Do you see any advantage in turning simple and obvious
operation at the level of XSLT ( (exsl|xt):node-set ) into
something optimization-based?
I'm sorry, I don't understand the assumptions behind this line of
reasoning. To my mind, the RTF in 1.0 was a ghastly mess, with it's
rules that say "you can use it anywhere that a string can be used, it
then behaves like a document node converted to a string, but you can't
use it anywhere you can use a document node". Enforcing these
restrictions was a nightmare and led to really buggy and inefficient
code which I was very happy to throw away. Where exactly do you see the
merits of RTFs?
I must think about it, I don't know how to explain my doubts.
In SAXON 6, I had to switch between tiny tree and normal tree models
to get performance close to XT for various stylesheets (which was
still almost twice faster -- in one particular but rather important
case).
My impression is that restrictions imposed by RTF could be used to
implement RTF generation more efficiently than node-sets.
Since many stylesheets just generate result tree in one pass and
do not convert them into node-sets (in XSLT 1.0), a good implementation
could implement RTF in a much more efficent way than node-set.
David Tolpin
XSL-List info and archive: http://www.mulberrytech.com/xsl/xsl-list