XPath ambiguity

Question

I am working with the following fragment of an XPath

ancestor::contribution[1]/preceding-sibling::contribution[@speaker-reference][1]

My Java application (using JDOM for XPath queries) interprets this differently from our Oracle database (11g). I was able to solve the problem by using brackets as follows:

 (ancestor::contribution[1]/preceding-sibling::contribution[@speaker-reference])[1]

So it seems that JDOM is reading the xpath as "the first of all preceding contributions with an attribute @speaker-reference" while Oracle thinks "the preceding contribution with an attribute @speaker-reference and position()=1".

I had actually intended the first interpretation. I am wondering which of the two interpretations is correct according to the XPath specification (couldn't find the right place to look) or whether the specification allows expressions to be ambiguous.

Answer 1

According to the XML Path Language Specification, square brackets operator [..] has precedence 19, while slash / has precedence 18. This means that the last square brackets [1] should be applied to the portion of the expression after the slash /, not to the entire expression. In other words, Oracle's interpretation is right.

Java's implementation^*, which gives you the correct result without parentheses around the / expression, is not compliant with the standard. Consider filing a bug with a short example and an explanation of what is going on.

^* Ironically, that's Oracle's implementation as well.

Answer 2

Given your description it is not easy to see what JDOM and Oracle are doing. But their different behaviour is apparently caused by a different implementation of the proximity position of a reverse axis.

Since ancestor::contribution[1] evaluates to an empty node-set or a single node we can simplify your example to the following situation, using element x as context node:

<doc>
   <contribution speaker-reference="a"/>
   <contribution speaker-reference="b"/>
   <contribution/>
   <x/>
</doc>  

Selecting preceding-sibling::contribution[speaker-reference] returns the two contribution nodes with a speaker-reference attribute in document order.

Selecting (preceding-sibling::contribution[speaker-reference])[1] returns the first of these two nodes which is the one with speaker-reference = a.

Selecting preceding-sibling::contribution[speaker-reference][1] the position predicate [1] must now be interpreted according the axis order. Since preceding-sibling is a reverse axis the selected node-set must treated in reverse document order. The first position in this node-set is the node with speaker-reference = b.

Hope this allows you to clarify which implementation has got right.