At 3:48 PM -0500 2/10/04, jco...@reutershealth.com wrote:
In addition, recursion is dangerous in Java in library routines that recurse
over user-supplied data, because the per-thread stack is of fixed size,
is typically set globally, and no recovery is possible if it overflows.
OK. Based on this I made that change. Code will be in CVS soon if it
isn't already. However, my benchmarks were not able to measure any
performance difference between the two cases. If there is one, it's
below my resolution.
One thing I've noticed while measuring XOM is that any changes that
don't affect parsing or serialization are pretty much irrelevant. The
times required to walk even a medium sized tree while modifying it
are somewhere below 1 millisecond. Possibly this would be measurable
for large documents, but in that case it's more important to optimize
for space rather than speed. Parsing and serialization are the only
things worth optimizing for speed.
The flip side is that if anyone has suggestions for saving time
during parsing and serialization, send them in. Even a 1 or 2%
improvement should be measurable.
--
25 messages in org.ibiblio.lists.xom-interest[XOM-interest] Three cosmetic patches
