Parameterizing on incomplete types

[Jamison Hope]

On Dec 1, 2011, at 4:28 PM, Per Bothner wrote:

> On 12/01/2011 01:21 PM, Jamison Hope wrote:
>>> It probably makes sense to generate the bridge methods at the same
>>> time as we check for missing implementations.
>>
>> Where are they generated now?
>
> My point is:  they're not.  That needs to be implemented.

At all? I thought you meant they weren't being generated here as a  
result
of you commenting out the relevant part of ClassExp or because they  
don't
yet work with this circular parameterization nonsense.

I just meant for the regular (non-parameterized) case. For instance,
when javac compiles:

> // A.java
> public class A
> {
>   public A get() { return null; }
> }
>
> // B.java
> public class B extends A
> {
>   public B get() { return null; }
> }

B.class ends up with an extra method:

> Compiled from "B.java"

> public class B extends A{
>     public B();
>     public B get();
>     public A get();
> }

The third method is marked SYNTHETIC|PUBLIC|BRIDGE (0x1041) and its
implementation is

aload_0
invokevirtual 2   // call public B get()
areturn

and it exists so that B actually overrides A's get() method with an
identical signature.


But it turns out that when Kawa compiles the Scheme equivalent,

> (define-simple-class A ()
>   ((get) ::A #!null))
>
> (define-simple-class B (A)
>   ((get) ::B #!null))

B.class doesn't get the extra method:

> Compiled from "bridge.scm"
> public class bridge.B extends bridge.A{
>     public bridge.B get();
>     public bridge.B();
> }


But maybe that's not what you're referring to as a bridge method. I  
can't
find any instance of the word "bridge" or 0x0040 as anything but  
ACC_VOLATILE
in the JVM spec.


>> Is that what the recursive call to
>> ClassExp#getImplMethods() is doing?
>
> No, that just searches for possible implementation methods.
> This is done for each abstract method.
>
> The key parts of the logic is:
>
>            methods = compiledType.getAbstractMethods();
>            nmethods = methods.length;
>        ...
> 	for (int i = 0;  i < nmethods;  i++)
> 	  {
> 	    Method meth = methods[i];
>            ...
>            Vector vec = new Vector();
> 	    getImplMethods(compiledType, mname, ptypes, vec);
>            if (vec.size() != 1)
> 	        error(...)
>
> This logic needs to be made aware of parameterized types.
> -- 
> 	--Per Bothner
> per@bothner.com   http://per.bothner.com/

--
Jamison Hope
The PTR Group
www.theptrgroup.com