meth review

[John Rose]

On Sep 13, 2008, at 3:24 PM, Rémi Forax wrote:

> I've done a more deep review of the code.
> Two codes are weird :
> MethodType.java, line 155
>         if (!trusted)
>             // defensively copy the array passed in by the user
>             ptypes = ptypes.clone();
>
> ptypes is not used after this line.
> I think this test should be moved before the creation of mt1
> (before line 146).

Excellent; thanks.  That line should be:
   mt1.ptypes = ptypes.clone();

But it should not be moved.  The initial query can be done with an  
untrusted array; that's the common fast case.

The query is re-done from scratch afterward with the trusted copy.

Also, I noticed that the trust bit was set wrong for the version of  
make taking Class<?>... ptypes.

> MH.java, line 234
> static void setVMDataArgSlot(MH self, int argSlot) {
>         long index = self.vmdata >> 32;
>         self.vmdata = (index << 32) | (argSlot << 32 >>> 32);
> }
>
> knowing that argSlot is an int, I don't understand 'argSlot << 32'.

Heh; argSlot wants to be a long.

Thanks for the close reading!

-- John

P.S.  I'm working in the JVM code right now, hooking up the  
unimplemented compiled-code paths.