RFR: CSR JVM support for records

Harold Seigel harold.seigel at oracle.com
Thu Nov 7 14:08:37 UTC 2019


Thanks Alex!

Harold

On 11/6/2019 6:57 PM, Alex Buckley wrote:
> Thanks for updating. I made small edits (assuming that "The format of 
> the record attribute is consistently checked" is meant to refer to 
> consistency checking a.k.a. format checking, and not to checking being 
> performed on a consistent basis) ... a CSR isn't really the place to 
> suggest that a Record attribute could potentially be useful for X or 
> Y, but it's time to move on.
>
> Alex
>
> On 11/6/2019 11:17 AM, Harold Seigel wrote:
>> Hi Alex,
>>
>> I updated the CSR, hopefully with the info you requested.
>>
>> Thanks, Harold
>>
>> On 11/6/2019 1:28 PM, Alex Buckley wrote:
>>> On 11/6/2019 10:14 AM, Harold Seigel wrote:
>>>> Note that the JVM does consistency check the Records attribute at 
>>>> class load time, not at first use by reflection. So, perhaps this 
>>>> sentence:
>>>>
>>>>     Note that if no reflection is performed then the abstract JVM does
>>>>     not care about the Record attribute in any way.
>>>>
>>>> to something like
>>>>
>>>>     The format of the record attribute is checked even if no 
>>>> reflection
>>>>     is performed.
>>>
>>> Given how Record is described in JVMS 4.7 ("each of these attributes 
>>> must be recognized and correctly read by an implementation of the 
>>> Java Virtual Machine"), it makes sense that the HotSpot JVM is 
>>> format checking a Record attribute at load time. So, yes, please 
>>> make the change you describe above, and please explicitly compare 
>>> Record to the format checking performed for Exceptions, 
>>> InnerClasses, etc, and contrast Record with the lack of format 
>>> checking for MethodParameters, Module, etc.
>>>
>>> Alex


More information about the hotspot-runtime-dev mailing list