I create https://bugs.openjdk.java.net/browse/JDK-8246634 to track this.
thanks
Mandy
On 6/4/20 4:35 PM, John Rose wrote:
P.S.C. (post-send clarification)
The workflow would be:
static final MethodHandle MH_ensureInit
= publicLookup().findVirtual(L…, “ensureInitialized”…);
By that I mean the workflow of the dynamic language
implementor. And after hitting “send” I realized that optimizing
that one case (of a findVirtual -> bindTo -> bindTo) is harder
than I thought. The entire workflow below could instead be
a call to a wrapping function that takes an arbitrary “mh”
(as below) and wraps an init-barrier onto the front *if needed*.
It would have to take a Lookup and a Class, plus the mh.
It would handle all the internal stuff. If the user wanted
a bare init MH, just pass in an empty MH as a “seed”.
And maybe take a short-circuit action if the “seed” comes
back unchanged; this is an alternative to isFullyInitialized,
but it seems sneaky.
class Lookup { …
/**
* To the given method handle, prepend an action, if necessary, to trigger
* initialization of the given class the first time the method handle is
called.
* If it would be illegal to call ensureInitialized on the given class from
this lookup,
* immediately report an error. Otherwise, if the class is already fully
initialized
* return the method handle unchanged. Otherwise, return a new method handle
* (of the same type) which incorporates the necessary initialization action,
* as if by a call to ensureInitialized.
*/
MethodHandle ensureInitialized(Class target, MethodHandle mh) {
checkAccess(target);
if (isFullyInitialized(target)) return mh;
// we need to tack a “before action” onto mh to trigger inits:
MethodHandle before = MH_ensureInit.bindTo(target);
return collectArguments(mh, 0, before);
}
// note: IMPL_LOOKUP requires previous checkAccess call
private static final MethodHandle MH_ensureInit
= publicLookup().findVirtual(L…, “ensureInitialized”…).bindTo(IMPL_LOOKUP);