>Now, if I create a custom implementation of org.w3c.dom.Node that makes >the database queries when the Nodes are expanded, can I trust the XPath >API not to expand the nodes unless demanded by the XPath queries I am >using?
I think the answer is "no". XSLT doesn't make any promises about the order in which the source document is accessed, nor when data is or isn't read from that document. Xalan _currently_ reads DOM nodes in document order, on an as-requested basis. That may change in the future. My recommendation would be that you break this up into stages -- perform a preprocessing stage which causes your DOM to be expanded (or at least stable), _then_ run XSLT on the result.
