Parameterizing on incomplete types

[Per Bothner]

On 12/02/2011 10:33 PM, Jamison Hope wrote:
> On Dec 2, 2011, at 2:48 AM, Per Bothner wrote:
>
>> I checked in a patch that effectively special-cases Object.
>> (Specifically, I added an "optimization" is ClassType#isSubclass.)
>> So we no longer get that error - now we just have to deal with
>>
>> <string>:1:1: missing implementation for
>> java.lang.Comparable.compareTo(T)int
>
> That patch works for parameterizations like <T> (which is really <T
> extends Object>),
> but not the general case where the type could have some other restriction
> (e.g. <T extends java.util.List>).

Right.  But handling the general case looks non-trivial.

>> How ambitious are you about getting into Deep Kawa Stuff? :-)
>
> Sounds awesome, let's do it.. um, how?
>
> There are 2 parts, I think - (1) moving the type comparison from
> BracketApply to
> a later compilation stage, and (2) determining and injecting appropriate
> bridge
> methods. I have a vague notion of how someone might do #1. Not so much
> for #2..

(1) isn't fundamentally difficult, but it's tricky figuring out where
and how to do it.  It should probably be done as part of the InlineCalls
tree-walker.  And (1) isn't very useful until we solve (2), which
is why I suggest deferring it.

(2) mostly just a matter of code: generating the needed
bridge methods.  How the <init> methods are generated is probably
a helpful illustration.  The other (more interesting?) part is
figuring out *when* we need a bridge method, and the necessary
signature.
-- 
	--Per Bothner
per@bothner.com   http://per.bothner.com/