[Nestmates] RFR 8187567: [Nestmates] Enable nestmate changes in javac by default

David Holmes david.holmes at oracle.com
Fri Sep 15 09:02:19 UTC 2017


On 15/09/2017 6:15 PM, Maurizio Cimadamore wrote:
> Looks great.

Thanks for taking a look Maurizio!

> The only thing we might need (not now, maybe later) is a 'reverse' javac 
> flag that forces generation of accessor and/or invokevirtual on private 
> in case you want to generate classfile with latest version number but 
> with old properties.

Yes. If we need it I'll add it. It is a pity the -XDxxx can't be more 
like the hotspot -XX:+xxx and -XX:-xxx flags to allow for enable/disable.

Cheers,
David

> Cheers
> Maurizio
> 
> 
> On 15/09/17 08:12, David Holmes wrote:
>> bug: https://bugs.openjdk.java.net/browse/JDK-8187567
>> webrev**: http://cr.openjdk.java.net/~dholmes/8187567/
>>
>> ** test/runtime/Nestmates/privateConstructors/TestInvokeSpecial.java 
>> contains additional changes for 8187536. They will be checked in 
>> separately.
>>
>> While nestmates will ultimately be associated with a future 
>> source/target version of the JDK and an updated classfile version (see 
>> JDK-8187302) we want to be able to enable them by default as-if 
>> executing in that future release.
>>
>> The primary changes are confined to javac and are very straight-forward:
>> - generate the nestmate attributes in the classfile
>> - use virtual invocation bytecodes for private method invocations
>> - don't generate private accessors
>>
>> With this change in place we can update the hotspot nestmate tests to 
>> not provide explicit "@compile" directives. We also add a test to 
>> verify that invokeinterface is rejected for private interface methods 
>> in pre-nestmate classfile versions (v52 works fine for this).
>>
>> We can now build the JDK with the new nestmate features enabled, and 
>> run tests without needing to provide additional javac options.
>>
>> I also added a missing testcase in 
>> test/runtime/Nestmates/privateMethods/TestInvoke.java (just to ensure 
>> there's no unexpected difference with this$0 usage).
>>
>> Thanks,
>> David
> 


More information about the valhalla-dev mailing list