In Java, the only way to change the behavior of a class is to subclass it, adding the new behavior to that subclass. In XSD, you can extend a data type by restricting it: that is, constraining the valid values for the elements and attributes. For instance, consider the following example:

<simpleType name="AirportCode">
  <restriction base="string">
      <pattern value="[A-Z][A-Z][A-Z]"/>
  </restriction>
</simpleType>

This type restricts a XSD string by ways of a regular expression, allowing only three upper case letters. If this type is converted to Java, we will end up with an ordinary java.lang.String; the regular expression is lost in the conversion process, because Java does not allow for these sorts of extensions.

One of the most important goals of a Web service is to be interoperable: to support multiple platforms such as Java, .NET, Python, etc. Because all of these languages have different class libraries, you must use some common, interlingual format to communicate between them. That format is XML, which is supported by all of these languages.

Because of this conversion, you must make sure that you use portable types in your service implementation. Consider, for example, a service that returns a java.util.TreeMap, like so:

public Map getFlights() {
  // use a tree map, to make sure it's sorted
  TreeMap map = new TreeMap();
  map.put("KL1117", "Stockholm");
  ...
  return map;
}

Undoubtedly, the contents of this map can be converted into some sort of XML, but since there is no standard way to describe a map in XML, it will be proprietary. Also, even if it can be converted to XML, many platforms do not have a data structure similar to the TreeMap. So when a .NET client accesses your Web service, it will probably end up with a System.Collections.Hashtable, which has different semantics.

This problem is also present when working on the client side. Consider the following XSD snippet, which describes a service contract:

<element name="GetFlightsRequest">
  <complexType>
    <all>
      <element name="departureDate" type="date"/>
      <element name="from" type="string"/>
      <element name="to" type="string"/>
    </all>
  </complexType>
</element>

This contract defines a request that takes an date, which is a XSD datatype representing a year, month, and day. If we call this service from Java, we will probably use either a java.util.Date or java.util.Calendar. However, both of these classes actually describe times, rather than dates. So, we will actually end up sending data that represents the fourth of April 2007 at midnight (2007-04-04T00:00:00), which is not the same as 2007-04-04.

Imagine we have the following simple class structure:

public class Flight {
  private String number;
  private List<Passenger> passengers;
    
  // getters and setters omitted
}

public class Passenger {
  private String name;
  private Flight flight;
    
  // getters and setters omitted
}

This is a cyclic graph: the Flight refers to the Passenger, which refers to the Flight again. Cyclic graphs like these are quite common in Java. If we took a naive approach to converting this to XML, we will end up with something like:

<flight number="KL1117">
  <passengers>
    <passenger>
      <name>Arjen Poutsma</name>
      <flight number="KL1117">
        <passengers>
          <passenger>
            <name>Arjen Poutsma</name>
            <flight number="KL1117">
              <passengers>
                <passenger>
                   <name>Arjen Poutsma</name>
                   ...

which will take a pretty long time to finish, because there is no stop condition for this loop.

One way to solve this problem is to use references to objects that were already marshalled, like so:

<flight number="KL1117">
  <passengers>
    <passenger>
      <name>Arjen Poutsma</name>
      <flight href="KL1117" />
    </passenger>
    ...
  </passengers>
</flight>

This solves the recursiveness problem, but introduces new ones. For one, you cannot use an XML validator to validate this structure. Another issue is that the standard way to use these references in SOAP (RPC/encoded) has been deprecated in favor of document/literal (see WS-I Basic Profile).

As mentioned earlier, the contract-last development style results in your web service contract (WSDL and your XSD) being generated from your Java contract (usually an interface). If you are using this approach, you will have no guarantee that the contract stays constant over time. Each time you change your Java contract and redeploy it, there might be subsequent changes to the web service contract.

Aditionally, not all SOAP stacks generate the same web service contract from a Java contract. This means changing your current SOAP stack for a different one (for whatever reason), might also change your web service contract.

When a web service contract changes, users of the contract will have to be instructed to obtain the new contract and potentially change their code to accommodate for any changes in the contract.

In order for a contract to be useful, it must remain constant for as long as possible. If a contract changes, you will have to contact all of the users of your service, and instruct them to get the new version of the contract.

When Java is automatically transformed into XML, there is no way to be sure as to what is sent across the wire. An object might reference another object, which refers to another, etc. In the end, half of the objects on the heap in your virtual machine might be converted into XML, which will result in slow response times.

When using contract-first, you explicitly describe what XML is sent where, thus making sure that it is exactly what you want.

Defining your schema in a separate file allows you to reuse that file in different scenarios. If you define an AirportCode in a file called airline.xsd, like so:

<simpleType name="AirportCode">
    <restriction base="string">
        <pattern value="[A-Z][A-Z][A-Z]"/>
    </restriction>
</simpleType>

You can reuse this definition in other schemas, or even WSDL files, using an import statement.

Even though a contract must remain constant for as long as possible, they do need to be changed sometimes. In Java, this typically results in a new Java interface, such as AirlineService2, and a (new) implementation of that interface. Of course, the old service must be kept around, because there might be clients who have not migrated yet.

If using contract-first, we can have a looser coupling between contract and implementation. Such a looser coupling allows us to implement both versions of the contract in one class. We could, for instance, use an XSLT stylesheet to convert any "old-style" messages to the "new-style" messages.