Looks OK. One comment, should we include the CR<id> in comment? I
remember it should not be there.
--Yumin
On 1/12/2012 12:24 PM, Bill Pittore wrote:
Hi David,
Thanks for forwarding this to the SA list. I did run the nsk/jvmti
tests from the UTE area on sqenfs-1.us.oracle.com. There were no new
failures with this change.
bill
On 1/12/2012 2:54 AM, David Holmes wrote:
Hi Bill,
I believe this should have gone out to the Serviceability list
(cc'ed) instead of, or perhaps as well as, the runtime list.
The change looks okay to me in that is does what you described it
would. With JVMTI the proof-of-the-pudding is always in the testing
and I assume the various JVMTI test suites have been thoroughly
exercised ?
Thanks,
David
On 11/01/2012 8:40 AM, Bill Pittore wrote:
Webrev for fix for 6972759 is at
http://cr.openjdk.java.net/~bpittore/6972759/webrev.00/
This bug has to do with single stepping after hitting an exception
breakpoint. The short story is that JVMTI maintains some thread state
information that was not updated correctly if a frame was popped off
the
stack. When you pop a frame the exception state should go back to
_exception_detected=false. Otherwise when MethodExit event is sent at
some time in the future the 'was_popped_by_exception' flag will be set
and the agent receiving the event may not re-enable single stepping
properly. This problem will also affect ForceEarlyReturn code.
Hacked up
a couple of existing nsk/jvmti tests to issue the necessary sequence of
JVMTI calls/events in order to test original bug and this fix.
bill
