Poor performance when processing schemas with high values of maxOccurs

• Key: XERCESJ-773
• Type: Bug
• Status: Resolved
• Resolution: Fixed
• Priority: Blocker
• Assignee: Michael Glavassevich
• Reporter: Steve Baker
• Component/s: XML Schema 1.0 Structures
• Affects Version/s: 2.5.0
• Fix Version/s: 2.9.1

Environment

Operating System: Linux
Platform: PC

Description

When validating an instance of an XML order against its schema, the process took
 5 minutes 7 seconds and required an increased maximum heap size to run to
completion.

The problem was traced to a maxOccurs value of 9999 on an element (the
OrderLine). Changing this to unbounded reduced the time taked to 1.16 seconds
and required no increase in the default heap size.

Comments

[Michael Glavassevich]

I've just committed an improvement to code in SVN which covers a number of common cases. We now build a representation of large minOccurs/maxOccurs in constant time and memory (which uses a counter during validation) for element and wildcard particles when each model group particle in the content model:

* has minOccurs/maxOccurs == 1; or
* contains only one element/wildcard particle with minOccurs/maxOccurs == 1

Handling sequences, choices and nested minOccurs/maxOccurs is somewhat tricker. We would need a more general solution than the one implemented for elements and wildcards to process those gracefully. I've opened a new JIRA issue [1] for tracking that.

[1] http://issues.apache.org/jira/browse/XERCESJ-1227