The Tools Landscape | SYS-CON AUSTRALIA

The Tools Landscape

Sep. 23, 2002 12:00 AM

Web services products have matured rapidly over the past 12 months, to the point where it's become acceptable to utilize the technology in major projects. However, though improving, the currently available toolkits can't be considered complete. This article surveys the rapidly evolving development tools landscape and addresses what tools a fully featured environment should provide to the Web-services developer.

IDE Integration
Historically, developers have toiled with an array of utilities, editors, and command-line tools. The ability to effectively navigate this maze of programs is a source of pride to many of them; however, this maze introduces a steep learning curve and, in the long run, lowers productivity. Some of the costs associated with such a nonintegrated approach are time lost switching back and forth between programs, time required to learn a new user interface, and cost of owning and managing multiple applications. It's also been costly for the tools vendors; each has written many thousands of lines of code that could have been shared between implementations. This includes code for the user interface, file and project management, version control, and many other housekeeping tasks.

The past year has seen major changes in the development landscape. Proprietary environments have fast lost ground to the concept of a single development framework into which additional tools can be deployed via a plug-in architecture. Products like JBuilder and NetBeans have provided this type of framework for some time, but it's fair to say that an industry-wide paradigm shift has coincided with the arrival of two über-frameworks: Microsoft's Visual Studio .NET and the open-source Eclipse project.

IBM's reasons for donating Eclipse to the open-source community are no mystery. If the non-Microsoft community doesn't close ranks around a single framework, it will not be economically feasible to compete with the de facto IDE monopoly for Windows development: Visual Studio. Developing an IDE is an expensive business - IBM spent $40 million on Eclipse. This may seem a costly bit of software to give away, but having the developer community adopt its technology gives IBM a significant head start since IBM's tools already run in Eclipse, while other vendors must go to considerable lengths to move to a plug-in architecture. Tools vendors may not wish to swallow this bait, but the drive toward a single IDE platform is given unassailable momentum by two converging forces: the developer-side desire for an integrated experience, and the economics of proprietary development.

One of the consequences of IDE integration is that products built on the same framework will largely look the same. The mainstream toolsets will also have significant overlap in functionality, though vendors will try hard to differentiate their offerings. This commoditization of the tools market is bad news for smaller vendors. In a battle of commodities, the larger, better-resourced players tend to win. It's not all gloom though. As vendors strive to outshine the competition, we can expect to see higher quality and more innovative software. Another plus: open-source groups and creators of niche or boutique software can take advantage of the IDE frameworks to deliver their software in a first-rate package. Although developers will initially lose some choice due to the inevitable process of consolidation in the tools market, this will be more than offset by improved product sets from surviving vendors and a more diverse range of software from smaller software houses. Web-services developers in particular stand to gain, as there are a great number of companies targeting this space.

Basic Tools
Before we survey some of the more interesting tools that are becoming available, let's establish the basic functionality required to develop Web services. From within the IDE, the developer should be able to:

Generate Web Services Description Language (WSDL) from a component: This should be as simple as right-clicking com. acme.MyClass and selecting "Generate WSDL...". This command should bring up a dialog that allows the developer to specify which methods to expose, what to name the Web service, and other basic options.

Generate stubs and skeletons from WSDL: It should be possible to generate both client-side stubs and server-side skeletons, preferably in a variety of languages. For Java, the developer should have the option of generating EJB or regular Java skeletons.

Deploy a Web service: Deployment can be an intricate process, so the IDE should provide a wizard to help assemble the classes, documents, and configuration data necessary to successfully deploy. Several products have introduced the concept of a Web Services Archive, similar to a JAR or WAR file. This can make it considerably easier to manage deployment, especially across multiple machines.

Note that the degree to which an IDE exposes WSDL can vary. In Visual Studio, the developer adds WSDL to a project by invoking the "Add Web Reference..." command. This automatically generates client proxy code and adds it to the project. The WSDL is effectively hidden from the developer. Hiding the complexity of WSDL is fine if your application is a consumer of Web services, but server-side development requires more control. This is particularly the case with Java because Web services haven't been tightly integrated into the JVM as Microsoft has done with the .NET Common Language Runtime (CLR). We might expect the toolset to grant the developer considerable control over the WSDL. In particular, it's important to be able to customize the mapping between WSDL and the native language.

WSDL Mapping
When WSDL is generated, the created schema types are typically named after classes, with elements named after the members of the class. Frequently, the developer wants to rename or omit a member, or otherwise change how the class is serialized. Why? Perhaps the Web service has to conform to a specific WSDL interface defined by a standards organization or commercial partner. Or it might be for more cosmetic reasons, such as changing unhelpful element names; or for business reasons, such as not exposing sensitive information.

To serialize an object, a Web services platform uses a mapping layer that defines how the native representation (e.g., a Java class) is mapped to XML (see Figure 1). The method and degree of control over this layer varies by product, but there are three common ways to provide customization:

Configuration file: This will usually allow fairly limited customization, such as changing member names or omitting elements.

XSLT: The native object is first serialized per the default rules, and then an XSLT transform is performed, allowing a greater control. However, developers can find XSLT difficult to work with, and the transform is processor intensive.

