XPath on DOM nodes not inserted in the owner document

The XPath data model is defined on a complete document tree and only for 
nodes in that document tree. The W3C DOM Level 3 XPath note however 
defines XPath on DOM nodes and with the DOM you can have nodes that are 
not (currently) inserted in the DOM document tree (e.g. by creating a 
node with a factory method like createElement and not inserting it in 
the document or by taking a node in the document and calling 
node.parentNode.removeChild on the node). Is there any specification 
that defines how the XPath data model with things like the root node 
denoted by '/' applies to DOM nodes that have parentNode as being null?

Mozilla and Opera 9 behave differently here, with Mozilla the XPath / 
evaluated relatively to a node with no parent node does not find any 
node while with Opera 9 the node itself is returned as the "root node".

I am considering filing a bug on Opera to have both browser implementing 
DOM Level 3 XPath behave the same but I would like to know whether the 
Mozilla implementation of this detail is based on some spec or just the 
decision of the implementor.

A test case is here:
<http://home.arcor.de/martin.honnen/javascript/2006/12/test2006120701.html>
it runs XPath expressions like /, .., and ancestor::node() first on two 
nodes inserted in the document, then remove the nodes from the document 
and runs the XPath expressions again.

-- 

	Martin Honnen
	http://JavaScript.FAQTs.com/