[RFA] Fix JVMTI/JDWP Exception Handling

[Keith Seitz]

Tom Tromey wrote:
>>>>>> "Keith" == Keith Seitz <keiths@redhat.com> writes:
> 
> Keith> Whoops. I misread that, sorry. Hmm. I thought one could not run
> Keith> compiled AND interpreted together. Perhaps my old age hallucinations
> Keith> are kicking in again?
> 
> Yeah, afraid so.  We mix compiled and interpreted all the time.
> The most basic case is that libgcj itself is compiled :-)
> The bulk of libgcj is not actually special in any way, it is just
> plain old java code...

Ugh, I did it again. I was thinking CNI & interpreted. Of course we run 
with compiled CLASSES and interpreted code. But the result is the same: 
I was planning to treat classes compiled to native the same as a JNI 
native methods. So if one attempts to set a breakpoint on 
java.lang.Object.wait (for example), JDWP/JVMTI would return a "native 
method" error. [Heck, it would bail long before that, but it was just an 
example.]

> Keith> I believe Kyle has already run into one instance where the compiled vs
> Keith> bytecode vs JNI distinction has been forgotten, and gij crashes...
> 
> Ah, I'd like to hear more.

A bug in _Jv_FindInterpreterMethod when you feed it something like 
java.lang.Object. He discovered it earlier today. I'm sure we'll have a 
patch sometime early tomorrow. [It's just a simple bug. How tough can it 
be? (TM)]

Are you sure about the patch? I don't want to commit anything that gives 
you the willies or keeps you up at night...

Keith