RFR: 8337980: Javac allows invocation of an inherited instance method from a static method [v2]

[Maurizio Cimadamore]

On Mon, 7 Oct 2024 19:58:44 GMT, Archie Cobbs <acobbs at openjdk.org> wrote:

>> I guess what I'm asking is whether there's some way to alter the declaration of the problematic methods so that the early-constructor context error is not reported - this seems possible if mergeAbstracts merges info using the interface method as a base symbol?
>
> I think I understand what you're saying (but apologies in advance if not :)
> 
> This particular bug can only occur in situations where an error should be reported, but you're wondering whether a similar situation could happen in valid code, which would mean the `AmbiguityError` is allowed to float downstream and possibly cause problems later... is that correct?
> 
> I believe that this can't happen because it always eventually gets resolved via `BasicLookupHelper.lookup()` which is used by `resolveMethod()`, `resolveQualifiedMethod()`, etc.
> 
> For example, this class:
> 
> public class D {
> 
>     public interface A {
>         String op();
>     }
> 
>     public abstract static class B {
>         public abstract Object op();
>     }
> 
>     public abstract static class C extends B implements A {
> 
>         public void foo(Object x) {
>         }
> 
>         public void bar() {
>             foo(op());
>         }
>     }
> } 
> 
> compiles successfully, and `D.C.bar()` will invoke the `A.op` variant:
> 
> public abstract class D$C extends D$B implements D$A {
> ...
>   public void bar();
>     Code:
>        0: aload_0
>        1: aload_0
>        2: invokevirtual #7                  // Method op:()Ljava/lang/String;
>        5: invokevirtual #13                 // Method foo:(Ljava/lang/Object;)V
>        8: return
> }

What I'm worried about is a situation where:
* multiple abstract methods are applicable
* one of the abstract method belongs to a class in an early construction context
* their signature needs to be merged
* when the signature is merged we use an interface as the owner for the symbol
* as a result, the early construction context error is missed (and I think the generated code will not pass the verifier)

-------------

PR Review Comment: https://git.openjdk.org/jdk/pull/20533#discussion_r1790860559