Custom serialization class: Instead of using reflection to inspect and serialize the object, the serialization process can be delegated to a helper class created by the developer. This mechanism provides almost unlimited control, although modifying and managing the additional classes is cumbersome.

Most platforms support the use of at least one method of directly editing the mapping layer. However, tools support is still weak. In the near future, developers can expect to see software that provides full round-trip support for editing WSDL. Such an editor will allow developers to modify elements in the WSDL and will automatically propagate these changes to the mapping layer. Note that it should be possible to perform the converse operation: modify the source component, and the changes are propagated toward the WSDL. Mapping is one concept that developers will encounter frequently with Web services. This is because of an under-appreciated enabling technology: XSLT.

XSLT and Graphical Mapping
XSLT is a technology that most developers haven't had significant contact with. It's often thought of in the context of generating HTML pages, but its utility extends far beyond that. As mentioned earlier, XSLT can be used to control the mapping between WSDL and a native language. Some other uses are:

Map requests from an older version of a Web service into the new version's format
Map non-SOAP XML documents into SOAP requests

XSLT has a steep learning curve - for many developers the "side effect-free" programming model is less than intuitive to grasp, so good tools are essential. Thankfully XSLT lends itself quite well to graphical editing, where two schemas are laid out side by side and related elements are linked by dragging one to the other (see Figure 2).

Let's look at an example. AirportWeather is a publicly available Web service that reports on conditions at airport weather stations around the world. There's also a recently created successor Web service, GlobalWeather, which is faster and has a richer data model. Both of the services have an operation that takes one parameter, the weather station code (e.g., JFK or SFO), and returns a weather report for that location.

We can use XSLT to transform a SOAP request for AirportWeather into a SOAP request for GlobalWeather. Instead of creating the XSLT by hand, we load the source schema (AirportWeather.wsdl) and the target schema (GlobalWeather.wsdl) side by side. Then we graphically link the two parameters - AirportWeather's nonintuitively named "arg0" and GlobalWeather's "code" - and the mapper will generate the necessary XSLT. In a similar manner, we can use XSLT to transform a non-SOAP XML document into a GlobalWeather request (see Figure 3).

XSLT is an underused technology with great potential, in particular for integrating Web services and legacy XML applications. A good development environment will support the graphical creation of XSLT, and it should allow the easy deployment of XSLT documents into your Web-services platform.

Testing
Once a Web service is built and deployed, the next step is to test it. Testing can range from merely verifying that a service is operational to a full automated test suite. The most common testing approach is to generate a client proxy from the WSDL and write a test class to invoke the proxy. But there's a better way: the IDE can generate a test class which invokes the Web service's operations using default parameter values. For Java, the test cases can make use of the JUnit test framework and the Ant build system to make it easy to integrate Web service testing into existing automated test suites.

Another approach is to generate a SOAP document from the schema information in the WSDL and send it directly to the Web service endpoint using HTTP. This isn't amenable to automated testing but is useful if the developer is interested in the lower-level details of the SOAP messages. However, the friendliest approach is to generate an HTML Web client that's deployed on the Web server alongside the Web service. This Web client is typically based on JSP or ASP and consists of one page for each operation. The form fields correspond to the operation parameters, and submission of the form results in the Web service being invoked. As well as being a fast means of verifying that a service is alive, this is also an effective way of demonstrating a developing Web service.

Debugging
Debugging a Web service is essentially the same as debugging any other server-side software, such as an EJB. The IDE's debugger attaches to the service process using the standard mechanisms for the platform, which usually involves starting the server in debug mode. But unlike an EJB, the raw messages between a client and a Web service are intelligible and of interest to the developer.

There are two quite similar ways of monitoring SOAP messages. A TCP tunnel is a program that listens on a port and directs all traffic to a specified host and port (e.g., tunnel from localhost:7000 to server.acme. com: 8000). The tunnel program can then display the messages in a GUI. To enable tunneling, the SOAP endpoint URL used by the client must be modified to point at the tunnel's listening port.

Like a TCP tunnel, an HTTP proxy listens on a port, but it has specific knowledge of the HTTP protocol. The proxy reads the "Host" field in the HTTP header and directs traffic to the indicated location. A proxy is more transparent than a tunnel in that proxy settings can generally be set on a process-wide basis and thus modification of individual endpoints is not required.

Conclusion
Monitoring tools were an indispensable item in the early days of Web services, when developers spent a lot of time working with raw SOAP messages. Although a development environment should still provide the ability to work at this level, many developers will never need to do so. It's a sign of the maturity of the technology that this is the case. Web services tools are entering the mainstream.

Related Links

▪ Figure 1

▪ Figure 2

▪ Figure 3

About Neil O'Toole
Neil O'Toole is a Technology Evangelist at Cape Clear, where he is responsible for promoting the adoption and effective use of Web Services by developers. He manages the CapeScience developer network (www.capescience.com), moderates the Cape Clear newsgroup, and works with the Web Services development community to help define Cape Clear
product direction.
He is the author of the popular NetTool debugging utility, has written dozens of technical articles and papers, and is an accomplished public speaker.
Before beginning his career in evangelism, Neil was a senior developer at Cape Clear. Prior to that he worked with Goldman Sachs (an investment bank), CR2 (banking software), and Esat Net, part of British Telecom. He holds a first class degree in Computer Science from Trinity College Dublin.