[13] RFR(XXS) 8225369: [AOT] vm/classfmt/cpl/cplres001/cplres00101m004/cplres00101m004.html fails

dean.long at oracle.com dean.long at oracle.com
Fri Jun 21 20:14:07 UTC 2019


If we only have one API, renaming resolvePossiblyCachedConstantInPool to 
resolveConstantInPool would make it more consistent with other resolve 
methods.  What do you think?

dl

On 6/21/19 1:09 PM, dean.long at oracle.com wrote:
> On 6/21/19 9:42 AM, Tom Rodriguez wrote:
>> Doesn't that make resolveConstantInPool completely unused?
>
> Yes, except for a jtreg test, which can be removed.
>
>> It should probably be deleted and the javadoc updated for 
>> resolvePossiblyCachedConstantInPool.
>
> OK.
>
> dl
>
>>
>> tom
>>
>> dean.long at oracle.com wrote on 6/18/19 10:52 AM:
>>> https://bugs.openjdk.java.net/browse/JDK-8225369
>>> http://cr.openjdk.java.net/~dlong/8225369/webrev/
>>>
>>> This test triggers a race, which can result in different resolved 
>>> values of the MethodHandle constant.  The VM chooses a winner and 
>>> stores that value in the constant pool cache, however Graal is not 
>>> checking the cache.  The trivial fix is to use the proper API that 
>>> goes through the cache.
>>>
>>> dl
>



More information about the hotspot-compiler-dev mailing list