XOM API for Processing XML with Java

[Elliotte Rusty Harold <elharo AT metalab.unc.edu>]

At 8:10 PM -0500 5/9/03, Bradley S. Huffman wrote:


> I'm assuming you mean spped when building from a SAX source. First in XOM,
> get rid of the parent stack, it's not needed, you can call getParent() in
> endElement(...).

I don't see how that would work. getParent() returns null before I've 
added it to the parent. That's why I need the parent stack. I could 
move more work into endElement() but I'm not sure how that would help.

>  Also I think you can add attributes and namespace
> declarations in one loop instead of two.

Good idea. Probably a small improvement, but couldn't hurt.

> Now obviously, what would be ideal is to know the number of children at
> element creation, but that's not going to happen with SAX. So here's a
> wild idea, have a large buffer to hold newly created nodes.  In all node
> creation methods (startElement, comment, flush*(), etc.) buffer the newly
> created node but **don't** add it to the current parent (as JDOM and XOM
> currently do), instead just increment the child count of the current parent
> (either find some way to use the parent to hold this info, or stack it).
> Finally in endElement(...) create the children list in the current
> parent according to the current child count and move nodes from the buffer
> to the parent. This should eliminate most unnecessary reallocation and
> copying of nodes. Hmmm, guess it would also eliminate the wasted overhead :)

Sounds pretty complex. Right now my tests are showing this is not 
where my hot spots are during building. Verification and parsing 
overhead just swamp this. More on that shortly.
--

  Elliotte Rusty Harold
  elharo AT metalab.unc.edu
  Processing XML with Java (Addison-Wesley, 2002)
  http://www.cafeconleche.org/books/xmljava
  http://www.amazon.com/exec/obidos/ISBN%3D0201771861/cafeaulaitA