1846 2005-08-15 09:49:47 +0000 [XSLT2.0] xsl:number applied to attributes and namespaces 2005-09-29 12:51:57 +0000 1 1 1 Unclassified XPath / XQuery / XSLT XSLT 2.0 Last Call drafts PC Windows XP CLOSED FIXED P2 normal --- 1 mike mike public-qt-comments oldest_to_newest 5406 0 mike 2005-08-15 09:49:48 +0000 If the "from" attribute of xsl:number is omitted, then the effect in XSLT 1.0 is that numbering (for level="single" and level="multiple") considers the entire tree. The way we have phrased the rules in 2.0 is subtly different: we say that if "from" is omitted, the effective value of "from" is the root of the tree, and the nodes that we consider are those that are descendants-or-self of the root. The effect of this is that we don't consider attribute and namespace nodes, except in the special case that the attribute or namespace is parentless. The consequence is that <xsl:number/> with no "from" or "count", when applied to an attribute node, returns "1" if the attribute is parentless, and () when it has a parent. In XSLT 1.0, it always returned "1" (but don't try this in Saxon). I think the cleanest fix for this would be that instead of considering only descendants-or-self of $F (the innermost node that matches the "from" pattern), we should consider all nodes that have $F as an ancestor-or-self (or equivalently, we should consider descendants-or-self plus their attributes and namespaces). That is, for level="single" and level="multiple", we should change the clause Let $AF be the value of: $A intersect ($F/descendant-or-self::node()) to: Let $AF be the value of $A[ancestor-or-self::node()[. is $F]] In practice "from" is more important for level="any". Here the test is [. is $F or . >> $F], which works fine for attribute and namespace nodes, so this case does not need to be changed. Michael Kay 6306 1 mike 2005-09-15 22:19:31 +0000 The WG agreed to this change at its meeting on 15 September 2005. The spec has been updated. Michael Kay