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

[Jiangli Zhou]

Hi Serguei,

Thanks for the review! Please see comments below.

On 10/09/2012 06:05 PM, 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.
> 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);
> assert(java_lang_String::is_instance(entry));
>
> In this case, a reference from CP to string is reported regardless it 
> is pseudo string or not.

Those are very good questions. Here is an example of the specific case 
with a generated JSR292 java.lang.invoke.LambdaForm$MH class. The follow 
are two constant pool entries of the class. Entry 16 is a 
"sun/misc/Unsafe" String, which could be patched using a sun.misc.Unsafe 
klass oop.

         ...
         (0016) tag=Utf8 "sun/misc/Unsafe"
         (0017) tag=Class "sun/misc/Unsafe"
         ...

I discussed with John Rose about different ways to fix the assertion 
issue. The code you have above uses pseudo_string_at() to access the 
'entry', which is similar to one of the version we discussed. The 
current change was chosen as it's the simplest solution.

Thanks,

Jiangli
>
>
> 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
>