Re: Server JVM

Jimmy Zhang wrote:
> I can help you set up server JVM, just a dll file you need to put in
> the right place... let me know if you need any help

I installed the JDK and tested with the server JVM.
It does indeed increase performance significantly, beating Expat on files
with lots of markup vs character data (27ms for benchmark_BR vs 49ms for 
Expat).
Question, why should one not always use the "buffer-reuse" version?

Expat was still faster on the file with lots of character data (13ms vs 
17ms for benchmark_BR).
I recompiled the Expat library with all optimizations for speed.

Now given that vtd-xml is quite fast, you still have to prove two things:

1) It does everything a conforming non-validating parser must do. How 
many of
the tests in the XML-Test-Suite does it pass? Expat passes all
the tests for a non-validating parser except a handful that are optional 
or in doubt.
Example: vtd-xml failed to detect duplicate attributes when they had
different prefixes pointing to the same namespace. Completing vtd-xml
to conform as well as Expat may well add more overhead.

2) To which degree does it pay off to delay the work of encoding 
transformation
to the point when the data is actually needed, as in a real-world 
application?
If the document is encoded in UTF-8 and your application requires UTF-16,
then this is already done by Expat, but for vtd-xml this work still has 
to be performed.

Whether it will be preferable over Expat for documents where memory
usage is not an issue, will depend on the answers to these questions.

Overall I do like your approach, and I think it is excellent for random 
access with an
in-memory document. It may also do very well on SOAP processing for 
smaller messages.

Karl