Well, it might have to do with the security. Bootstrap method implementations
will likely use MethodHandles.* methods -- unreflect(), findVirtual() etc. The
access checks are performed at the point of invocation of these discovery
methods, resolved against the caller of these methods. A "rogue" subclass could
inherit its bootstrap from a trusted class and link to and invoke methods it
would otherwise not have access to. (For purpose of access checks, the caller
would still be the superclass, since the bootstrap methods are static.)
That's what I think; John Rose might clarify :-)
Attila.
On 2010.08.24., at 22:15, Jim Laskey wrote:
> Likely some subtlety I'm missing here.
>
> I meant inherited in the general sense. Maybe hierarchically blest is a
> better term. That is, that during initialization, a subclass copies the
> bootstrap MH from it's parent class (no instances involved.) The subclass
> can still override the bootstrap with a static initializer. The point is to
> simplify the launch effort/size/performance in each of the subclasses.
>
> So instead of
>
> class LangBaseClass {
> public static CallSite bootstrap(Class caller, String name, MethodType
> type) {
> ...
> return callSite;
> }
>
> static {
> java.dyn.Linkage.registerBootstrapMethod(LangBaseClass.class,
> "bootstrap");
> }
> }
>
> class LangSubClass1 extends LangBaseClass {
> static {
> java.dyn.Linkage.registerBootstrapMethod(LangBaseClass.class,
> "bootstrap");
> }
>
> ...
>
> InvokeDynamic.foo();
>
> }
>
> class LangSubclass2 extends LangBaseClass {
> static {
> java.dyn.Linkage.registerBootstrapMethod(LangBaseClass.class,
> "bootstrap");
> }
>
> ...
>
> InvokeDynamic.bar();
>
> }
>
> ...
>
> We have;
>
>
> class LangBaseClass {
> public static CallSite bootstrap(Class caller, String name, MethodType
> type) {
> ...
> return callSite;
> }
>
> static {
> java.dyn.Linkage.registerBootstrapMethod(LangBaseClass.class,
> "bootstrap");
> }
> }
>
> class LangSubClass1 extends LangBaseClass {
>
> ...
>
> InvokeDynamic.foo();
>
> }
>
> class LangSubclass2 extends LangBaseClass {
>
> ...
>
> InvokeDynamic.bar();
>
> }
>
> ...
>
>> Date: Tue, 24 Aug 2010 16:01:00 +0200
>> From: R?mi Forax <fo...@univ-mlv.fr>
>> Subject: Re: Common bootstrap
>> To: mlvm-dev@openjdk.java.net
>> Message-ID: <4c73d09c.60...@univ-mlv.fr>
>> Content-Type: text/plain; charset=ISO-8859-1; format=flowed
>>
>> Le 24/08/2010 15:36, Jim Laskey a ?crit :
>>> Excuse me for being late in the game but I have a na?ve question.
>>>
>>> Why not have the bootstrap method be inherited? Isn't it likely that a
>>> j-lang would have a base class and that often a common bootstrap might be
>>> used? (I'm assuming it's an implementation/security issue, but asking
>>> anyway.)
>>>
>>> Cheers,
>>>
>>> -- Jim
>>>
>>
>> Before bootstrapping you have nothing :)
>> so you have no object to do a virtual call.
>>
>> Attila Szegedi has an interesting answers on how to share/reuse some
>> infrastructure on top of invokedynamic:
>> http://wiki.jvmlangsummit.com/images/7/73/Metaobject_Protocol_Meets_Invokedynamic.pdf
>>
>> R?mi
Attila.
--
home: http://www.szegedi.org
twitter: http://twitter.com/szegedi
weblog: http://constc.blogspot.com
_______________________________________________
mlvm-dev mailing list
mlvm-dev@openjdk.java.net
http://mail.openjdk.java.net/mailman/listinfo/mlvm-dev
