On Sat, Jan 31, 2004 at 05:50:22PM -0000, Michael Kay wrote:
My experience is that node-set() is a badly broken
extension to XSLT-1.0
I think that describing an extension as broken because you have
difficulty implementing it using your chosen implementation technology
is a little unreasonable.
Well the Working Group differenciated Result Value Tree from the
node set when producing XSLT-1.0 . It was guided by implementation
issues IMHO, from that perspective node-set() runs againts that
decision, and even worse allowing RVT to be directly usable where
it shouldn't really confused people. That is where the people
started to expect this to work on XSLT processors, develop
algorithms based on this and considering this a "feature" of the
language. I think I did complain clearly and directly about this,
but that was an implementation issue, so I complained here and not
in the XSLT-1.0 spec comment list because well it wasn't really
an XSLT-1.0 problem.
I was not aware before now that anyone opposed this facility. I'm
surprised that if you object to it so strongly, you haven't made your
views known until now. Given that there are many successful
Ohhh I think it's not the first time I complain about it :-)
implementations, I don't think the working group would be greatly swayed
by an argument based on implementability; but there have been other
features (like xx:evaluate()) excluded on similar grounds and you should
at least try to make your views known at the right time in the right
place.
Well arguing w.r.t. XSLT-1.0 is useless because node-set() is not
in that spec. Arguing against 1.1, I did a long time ago and had to wait
months and months before the deprecated Working Draft finally got flagged
as such, at the time I was on the W3C XML CG !
W.r.t. XSLT-2.0 there is more issues than simply node-set(), implementation
are anyway very different and expecting to keep 1.0 design issues
isn't technically sound, XSLT-2.0 will be reimplementations not simple
evolutions so complaining about node-set() being a big change in comparison
to XSLT-1.0 implementation design makes little sense anyway.
I stated my view on 2.0 on this list and I think I CC'ed the
comment list at the time, but it was useless to point at node-set()
specifically.
Finally I don't see myself argue against EXSLT:node-set(),
it's clearly an extension, so the "XSLT-1.0 compatibility" viewpoint
can't really hold.
I think I did made my views known at the right time and in the
right places, i.e. where that made sense. I think I also complained
to Sharon and you at the time when Saxon used liberally RVT as
node-set inputs.
Daniel
--
Daniel Veillard | libxml Gnome XML XSLT toolkit http://xmlsoft.org/
daniel(_at_)veillard(_dot_)com | Rpmfind RPM search engine http://rpmfind.net/
http://veillard.com/ |
XSL-List info and archive: http://www.mulberrytech.com/xsl/xsl-list