Review request: 8000459: assert(java_lang_String::is_instance(entry)) failure with various mlvm tests

[Jiangli Zhou]

Thanks, Serguei!

Jiangli

On 10/09/2012 07:42 PM, serguei.spitsyn at oracle.com wrote:
> Got it. Thanks, David.
> It means, pseudo string is not a java_lang_String.
>
> Jiangli,
> Thumbs up.
>
> Thanks,
> Serguei
>
>
> On 10/9/12 6:46 PM, David Holmes wrote:
>> On 10/10/2012 11:05 AM, serguei.spitsyn at oracle.com wrote:
>>> Hi Jiangli,
>>>
>>> I'm not very familiar with the pseudo strings, and so, just have a
>>> question:
>>>
>>> *src/share/vm/prims/jvmtiTagMap.cpp*
>>>
>>> 2927 entry = pool->resolved_string_at(i);
>>> 2928 assert(java_lang_String::is_instance(entry) ||
>>> 2929 pool->is_pseudo_string_at(i), "must be string");
>>>
>>> I have a little of doubt the above is correct.
>>>
>>> What value the "entry" will have in a case of pseudo string cp entry?
>>> Just want to understand what constant pool reference will be reported.
>>
>> The pool stores strings and pseudo-strings in the same way. So 
>> pool->resolved_string_at(i) and pool->pseudo_string_at(i, 
>> cp_to_object_index(i)) return the same thing.
>>
>> So entry is either a real string or a pseudo-string - hence the 
>> assert checks for either one.
>>
>> There were a number of ways this code could have been modified to 
>> handle this, but augmenting the assert was simplest.
>>
>>> Do you see all related failing tests passed?
>>>
>>> Would something like the following be more accurate ?
>>> entry = pool->is_pseudo_string_at(i) ?
>>> pool->pseudo_string_at(i, cp_to_object_index(i)) :
>>> pool->resolved_string_at(i);
>>
>> entry is the same no matter which path is taken.
>>
>>> assert(java_lang_String::is_instance(entry));
>>
>> That assert fails if entry was a pseudo-string.
>>
>> David
>> -----
>>
>>> In this case, a reference from CP to string is reported regardless 
>>> it is
>>> pseudo string or not.
>>>
>>>
>>> Thanks,
>>> Serguei
>>>
>>> On 10/9/12 4:26 PM, Jiangli Zhou wrote:
>>>> Please review following fix for 8000459:
>>>>
>>>> http://cr.openjdk.java.net/~jiangli/8000459/webrev/
>>>>
>>>> With JSR292, a constant pool String entry could be patched to a
>>>> non-string oop. The assert in
>>>> VM_HeapWalkOperation::iterate_over_class() needs to be adjusted to
>>>> reflect that.
>>>>
>>>> Thanks,
>>>>
>>>> Jiangli
>>>
>