John Rose a écrit :
> I hope b59 smells better. I just pushed the changes I've chosen for
> b59 to the patch repo.
>
> I am working on other changes too, but I am in point-fix mode, as JDK7
> M3 settles down for JavaOne.
>
> My goal at the moment is to find workarounds for broken stuff, so as
> not to be making widespread changes.
>
> Here are some workarounds that will help:
>
> --- Use MethodHandles.lookup().findFoo(...) if you run into trouble
> with cached or public lookup-factories.
>
> Try to access only methods on the current class (the one calling
> lookup()), or else public methods in public classes.
>
> This avoids bugs with the access checking logic, by exercising the
> simplest forms of access checking.
>
Or use reflection (getMethods()), set accessibility
(setAccessible(true)) and
use MethodHandle.unreflect(java.lang.reflect.Method).
> --- In your method handle invocation sites and in invokedynamic sites,
> prefer method handles of built-in types.
>
> A symptom of when you might want to do this is "NoClassDefFound" at an
> invoke site, for an application class that obviously exists.
>
> Generally speaking, the inserted leading argument type to
> Lookup.findVirtual only gets in the way. Use convertArguments to
> change the leading argument to Object, or bind it to an instance
> immediately.
>
I've found that if you convert the leading argument of the method handler,
you have to cast the leading argument of MethodHandle.invoke() too.
// in my example
methodHandle = MethodHandles.convertArguments(methodHandle,
MethodType.make(void.class,
/*AbstractVisitor.class*/ Object.class, // workaround
to NoClassDefFound bug
Object.class));
// and then
methodHandle.<void>invoke((Object)this, target);
cast is needed here ------^
> --- When currying method handles with insertArgument, insert only a
> leading argument (#0) to a direct method handle.
>
> If you have something more complicated you want to do with currying,
> consider using a JavaMethodHandle instead.
>
> (Yes, all this should be taken care of for you, and it will be!)
>
> -- John
>
And it works like a charm :)
Rémi
_______________________________________________
mlvm-dev mailing list
mlvm-dev@openjdk.java.net
http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
