[
Lists Home |
Date Index |
Thread Index
]
Mike,
> From: Mike Champion [mailto:mc@xegesis.org]
>
> Is the XML InfoSet just too low-level to bother
> trying to expose to ordinary users? Do most people want to
> use a data binding tool to
> hide the XML as a serialization format, or would a clean API
> (e.g., that was integrated with XPath and XSLT, hid bizarre
> syntax such as CDATA sections, and had a sensible model of
> how text was related
> to elements) find some customers?
Definitely. An Infoset Object Model would be received with gratitude.
The thing got do would be to key off the pull based APIs and add
visitation. That way we treat an IOM as an AST. I didn't like the pull
based stuff and said as much here a while back, but I'm coming around.
What I didn't like is the way they use external iteration when a visitor
is a better design choice. Having both is fine, but over time my
codebase will be in better state for using visitors instead of external
iterators.
Piling on convenience functions (ie XPath) into an IOM is bad design.
With an AST data structure, they are properly applied as operations over
that tree. The only excuse for it is trying to build an architectural
franchise around an API's feature set (anyway APIs are not the right way
to lock in developers anymore).
Strictly speaking SAX isn't a proper visitor; you don't control the
descent, the parser does and there's no notions of leaf or position in
SAX. The future of SAX would seem to be a compiler target for a
declarative way of handling infoset events (cf Benoit Marchal's work).
Anyway Elliotte said he's going to sort it out for us ;) I'm betting
he's looking hard at how we walk the tree.
regards
Bill de hÓra
..
Propylon
www.propylon.com
|
