[prev in list] [next in list] [prev in thread] [next in thread] 

List:       openjdk-serviceability-dev
Subject:    Re: JDK-8203350: Crash in vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/TestDescription.jav
From:       Chris Plummer <chris.plummer () oracle ! com>
Date:       2018-09-28 20:17:51
Message-ID: 33e4df93-2f6a-3eb6-aded-cb4c26f6fe0b () oracle ! com
[Download RAW message or body]

Hi Gary,

If you haven't been able to reproduce the failure and just want to take 
it off the problemlist, I'm ok with that. I'd rather not see any test 
changes that could potentially hide the crash. If it eventually fails 
for some other reason and it can be shown that it is testbug due to 
improper classloading, then I can see adding the change you suggested.

Chris

On 9/28/18 5:17 AM, Gary Adams wrote:
> At this point my first recommendation is to
> simply remove the test from the ProblemList,
> since it has not been reproduced after 1000s of
> testruns with debug and release builds.
>
> What is not clear to me at the moment is -
> was it intentional that hs201t001 intentionally placed
> the classloader operation outside of doInit() and
> hs201t002 placed the classloader operation in the
> doInit() method. It appears that the primary
> purpose of these two tests is the difference in behavior
> for IsMethodObsolete.
>
> My best guess is that at some point the classloader operation
> was failing after a RedefineClasses, and a workaround
> was introduced into hs201t001. Later on the same
> issue was observed with hs201t002, which did not have the
> same workaround in place.
>
> My primary goal is to make the test reliable for
> functionality it is intended to exercise. My secondary
> goal is to have some consistency   between these
> similar tests.
>
> From the discussions in the related issues, the
> crashes were not able to be reproduced so the issues
> were resolved as CNR. It was presumed that some other
> fix accounted for the problem no longer being visible.
>
> On 9/27/18, 2:34 PM, Chris Plummer wrote:
>> Hi Gary,
>>
>> Aren't you just hiding a potential jvmti bug with this change? If you 
>> think this is a test bug and this is a proper fix, I'd like to see an 
>> explanation of how the test is causing the crash. The explanation 
>> would need to involve native code, since pure java should never crash.
>>
>> thanks,
>>
>> Chris
>>
>> On 9/27/18 5:18 AM, Gary Adams wrote:
>>> I've been unsuccessful trying to reproduce the failure in hs201t002.
>>>
>>>    Issue: https://bugs.openjdk.java.net/browse/JDK-8203350
>>>
>>> Colleen made a comment on the bug that the reference
>>> from hs201t002a to class hs201t002 might be an issue
>>> for the redefined class.
>>>
>>> I found in test hs201t001 that an intentional reference
>>> before entering hs201t001a.doInit() is made to avoid
>>> that classloader operation.
>>>
>>> It's not clear to me why that was done, but the same workaround
>>> could be used in hs201t002a, if it would make the test more robust.
>>>
>>>
>>> diff --git 
>>> a/test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/newclass/hs201t002a.java 
>>> b/test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/newclass/hs201t002a.java 
>>>
>>> --- 
>>> a/test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/newclass/hs201t002a.java
>>> +++ 
>>> b/test/hotspot/jtreg/vmTestbase/nsk/jvmti/scenarios/hotswap/HS201/hs201t002/newclass/hs201t002a.java
>>> @@ -26,6 +26,7 @@
>>>   public class hs201t002a extends Exception {
>>>
>>>          public hs201t002a () {
>>> +               System.out.println("Current step: " + 
>>> hs201t002.currentStep); // Avoid calling classloader to find 
>>> hs201t002 in doInit()
>>>                  doInit();
>>>          }
>>>
>>
>>
>


[prev in list] [next in list] [prev in thread] [next in thread]