OASIS Mailing List ArchivesView the OASIS mailing list archive below
or browse/search using MarkMail.


Help: OASIS Mailing Lists Help | MarkMail Help



   Re: XML Memory Requirements (was Re: Feeling good about SML)

[ Lists Home | Date Index | Thread Index ]
  • From: Tyler Baker <tyler@infinet.com>
  • To: David Brownell <david-b@pacbell.net>
  • Date: Thu, 18 Nov 1999 19:28:43 -0500

David Brownell wrote:

> Tim Bray wrote:
> >
> > The performance is good enough that in Lark, the limiting factor was
> > my lousy input buffering, not the character class checking; and Lark
> > was always in the top half of the performance table. -Tim
> I think that's pretty typical.  I've profiled several parsers by now,
> and character class checking hasn't been a major issue ... usually
> one or two percentage points CPU time for testing whether a character
> works in a name (first/rest) or not.
> The technique I used in Sun's parser may be good for many folk to steal.
> It involves using the standard Character.getType() method (which has
> access to lots of Unicode tables, and in recent JVMs uses native code
> to quickly access them) and then filtering that output by the rules in
> the XML spec.

That is one way to do it. What I found worked optimally in my case is to pregenerate a bitmap
of all legal name and name start characters, save it to a file, and load it when using the XML
parser into a constant array of integers (integer arrays in Java are faster to access because
comparisons don't require a cast after retrieval). It takes up a little bit of extra memory
but I think it is well worth it. Other Java XML parsers I have seen use a similiar approach
except that the store the character bitmap in the class file itself which has the unfortunate
problem of bloating the size of your class file immensely, and slowing down startup time
because the stupid way Java deals with constant arrays of primitive types in classes is that
it generates instructions for initializing each section of the array to the desired value
instead of just having a stream of bytes that are read into the array one by one. I opted to
initialize the array of name and namestart characters from a file because the startup time for
the class was considerable as you needed to loop 128,000 times and test each index against a
isNameChar() function or a isNameStartChar() function or something like that.

I could use this method for testing whitespace as well as well as other characters, but it
hardly seemed worthwhile because all of the whitespace characters can be tested with a
switch/case statement and testing for all of the other special characters do not occur enough
in an XML document to mandate adding another 2048 integers into memory.


xml-dev: A list for W3C XML Developers. To post, mailto:xml-dev@ic.ac.uk
Archived as: http://www.lists.ic.ac.uk/hypermail/xml-dev/ and on CD-ROM/ISBN 981-02-3594-1
To unsubscribe, mailto:majordomo@ic.ac.uk the following message;
unsubscribe xml-dev
To subscribe to the digests, mailto:majordomo@ic.ac.uk the following message;
subscribe xml-dev-digest
List coordinator, Henry Rzepa (mailto:rzepa@ic.ac.uk)


News | XML in Industry | Calendar | XML Registry
Marketplace | Resources | MyXML.org | Sponsors | Privacy Statement

Copyright 2001 XML.org. This site is hosted by OASIS