RE: Removing namespaces without escaping CDATA???

> While I haven't implemented an engine myself, I can imagine 
> it would be architecturally very disruptive since you are 
> conflating syntax with structure for this instruction.  Right 
> now the <xsl:element> instruction and all of its attributes 
> construct result nodes in the result tree, while saying 
> nothing about serialization syntax.  Adding your suggested 
> change would be introducing syntactic awareness to the 
> construction of the result tree for this instruction.
> 
> I acknowledge this is happening with the <xsl:text> 
> instruction and d-o-e, but I don't know of any other result 
> tree construction instruction that impacts on serialization.
> 

Correct analysis. The architectural problem with d-o-e is that it requires
out-of-band communication between the XSLT processor and the serializer.
Instead of just passing a result tree and a set of serialization parameters
from the transformer to the serializer, there is a need to annotate
individual nodes of the tree with serialization properties (or in the case
of d-o-e, individual characters in a text node). This proposal has the same
problem.

Michael Kay
http://www.saxonica.com/