the implementing class must provide the method bodies for all methods defined in the interfaces it implements. While this does solve the problem of multiple-typing, it does so at a cost. If we need to change the implementation of a Player, it must be done in both the Player class and the Prop class. This violates what is sometimes called the DRY (Don't Repeat Yourself) principle. According to Dave Thomas, "DRY says that every piece of system knowledge should have one authoritative, unambiguous representation." Here, we have (at least) two representations for each Player method.

For the many ColdFusion developers who dutifully provide type properties for their arguments and returntype properties for their methods, this has caused ineluctable problems that have not been resolved. But again, let's ask the question: Is this a defect in ColdFusion or merely a difference?

If we embrace ColdFusion's weak typing, there is no problem. As long as both the Player object and the Prop object can respond to the methods needed to play in the tournament (bet, check, fold, and raise), our code will run without a snag. "But it's not type safe!" the strong type proponent might argue. That's true, but ColdFusion doesn't offer type safety (nor do Smalltalk, Python, or Ruby). We might say that if an object can bet, check, fold, and raise like a Player, then it is a Player - regardless of its class.

Welcome to duck typing. All that is required is that we give up the illusion of type safety (and with it, the returntype and type properties of methods and arguments, respectively). And there are other benefits to duck typing. One informal study found that Java applications compared to the same applications in Smalltalk required between 85-230% more code. That's a lot of finger typing.

Of course, speed doesn't mean much if we are writing fragile code that will be hard to maintain. Since the overwhelming majority of the cost of code over its lifetime is spent on maintenance, anything that compromises the maintainability of code must be highly suspect. And strong typing must be credited with catching many errors that may not be at first apparent, but which will surface over an application's lifetime.

As you might guess, the duck typers have an answer for this: unit testing. Unit testing is a part of a movement/philosophy known as test-driven development. In test-driven development, automated tests are written as part of the software design cycle. Such tests act like "mini clients," exercising the various methods of each class to uncover any bugs.

For such testing to be feasible, it must be automated. Kent Beck and Erich Gamma produced a tool to do just this for Java programmers: JUnit. It has since been implemented in many languages. ColdFusion has two programs that provide automated test unit functionality, CFCUnit (www.cfcUnit.com) and CFUnit (www.cfunit.sourceforge.net). I strongly encourage readers to explore the difference using such tools can have on the quality of your software.

Perhaps you noticed that something was left out of our duck typing solution to the problem of prop players. If both the Prop and the Player classes have to independently implement the various player methods, are we not still in violation of the DRY principle? We are. The solution to this problem comes from applying the notion of mixins to ColdFusion.

Mixins are a technique, first introduced in a Lisp language variant known as Flavors, in which a class defines the attributes of the class, but does not define the methods of the class. Instead, the methods are defined in separate files known as mixins. In our example, both the Prop and the Player class would define their own instance variables, but would include the methods for bet, check, fold, and raise from a separate file - perhaps PlayerMixin.

The good news is that ColdFusion makes the use of mixins very simple and straightforward. Next month, we'll look at code that makes use of mixins and we'll see how mixins can either be applied to an entire class or to individual objects.

Add Your Feedback

In order to post a comment you need to be registered and logged in.

Register | Log in

Please wait while we process your request...