RFR(XS): 8209736: runtime/RedefineTests/ModifyAnonymous.java fails with NullPointerException when running in CDS mode

[Jiangli Zhou]

> On Aug 31, 2018, at 2:54 PM, Calvin Cheung <calvin.cheung at oracle.com> wrote:
> 
> 
> 
>> On 8/31/18, 1:52 PM, Jiangli Zhou wrote:
>> Hi Calvin,
>> 
>> 
>>> On 8/31/18 1:33 PM, Calvin Cheung wrote:
>>> Hi Jiangli,
>>> 
>>>> On 8/31/18, 1:10 PM, Jiangli Zhou wrote:
>>>> Hi Calvin,
>>>> 
>>>> How about use a separate loop (before the main test loop in runTest()) to print out the class names in the allLoadedClasses array?
>>> 109     static void runTest() {
>>> 110         while (!done) {
>>> 111             Class[] allLoadedClasses = inst.getAllLoadedClasses();
>>> 112             int len = allLoadedClasses.length;
>>> 113             for (int idx = 0; idx < len; idx++) {
>>> 
>>> Did you mean print out all the class names after line 111? There would be too much output since there's a while loop enclosing it.
>>> Or did you mean something else?
>> Yes. If the failure does re-manifest again, we can compare the output with a normal run. That would give some useful information. How big is the allLoadedClasses?
> On my local linux machine, it only loops through the while loop twice and I saw more than 1800 class names in the output.
>> 
>> Or, maybe we could print the classes if null is detected.
> This is a better option.

Probably dumping to a file.

>>>> If null is discovered, throw a RuntimeException(). That helps to provide more debugging information and also will not silent the issue that causes the NPE if it does' resurface again.
>>> I don't think throwing a RuntimeException() gives anymore info than the existing NPE.
>>> The idea of this change is that after a null clazz is encountered, let the test continue to run. It may fail in a different way which may give us more clues.
>> The NPE might be caused by a rare race condition, which might not affect the redefineClass process and the test could pass with the change. We should avoid such case.
> Maybe but we don't know for sure. The main purpose of the test is to make sure an anonymous class cannot be retransformed or redefined. So I don't think we're masking any problem if a null class is encountered and the test passed.

The issue is with getAllLoadedClasses. The proposed change would make the issue not being detected. I think we need to understand what’s the expected behavior of the API.

Thanks,
Jiangli 
> 
> thanks,
> Calvin
>> 
>> Thanks,
>> Jiangli
>>> 
>>> thanks,
>>> Calvin
>>>> 
>>>> Thanks,
>>>> 
>>>> Jiangli
>>>> 
>>>> 
>>>>> On 8/31/18 11:33 AM, Calvin Cheung wrote:
>>>>> bug: https://bugs.openjdk.java.net/browse/JDK-8209736
>>>>> 
>>>>> webrev: http://cr.openjdk.java.net/~ccheung/8209736/webrev.00/
>>>>> 
>>>>> The test failure is no longer reproducible; it happened only once in tier3 and once in tier6 testing.
>>>>> This simple change is just adding a null check on the variable clazz and prints the array size and index if it is null.
>>>>> 
>>>>> Testing: hs-tier{1,2,3} with default CDS archive in JDK.
>>>>> 
>>>>> thanks,
>>>>> Calvin
>>>> 
>>