Crash with invokeDynamic on MethodMissing attempt
John.Rose at Sun.COM
Thu May 21 04:59:45 PDT 2009
Yuck, a leftover from the invokeinterface encoding. It's even clearly
marked my code-distress marker (@@@@).
Yes, your fix looks OK....
But, the @@@@ line wants to be something better, and I think there are
better entry points in constantPoolOop since I refactored the logic
there. Try to use cp->signature_ref_at() or at least
nt_signature_ref_at(). (You can remove the distress marker.)
On May 21, 2009, at 4:31 AM, Christian Thalinger wrote:
> On Thu, 2009-05-21 at 11:32 +0200, Christian Thalinger wrote:
>> On Wed, 2009-05-20 at 21:41 -0700, John Rose wrote:
>>> Thanks. We have seen problems with indy in the JIT; that's
>>> probably it.
>>> (We're working on the JIT support right now.)
>>> Try running with java -Xint; does that make the crash go away?
>> Yes and no. Indeed it crashes in the compiler, but when running with
>> the interpreter it asserts too. The problem is
>> GenerateOopMap::do_method. That's the same point where I'm currently
>> hanging in the compiler. Let's see if I can come up with a fix...
> I think I have a fix for it. John, is that correct?
More information about the mlvm-dev