Daniel Veillard wrote:
>> Unclear it should. That mean the current document doesn't have any
>>key, the "temporal nodeset" actually behaves like a new document generated
>>on-the-fly. And this document is not indexed the source document for
>>the current node doesn't exist...
>
> Current conclusion is:
> your stylesheet relies on totally unspecified behaviour
> I have a hard time accepting as a bug, but if you
> have further evidence, I would take them...
Daniel, thank for taking care about this issue. You are right, I was
wrong when I though that key() function should always do lookup in the
source document even when context node is from other temporal document.
It seemed logical as such temporal document can't be indexed. But read
standard more carefully first and then start thinking applies here.
Sad conclusion for DocBook XSL stylesheets users is that single-pass
profiling won't work in xsltproc if you use links in your documents. It
actually works in Saxon but it is non-conformant behaviour. I can't
imagine XSLT 1.0 solution to this problem which will not cause very
hard-to-maintan fork between common and profiling stylesheets.
Jirka
--
-----------------------------------------------------------------
Jirka Kosek
e-mail:
http://www.kosek.cz
