XOM API for Processing XML with Java

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

On Feb 6, 2006, at 1:54 PM, Wolfgang Hoschek wrote:

> > You can get a list of the namespaces in scope on an element using an
> > XPath query such as
> >
> > Nodes namespaces = element.query("namespace::node()");
> >
> > This still ascends the tree, but at least you don't have to write  
> > your
> > own code for doing so.
> >
> > I've written a simple benchmark to see just how bad the various
> > methods
> > for iterating over namespaces really are.
> >
> > https://xom.dev.java.net/source/browse/xom/src/nu/xom/benchmarks/
> > NamespaceLister.java?rev=1.1&view=markup
> >
> > I'm not finished yet, but so far I'm not sure there's a practical
> > problem here. In my tests against http://www.rddl.org/ this code is
> > able
> > to retrieve each and every namespace URI, both in scope and declared,
> > for every element in the document in between .1 and .3 seconds. No
> > caching is done. i.e. each element is treated independently of the
> > others. Thus the time to retrieve namespaces declared or in scope for
> > just one element would be a small fraction of that.
>
> That doc is 50KB small, right? If so, that would mean about 500KB/s
> throughput - unusable from my point of view by about 2 orders of
> magnitude for non-XOM-core serialization algorithms related to XPath,
> XQuery, STAX, bnux, etc., as those need to iterate over the declared
> namespaces. I certainly wouldn't use anything like that.
>
> Also note that element.query("namespace::node()") is inefficient for
> reasons that have nothing to do with namespace iteration, but rather
> with XPath setup.
>
> Wolfgang.

I should add that using a data structure along the lines of  
nux.xio.io.NamespacesInScope instead of SAX NamespaceSupport in  
nu.xom.Serializer yields overall serialization speedup of about 1.5  
when using namespaced docs such as nux/samples/data/wsdl-ogsi.xml, as  
measure by BinaryXMLTest and BinaryXMLBench (which also support non- 
binary modes despite their name). NamespacesInScope also performs  
very well for non-namespaced docs.

This is because NamespaceSupport.declarePrefix() is really yet  
another "agressively suboptimal" path for namespaced docs. Apparently  
David Megginson once mentioned the whole class was one of his 4 hour  
hacks... and expressed surprise that folks wanted to seriously reuse  
it, or generalize it even further :-)

Wolfgang.