One of the things I did in my latest version of WSS RSS was add a configuration file. I didn’t just use web.config since WSS has a tendency to overwrite it when you upgrade a new virtual server to support WSS. However, I didn’t go the usual route of parsing the XML configuration information into objects via serialization or some other mechanism. Instead, I just loaded up an XmlDocument and queried it with XPath whenever I need to read configuration information. I do this multiple times per web request (once per web and list in the feed).
This leads me to my point – I’m beginning to think the whole idea of XML serialization is off base. Given an object graph, I can serialize it into XML and then deserialize it into an identical object graph. However, I don’t think I can take an arbitrary XML document and always be able to deserialize it into an object graph. XML concepts like derivation by restriction are difficult to model in a strongly typed object model like CLR’s. And since I have XPath, I don’t think I want to always deserialize my XML into objects – it was much simpler to execute XPath queries than it would have been to manually traverse the object graph or build my own query mechanism.