That's not necessary the case.

'tis so:-)

 (except for the special cases of replacing self::*[1] and
ancestor::*[1] by . and ..)


  Here is the example
  Node A:
  /A/B[2]/C[text()='test']/text()
  Node B:
  /A/B/D[@a='123']/comment()
  The short cut doesn't follow xxx::*[n]

I wrote

> then I think in general for two element nodes, 

Just to save me writing out the full version, which I will now do:-)

neither of those are element nodes. if you want to include text and
comment nodes then just replace * by node() (and use a different number
for n)

If you need to include namespace and attribute nodes as a possibility for
the target then you  can use the above format to find the parent and add
oe more step namespace::abc or attribute::abc.

So you never need more than two steps.

So the general format for the shortest relative Xpath between two nodes
in the same document is is one of


.
..
xxx::node()[n]
yyy:zzz
../yyy:zzz
xxx::node()[n]/yyy:zzz


where

 xxx is one of a
  ancestor self preceding following descendant
 yyy is one of
   namespace attribute
 zzz is a QName (name of an attribute or namespace node)
and n is a positive integer


You could get a shorter path than that eg if there is only one foo child
you could use foo rather than descendant::foo[1] but you can not use
less steps I believe.

David

________________________________________________________________________
This e-mail has been scanned for all viruses by Star. The
service is powered by MessageLabs. For more information on a proactive
anti-virus service working around the clock, around the globe, visit:
http://www.star.net.uk
________________________________________________________________________