I believe that there is a strong tendency for developers to think of Web services as methods. In reality it is not that simple. A Web service defines an XML conversation between client and server. Yet, when a developer sits down to create Web services he or she usually starts by creating a method.

From that starting point it is easy to get a software toolkit to turn your method into a Web service. By default or with a few settings you can allow the toolkit to define the conversation around your method. This method-first development technique works, especially for small or pilot projects, but it is not the best approach. A contract-first approach results in better long-term development, interoperability, and maintenance.

Convincing a development team to work contract-first is not as simple as it might seem. You have to convince the team to spend more effort on unfamiliar languages like XML Schema and WSDL, and rely less on familiar languages such as Java or C#. With that said, there are good reasons to code your Web service the hard way - contract-first. In this article we will look at six good reasons to develop Web services contract-first instead of method-first. By the end of this article I hope you will see that contract-first is a superior way to create Web services and will save you time and money.

RPC Thinking
Before we delve into the six reasons, I want to establish the difference between method-first and contract-first Web services. We can walk through the basic steps of both styles of Web services creation using a simple service. This will allow me to show some code and remove the abstractness from the topic. Forgive me for using an example that has been repeated countless times. If you're an old pro at Web services skip ahead to the "Reason 1" heading. The following code is a Calculator with one method. My preferred weapons are Java and Apache Axis, so all examples will be coded in Java. Your choice of language or platform should have little effect on which way you will make Web services. (see Figure 1)

Step1: Create a method that will become a Web service. The code below provides simple add method that will be turned into a Web service.

public int add(op1, op2)
{
    return op1 + op2;
}

Step 2: Run this method through a software toolkit that turns it into a Web service. There are many of these toolkits on the market. It might be as simple as adding an attribute, or saving the source code to a different extension. For the rest of this article we will call that machine the Web service toolkit. (see Figure 2)

Step 3: Deploy it to some Web server. The server will provide HTTP handling and leave you with a very simple Web service. Web service toolkits automatically create XML conversion code and a Web service description or WSDL (pronounced wizdle) document. This WSDL document is the key to getting a client to call your Web service and will play a prominent roll in contract-first thinking.

Step 4: A client can now obtain the WSDL document and create a client through your Web service. The cool thing is that the client can basically treat the Web service like a regular method call. Although XML is being passed via HTTP following the rules of SOAP, the client can be created very easily and the client doesn't need to know about the XML payload. (see Figure 3)

Contract-First Thinking
Taking a quick look at contract-first development, we can clearly see the differences between method-first and contract-first. In contract-first Web services you first create the WSDL document. The WSDL might have supporting XML Schema documents as well. As you can see from Listing 1, the WSDL code is a bit longer and more involved than the simple Java method. In fact, when you look at it in such a simple example, it almost seems ridiculous to suggest that you might write the WSDL before you write the Java code, but that is exactly what I'm suggesting. The steps to create the Web service are similar to what we've already seen, but kind of in reverse.

• Step 1: Create the WSDL and supporting Schema
• Step 2: Generate the service from the WSDL

Reason 1: Schema Is More Descriptive
Data types in Java and C# and other languages are described only in code, which is not shared with the client. Clients are generated via the WSDL document. Clients don't have the full definition of the data types unless you include that information in the XML Schema document. As a very simple example, say you are passing in a Person that consists of a firstName, middleInitial, and a lastName. In code this is simple, but some of the data may not be required. Say you want to require the firstName and lastName, but the middle initial is optional. If your WSDL is automatically generated, it is likely to turn out like Listing 2. The middleInitial is created by default, but with the XML Schema language we can get much more descriptive. With this simple example we can easily make the middleInitial optional by including one little attribute called minOccurs like the code below.

<xs:element name="middleInitial"
minOccurs="0" type="xs:string" />

I hope this simple example illustrates how schema can be more descriptive, but it is such a tiny example. Imagine a rich set of data, not just a simple person. If you want your clients to understand that the quantity element can only be non-negative integers, or have them understand the pattern that the SKU number must be three letters followed by a hyphen and four numbers, that can be described in XML Schema. By relying on the tool to generate the proper schema, you are passing only a vague description of the data where an e-mail address becomes a string and a month is an integer that can be negative.

Knowing that the WSDL is your client's interface it should be obvious that you want to make it as accurate and descriptive as possible. That might not be enough to write the WSDL first. You could still modify the WSDL after the fact and pass that out to your clients, but we have more reasons to go.

Reason 2: Your Language Is Not Alone
To understand the "we are not alone" idea, we only need to expand the Web service example a bit. You create a Web service and you require that data representing a customer be sent to invoke the service. You may have put some time into the customer type and you realize it should be shared. Other departments also use the same entity in their services, but wait, here's the problem. You code in C#, they code in Java. You can't just centralize the code. They need to convert it to Java. No big deal, as easily done as said practically, but what if the definition of a customer changes? If the Java coders change the customer your two versions no longer match. Which version is the "central" or "master" customer type? Do I need to change the Java code every time I change the C#? How do you communicate the change to the other team?