XOM API for Processing XML with Java

[Wolfgang Hoschek <wolfgang.hoschek AT mac.com>]

On Oct 31, 2006, at 10:25 AM, Elliotte Harold wrote:

> Wolfgang Hoschek wrote:
>
> > If you'd rather like to expose the namespace map in a strongly  
> > typed manner, fine. For example via getNamespaceCount(),  
> > getNamespacePrefix(int), getNamespaceURI(int).
>
> That's what I was wondering if you're doing. We could actually cut  
> that down to two methods, at some cost in speed by replacing  
> getNamespaceURI(int) with a getNamespacePrefix(int) followed by a  
> lookup. Usually there's less than a dozen mappings (often only or  
> two) so this might not matter a great deal.

Since the XPathContext doesn't maintain the current iteration state,  
it must recreate it for each iteration step. Thus, advancing to the  
next item is O(N). Enumerating all items is O(N^2). I can live with  
it in this particular case, because N is small for all practical  
purposes, and setting up an XPathContext is probably not a high- 
frequency operation, but I wouldn't advocate the pattern. In general,  
simulating a stateful iterator in a stateless way (e.g. via a  
positional index) in a system that's not distributed is a counter- 
intuitive, complicated, non-scalable pattern, and simply overkill for  
what is really a trivial problem. In a distributed (or concurrent)  
system it can make sense because systems with less state are simpler  
to reason, build, maintain and recover from. But XOM isn't a  
distributed system.

> Or I could just return a String array of prefixes, and make it one  
> method call.

Yes, that's much better: Simple, straighforward and relatively  
efficient.

Wolfgang.