I have the following XPath problem: How to select the first preceding node,
whether it is an ancestor, a preceding-sibling or a "preceding-siblings child"?
As I understand, the axis 'preceding' selects everyhting except any ancestors.
Let's say we have the following document:

<a>
  <b/>
  <c/>
  <d/>
</a>
<e>
  <f/>
  <g/>
</e>

If the context node is <f/>, preceding::*[position()=1] does
not select <e> but <d/>. And if I use the ancestor axis, it
always selects an ancestor (as it is supposed to) even if it weren't the first
preceding node (e.g. the context were <g/>). I guess the problem boils down
to figuring out whether the first ancestor is also the first preceding node,
or is there some other nodes in between.

When I was travelling "downwards" in the document, the problem was
easy to solve. So what's the catch?

Try

(preceding::* | ancestor::*)[last()]

Note it's [last()] rather than [1] because the result of a union is always in document order.

This expression may be expensive but it should give the right answer.

Michael Kay
http://www.saxonica.com/
Author, XSLT Programmer's Reference and XPath 2.0 Programmer's Reference

Yep, that did he trick. Thank you. I'm still on the learning curve on XSLT, and a union didn't even come to mind. Thanks again...