On 7/26/17 12:11 AM, David Holmes wrote:
On 26/07/2017 10:27 AM, Yasumasa Suenaga wrote:
Hi Dan,
I've added some analysis to the bug report
Thanks!
I tried to fix this issue in JvmtiEnvBase::get_owned_monitors() at
first.
But it is difficult because we cannot know pending monitor if thread
state is MONITOR_CONTENDED_ENTER when get_owned_monitor() is called.
I need to look closer at this when I get back from vacation next week.
Seems like you're back already. :-)
A pending monitor should not be reported as owned (unless the spec
says otherwise) and it seems odd to me to fix the current problem by
marking the monitor as pending earlier.
It's the updating of the _current_pending_monitor field that
allows JvmtiEnvBase::get_locked_objects_in_frame() to recognize
that the monitor observed in the frame is only pending and
is not owned.
I put a fairly detailed note in the bug yesterday, but you
should look at that when you're officially back!
Dan
Thanks,
David
Did you run the jdk repo's com/sun/jdi tests with your fix?
I have not done yet.
I have a trip until 28 July JST. So I will run it after that.
Yasumasa
On 2017/07/26 7:05, Daniel D. Daugherty wrote:
On 7/24/17 8:40 PM, Yasumasa Suenaga wrote:
Hi all,
I tried to get owned monitors in MonitorContendedEnter JVMTI event
handler.
However GetOwnedMonitorInfo JVMTI function returns a monitor which is
not yet owned.
I attached reproducer to JBS. Please read README.md.
I think GetOwnedMonitorInfo() should not return a pending monitor.
I uploaded webrev. Could you review?
http://cr.openjdk.java.net/~ysuenaga/JDK-8185164/webrev.00/
I hope this fix is applied to 8u or later release.
I cannot access JPRT. So I need a sponsor.
Thanks for the bug report. It's nice to have a test case and a proposed
fix all in the bug report! I've added some analysis to the bug report
and we'll need to run this fix through Oracle's JPDA test stack which
is not (yet) open.
Did you run the jdk repo's com/sun/jdi tests with your fix?
Dan
Thanks,
Yasumasa
