> Can the key function be used to index partial document trees. Is this
> correct? Then, why does the specification demand that the
> supplied third argument must have a document node as an ancestor?
I think we were basically making assumptions about implementation
strategies, which may or may not prove to be correct. In adding the feature
to search within a subtree, we wanted to give users extra capability without
putting a lot of extra burden on implementors. We reckoned many products
might choose to build indexes at the document level even though only a
subtree is being searched.