This mailing list entry describes the problem.
This problem is new with GCJ 3.3, 3.2 works fine.
I'm struggling with PR 10746, and before that is resolved, producing an easily reproducable case to fit into this report is going to be difficult.
I thought I'd report this problem regardless.
Created attachment 4627 [details]
This is a snapshot from the debugger. As you can see getClass() is trying to
access address 0.
Presumably it is trying to fish out the vtable from the class object...
Does this happen on the mainline after bug 10746 was fixed?
*** This bug has been marked as a duplicate of 10746 ***