[
Lists Home |
Date Index |
Thread Index
]
> One implementation is easy, just as proof of concept, but is
> not likely to be optimal. If it's your own flavor of DOM to
> which XPath processing is applied, and the DOM can be assumed
> to be read-only, each node can be constructed to carry its
> index in document order, allowing a trivial sort after nodes
> were found by the Xpath processor, using the DOM
> implementation secret sauce. You could relax the read-only
> assumption at the cost of a major renumbering each time the
> DOM was modified.
Yes, this is why most XSLT processors perform far better on their own
native tree structure than on a third-party DOM, especially a mutable
DOM. It's precisely the scheme Saxon uses (except that the node doesn't
"contain" its index, the tree is actually an array of nodes stored in
document order).
Michael Kay
Software AG
home: Michael.H.Kay@ntlworld.com
work: Michael.Kay@softwareag.com
|