It could be the extra position predicate that's causing the problem. Are
you running under release or debug? Debug will be _much_ slower.
This is an architectural deficiency of the C++ processor, because it uses
exceptions to restart an XPath if it detects a position predicate.
If you can send me your document privately, zipped up, I'll take a look at
what's happening and see if there isn't maybe something else going on.
Dave
"Cornelius, Martin"
<Martin.Cornelius@softw To:
"'[EMAIL PROTECTED]'"
areag.com> <[EMAIL PROTECTED]>
cc: (bcc: David N
Bertoni/CAM/Lotus)
07/19/2001 08:59 AM Subject: Xalan-C++ 1.1:
Catastrophic
Please respond to performance with prededing:: Axis
xalan-dev
Hi there, has anybody else noticed that the performance of Xalans-C++ Xpath
evaluation is degrading catastrophically with large documents, if the
preceding:: axis is used ?
I have built an application that uses only the Xpath component of Xalan (
no
XSLT ), to find nodes in an XML Document.
The Sample-Document has about 174000 Element nodes, maximum depth of
nesting
is about 20.
Searching for the 99999.th node in forward direction is quite fast:
(//*)[99999]
takes about 1.2 seconds (on a 700 MHz PentiumIII under WinNT) for the pure
Xpath Evaluation, ( plus 4 seconds for parsing the Document )
In contrast,the question for the same node, formulated as:
(//*)[100000]/preceding::*[1]
takes about 75 seconds, what is 2 ordes of magnitude slower.
BTW, Saxon needs about 5 Seconds for the complete Parsing and evaluation
for
both queries.
