It's expected that once you have an interface language, like WSDL or CORBA IDL, that you will have a suite of tools that allows you to generate stubs, skeletons, starting code, tests and so forth from the basic interface description in the implementation language of your choice. This is a very important productivity element for any developer, as having to code up server skeletons by hand is just an outrageously dull task.
When I was putting together CORBA systems, I used a tool called idlgen which saved huge amounts of time by generating complete test suites with randomized data, and allowed you to produce stubs, skeletons and mainline code in a jiffy. In my current work, I'm using EMF to represent a suite of models, from which I generate implementation code.
But there's one thing that's starting to bug me about some code generation tools, and here I'm talking about one particular tool that is almost always called wsdl2java, and that is there are so many of them. Everyone! Stop writing new wsdl2java tools and just write a wsdl2star tool instead! Take wsdl4j or Woden as your model and write some backend processing that allows people to develop Velocity templates, or write JavaScript/E4X code to generate the necessary. Wrap all of this up in a single Open Source project and kick-off a community of Open Source script/template developers so that other projects can contribute their code generation incantations (there's your regression test suite too, by the way).
Is there someone doing this already? Let me know if you find them!