What is the reason for the additional indirection in Handle?

[Paul Hohensee]

Yes, you can pass oops from Java -> native, but they're handlized and you
aren't supposed to use one to access a java object except via JNI calls.
The latter check for a safepoint in progress on entry and block if so.  If
that didn't happen, then you could end up with a stale oop in a register
someplace.  The vm also checks for a safepoint in progress on return from
native (i.e., native -> Java transition) and blocks there if so.  If it 
didn't,
then Java code, which uses un-handlized oops, could have those oops
changed out from under it.  You can also think of JNI calls as native ->
Java -> native transitions, albeit trivial ones for get and putfield.  They
do a Java operation, so effectively they transition in and out of Java 
state.

Paul

Gary Benson wrote:
> Keith McGuigan wrote:
>   
>> Transistions from Java -> native code (VM or library) and back
>> also have checks for safepoints.
>>     
>
> Do things safepoint on Java-native?  I thought it was only
> native-Java.  I thought Java-native didn't safepoint so you
> could pass oops as arguments to runtime calls.
>
> Cheers,
> Gary
>
>