Ran into a couple articles on unit testing today: 6 Rules of Unit Testing and JUnit Best Practices.  Between the two, 22 guidelines for using JUnit, all of which apply to NUnit as well.  Charles' rules are really important in that he's getting at test driven programming.  I started doing this when I went to add a major new feature to one of our web services and it worked well.  When you do test first, it's almost inevitable that you cover the failure cases first, because as Charles says, code never works right the first time.  For me, this means that NUnitGUI is my startup application when running out of the IDE.  One thing that I've found is that coding in this way, I spend a lot less time in the debugger.  The downside, though, is that this takes some discipline, because we've all built up a bad habit of doing the code/compile/debug cycle.  The cycle needs to be code/compile/test. 

One problem that I have is that I don't always understand the APIs I'm working with, so I learn them through the debugger.  This is sort of a related problem to the "prototype that becomes production code" problem.  You really need to get a clean separation between code that's experimental and code that's going to be used for something.

GDSs traditionally facilitate communication between the TPF and other applications using proprietary application programming interfaces. The Web services concept, however, uses such common Internet languages as XML to save programming time and money for GDSs and their partners. Galileo's first customer on Web services is AAA. Lutz said many more customers will use the services in the next few months.[Business Travel News]

An interesting article with a good overview of the general situation in the GDS business and different efforts to break away from the IBM TPF based systems that run most of the processing. The article starts off by citing a report which suggests that moving away from TPF might be a bad idea, but doesn't add anything to that assertion, except to suggest that the movement's driven more by programmer preference than real need. That does happen, but doesn't address the point that nobody's building new systems on TPF these days, either, or the basic fact that even the people who know TPF would rather program in C or C++.

AOP in Java. Rickard Öberg has been running a series of entries on implementing AOP in Java. He's working on his own AOP system, after having tried AspectJ and found it lacking.

Beyond Encapsulation Wrapping objects in web services is a bad idea, writes Adam Bosworth, an early champion of XML at Microsoft and now VP engineering at BEA Systems, in an article in XML & Web Services Magazine:

More on loosely coupled webservices, from, of all places, the Loosely Coupled Weblog. The Bosworth article that Philip references is a short read, but he sums up the points where I also take issue with the idea of passing objects across a SOAP interface.