Re: [jdt-core-dev] Help with ParameterizedQualifiedTypeReference

[Stephan Herrmann <stephan.herrmann@xxxxxxxxx>]

Hi Carmi,


On 08/02/2016 02:51 AM, Carmi Grushko wrote:
> Hi,
> I'm trying to fix Bug 462192, where the following happens.
> Suppose we have a Java file,
>
>     class Foo {
>      void foo() { bar(null); }  // <----- resolveBinding() on `bar` is null
>      static void bar(Clazz.Builder<?> b) { }
>     }
>
> Then the `bar(null)` on the 2nd line will fail to resolve.
>
> Deleting the dot between Clazz and Builder causes everything to work again.
>
>     class Foo {
>      void foo() { bar(null); }  // <----- resolveBinding() on `bar` is non-null
>      static void bar(ClazzBuilder<?> b) { } // <----- no dot
>     }
>
> I'd expect qualified and non-qualified types to behave the same.
> Is there a reason why we shouldn't resolve `bar` in the first case?

To double check I understand your case: I don't see any of the
types Clazz, Clazz.Builder, ClazzBuilder, so in both cases the
parameter of bar is unresolved, right?

To answer whether a MessageSend to bar() can be resolved, we
should probably first understand what MethodBindings are generated
for bar() in either case.

> In ParameterizedQualifiedTypeReference.internalResolveLeafType, I tried returning
> `this.resolvedType.closestMatch()` instead of null

If you modify the compiler in such ways expect lots of regressions
throughout the compiler test suite :)
If resolving fails, it must signal this to callers of internalResolveLeafType().
OTOH, I don't understand how a closestMatch() of an unresolved type could help any.

Let's continue this in the bug, OK?
Stephan