Probable bug in Kodo class handling code (Beta 2.2)

Question

I have a test class "Widget" that has two attributes, name and
mark. I am encountering a curious error in the no-args
constructor under some circumstances that looks like a bug.
public class Widget
private static Random random = new Random();
private String name;
private int mark;
private Widget()
public Widget(String name)
this.name = name;
mark = random.nextInt(1000);
Typically, the test application constructs Widgets with a name.
But since I believe that it is good practice for the application
to define the no-args constructor even when it is only for the
use of JDO, I have a private no-args constructor. As the class
stands above, it compiles, enhances, and runs just fine.
But if I copy the line "mark = random.nextInt(1000);" into the
no-args constructor, the code compiles and enhances just fine,
but bombs on running with the following error:
javax.jdo.JDOFatalDataStoreException:
The registered class "com.ysoft.jdo.book.widget.Widget"
is not compiled or not longer exists. If the class has been deleted,
unregister it before proceeding.
NestedThrowables:
java.lang.ExceptionInInitializerError
     at
com.solarmetric.kodo.impl.jdbc.schema.DB.getPersistentTypes(DB.java:270)
     at com.solarmetric.kodo.impl.jdbc.JDBCPersistenceManagerFactory.setup
(JDBCPersistenceManagerFactory.java:170)
     at
com.solarmetric.kodo.runtime.PersistenceManagerFactoryImpl.privateSetup
(PersistenceManagerFactoryImpl.java:501)
     at
com.solarmetric.kodo.runtime.PersistenceManagerFactoryImpl.getPersistenceManager
(PersistenceManagerFactoryImpl.java:61)
     at
com.solarmetric.kodo.runtime.PersistenceManagerFactoryImpl.getPersistenceManager
(PersistenceManagerFactoryImpl.java:50)
     at
com.ysoft.jdo.book.widget.WidgetHandler.<init>(WidgetHandler.java:43)
     at com.ysoft.jdo.book.test.widget.TestWidget.main(TestWidget.java:18)
Exception in thread "main"
Using the schematool for -action refresh will yield a null pointer
exception at the no-args constructor line that is setting the
value of mark.
After initially having the mark initialization code in the
no-args constructor and puzzling over the error, I realized that
it didn't make sense to set the value of mark in the no-args
constructor, since JDO would later set it to whatever the value
in the database was. But it seems to me that there should be no
great harm in it, either. If mark is initialized with the value
"42" instead of "random.nextInt(1000)", there are no problems.
Any insights?
David Ezzio
Yankee Software

Answer 1

David,
I'm not quite sure what to do about this. I'll have to dig into the spec
and get back to you.
Here's what's happening:
Upon compilation, your initialization of the static field 'random' is
inserted into a 'static' block.
Upon bytecode enhancement, a good deal of code is added to the 'static'
block. This code is responsible for registering the class with the JDO
system.
One step of this registration process involves instantiating an object
of type 'Widget' and passing it to JDOImplHelper.registerClass().
This step is inserted into the static block before the initialization
of 'random'.
So, when the no-args constructor is invoked in preparation for the call
to registerClass(), 'random' is still null. Hence the null pointer
instruction.
As a temporary workaround, move the 'random' initialization code into an
init () method that is invoked from the constructors, and which
initializes 'random' if it is null. Or just don't perform the
initialization in the no-args constructor.
-Patrick Linskey