[jboss-dev] JBAS-7650 - jboss-cl 2.2.x upgrade WAS Re: Planning to revert

[Adrian Brock]

On Mon, 2010-02-01 at 13:36 +0100, Adrian Brock wrote:
> Obviously, I still need to figure out why it is failing with
> useLoadClassForParent=false
> That's just an optimization to avoid locking the
> bootstrap classloader too much. It should have the same
> semantics.
> https://jira.jboss.org/jira/browse/JBCL-142

It took me a while to figure this out because it occurs
while the logging is getting reconfigured, meaning it was
difficult to get information about what was happening. :-)

I'm not really sure how to fix it?

The problem only occurs when you have a security manager
installed and you are logging access control requests.

What is happening is that stdio has just captured
System.out/err. At the next access control logging
stdio tries to write this to the logging framework.

It needs java.lang.Math to do this - see write():
http://viewvc.jboss.org/cgi-bin/viewvc.cgi/common/jboss-stdio/trunk/src/main/java/org/jboss/stdio/WriterOutputStream.java?revision=3915&view=markup
Its the first such event, so it needs to load the class.

But while it is loading the class, jboss-cl does a
getResource("java/lang/Math.class") which also
does some access control logging
which will again get intercepted by stdio.

It then of course goes through write()
java.lang.Math is not yet loaded so it repeats.

Its reported as a ClassCircularityError
as Brian said, but its really a recursion problem.

If you use loadClass() directly, there is no
logging of resource access so you don't get the problem.

-- 
xxxxxxxxxxxxxxxxxxxxx
Adrian Brock
Chief Scientist
JBoss by Redhat
xxxxxxxxxxxxxxxxxxxxx