[XMLBEANS-300] Threading issue leads to NPE in SchemaComponent$Ref.getComponent

• Type: Bug
• Status: Resolved
• Priority: Major
• Resolution: Fixed
• Affects Version/s: Version 2, Version 2.1, Version 2.2
• Fix Version/s: Version 2.3
• Component/s: XmlObject
• Assignee: Unassigned
• Reporter: Gordon Eastman
• Created: 05/Dec/06 18:49
• Updated: 27/Apr/07 00:21
• Resolved: 27/Apr/07 00:21

Description

We have seen this NPE during normal use of our application on a MP Solaris system using XmlBeans 2.0.0. I tested and confirmed it is still a problem in 2.2.0:

java.lang.NullPointerException
at org.apache.xmlbeans.SchemaComponent$Ref.getComponent(SchemaComponent.java:104)
at org.apache.xmlbeans.SchemaType$Ref.get(SchemaType.java:872)
at org.apache.xmlbeans.impl.schema.SchemaPropertyImpl.getType(SchemaPropertyImpl.java:92)
at org.apache.xmlbeans.impl.schema.SchemaTypeImpl.createElementType(SchemaTypeImpl.java:965)
at org.apache.xmlbeans.impl.values.XmlObjectBase.create_element_user(XmlObjectBase.java:895)
at org.apache.xmlbeans.impl.store.Xobj.getUser(Xobj.java:1592)
at org.apache.xmlbeans.impl.store.Cur.getUser(Cur.java:2627)
at org.apache.xmlbeans.impl.store.Cur.getObject(Cur.java:2620)
at org.apache.xmlbeans.impl.store.Cursor._getObject(Cursor.java:983)
at org.apache.xmlbeans.impl.store.Cursor.getObject(Cursor.java:2896)

Here is how to reproduce with the debugger:

• start two threads, each of which loads a different xml document with the same schema type
• each thread will traverse its document, calling getObject at each element
• break before traversal starts
• insert breakpoint at SchemaComponent.java:104, and let both threads run to this point
• resume one thread, and then the other
• the second thread fails with the NPE

Here is the code fragment from SchemaComponent $Ref:

public final SchemaComponent getComponent()
{
if (_schemaComponent == null && _handle != null)

{ _schemaComponent = _schemaTypeSystem.resolveHandle(_handle); // this is line 104 _schemaTypeSystem = null; }

return _schemaComponent;
}

Both threads use the same Ref object, and there is no protection between use of _schemaTypeSystem on line 104 and setting it null on the next line

I do not know enough about how thread protection of schema types is designed to know if a localized fix to synchronize the lazy instantiation is appropriate, or if synchronization at a higher level is appropriate.