RE: [boost] Proposal: Regression, XML, and Unicode Game Plan

From: Asger Alstrup Nielsen (alstrup_at_[hidden])
Date: 2001-09-27 02:54:33

> The main
> problem with xerces-c implementation is that it is a JAVA
> implementation remapped to C++.

Please note that in xerces 1.5.1, they have introduced an experimental
IDOM interface, which is native C++ based, rather than a port of the
Java interface (which is specified in the W3C specifications.)

I quote the motivation from

http://xml.apache.org/xerces-c/program.html#IDOMProgGuide

--
The performance of the C++ DOM has not been as good as it might be,
especially for use in server style applications. The DOM's reference
counted automatic memory management has been the biggest time consumer.
The situation becomes worse when running multi-threaded applications. 
The experimental C++ IDOM is a new alternative to the C++ DOM, and aims
at meeting the following requirements: 
- Reduced memory footprint. 
- Fast. 
- Good scalability on multiprocessor systems. 
- More C++ like and less Java like. 
--
The Apache tools are still restricted by their aim for maximum
portability, but you should still have a look at what they have done.
After all, they have a lot of field experience with XML parsers and
interfaces.
I propose a different way of attacking this problem: The main issue is
the new C++ interface, not the implementation. Therefore, it might make
sense to try to come up with an interface, and then use xerces to
implement this.
Once the interface is somewhat stable, an efficient, "modern"
reimplementation could be considered.
Greets,
Asger Alstrup Nielsen