MethodHandle.invoke* should only work with source 1.7

[Rémi Forax]

On 02/11/2011 01:01 AM, maurizio cimadamore wrote:
> On 10/02/2011 23:03, John Rose wrote:
>> On Feb 10, 2011, at 3:46 AM, Rémi Forax wrote:
>>
>>> The main problem is that when you will run it, with a 1.7 VM because
>>> MethodHandle is only available with a 1.7 VM,
>>> the VM will throw a WrongMethodTypeException because as you said the 292
>>> magic will be ignored.
>>>
>>> I think it's better for the users to don't compile instead of compiling
>>> something which is inherently wrong.
>> Try compiling mh.invokeExact using a 1.6 compiler against a 1.7 rt.jar.  You should get a compile-time error, since the class file version numbers are wrong.  The 1.6 compiler should refuse to load MethodHandle.class from rt.jar.
>>
>> If you compile against a non-standard rt.jar, you might get the error you are observing, but that error won't affect 1.7 users.  If it affects backport users, I suggest taking steps to put the java.dyn package jar in a special place where the 1.6 compiler is not going to see it, by default.
>>
>> There are lots of variations on this theme of version mismatches.  But the class file version number saves us in a lot of them.
> Note also that if you use the JDK 7 compiler with -source 6 flag to
> compile a class that contains MethodHandle, a warning will be emitted
> (not related to MethodHandle though):
>
> warning: [options] bootstrap class path not set in conjunction with
> -source 1.6
> 1 warning
>
> IMHO the above warning should be enough to justify the runtime
> failure... if you follow the warning suggestion and fix bootclasspath
> (to point to JDK 6's rt.jar) then the program won't compile anymore (as
> java.dyn is not available in JDK 6).

A warning is not enough, in my opinion.
Anyway, I wonder why hotspot works when it find a Methodhandle.invoke* 
in a 1.6 classfile.

> Maurizio
>> -- John

Rémi