Via Steve Vinoski's Answers for Sergey I stumbled upon Sergey Beryozkin's Questions for Steve which started of with the following question

1. Do you think client code generation is evil ? If yes, do you expect people to do manual programming on a large scale ?

The problem with the software industry [or should I say humanity?] is that we like to take absolutist positions because they are easier to defend or argue against than admitting that the world is full of shades of gray.

Starting with the first question, code generation isn't necessarily bad let alone evil. However there are lots of problems with how code generation is implemented by the major vendors that support the SOAP/WSDL/XSD/WS-* family of technologies. Steve does a good job of laying out the problems with these approaches.

The first problem Steve points out is that a lot of these SOAP toolkits have implement some form of location transparency which tries to hide as much as possible the differences between invoking a remote system and calling a method on a local object. This behavior even flies in the face of SOA since one of the four tenets of service orientation is that boundaries are explicit. Another problem is that the inflexible and rigid requirements of static typing systems runs counter to the distributed and flexible nature of the Web. I posted a practical example a few years ago in my post entitled Why You Should Avoid Using Enumerated Types in XML Web Services. In that example, I pointed out that if you have a SOAP Web Service  that returns an enumeration with the possible value {CDF, RSS10, RSS20} and in a future release modify that enumeration by adding a new syndication format {CDF, RSS10, RSS20, Atom} then even if you never return that syndication format to old clients written in .NET, these clients will still have to be recompiled because of the introduction of a new enumeration value. I find it pretty ridiculous that till today I have list of "people we need to call and tell to recompile their code whenever we change an enum value in any of our SOAP Web Services". Of course, some of this foolishness can be mitigated in a statically typed system using technologies like Java's dynamic class loading but that is really just an insecure workaround that tries to hide the fact that what you really need here is a dynamically typed system. 

The second question is really asking whether we want developers writing XML processing code by hand instead of having a code generator do this work. Even though I used to work on the XML team at Microsoft, I do agree that it is a valid concern that you shouldn't want to spend a lot of effort writing code for parsing and processing XML if that is not the core function of your application. Again, Steve Vinoski hits the nail on the head with the suggestion to use standard data formats and MIME types. For example, if I decide to use application/atom+xml MIME type for the data that is returned by my RESTful Web service then clients can choose from a vast array of libraries for processing Atom feeds [such as the Windows RSS platform, ROME, Mark Pilgrim's Universal Feed Parser, etc] without having to write a lot of XML parsing code. If you must provide your own custom formats then it is imperative to make sure that it is easy to consume these formats from any platform by using a consistent and simple data model for the data format. A number of popular Web service APIs like the Flickr API, and the Facebook platform have provided client libraries for their APIs, this should be considered the exception and not the rule. Even in their case, it is interesting to note that a large proportion of the client libraries for these services are not actually maintained or developed by the creators of the service. This highlights the value of utilizing simple data formats and straightforward protocols. That way it isn't actually a massive undertaking for client developers to build and share libraries that abstract away the XML processing code. Of course, all of this can be avoided by just using standard MIME types and data formats that are already supported on a wide array of platforms instead of reinventing the wheel.