a JNI backtrace

Thu Mar 22 22:17:00 GMT 2001

Bryce> It looks like a JNI bug to me. I suspect _Jv_JNI_PopLocalFrame
Bryce> is possibly passing some sort of invalid pointer (but not null)
Bryce> to unmark_for_gc(), which is subsequently calling
Bryce> Hashtable.get() with a garbage pointer. You may need to use
Bryce> your new GDB to poke around and figure out if this theory is
Bryce> correct and if so, where the garbage in env->locals comes from.
The mark/unmark_for_gc code is gross. We should replace it at our
earliest opportunity.
Marcus, unless you have a relatively simple test case I'm afraid
you'll have to debug this one on your own...
Possibly an object is being reclaimed early. That could cause hash
table operations using that object as a key to fail. You could do
something like set up a finalizer that calls `fail_on_finalization' in
mark_for_gc.
Tom