OK

I think we have a basic difference of opion between what COULD be returned
in the nodeset

... and what IS currently returned

> somebody has to remove node <c> from <a>. No magic here, it's tree, not
> flat list, if you select node <a>, then .xml property would give you
> full subtree of <a>.

I dont see any reason why this has to be done by additional coding
when  it is quite clear what the intent of the XPath ancessor::* is
- to create a single node subset

> What is weird is the following  code:
>
> for (i=0;i<nodes.length;i++ )
> {
>    Response.write(nodes.nextNode().xml)
> }
>
> It's too naive as it presumes nodes in the nodelist can't be overlapping
> as in ancesor::* case - each one node contains immediately preceding one
> as a child and others as descendants. And the first one - root element
> contains all nodes in the tree (except document prolog).
>

Not if take the view that "distinct" root  nodes COULD be returned in the
nodeset . Looking
at

http://www.zvon.org/xxl/XPathTutorial/Output/example10.html

its  quite clear what could be returned - a  3 root node nodeset

<BBB/>
<CCC/>
<CCC/>

or for http://www.zvon.org/xxl/XPathTutorial/Output/example12.html

its quite clear that two root nodes should be returned


<BBB>
      <CCC/>
</BBB>

<CCC>
      <EEE/>
</CCC>


and similarly for ancestors example

http://www.zvon.org/xxl/XPathTutorial/Output/example14.html

a single root node nodeset

<AAA>
     <BBB>
          <DDD>
                <CCC/>
         </DDD>
    </BBB>
</AAA>


Unfortunately the nodeset currently returned makes this unnecessary
additional task!