Non-cosmetic failures using @TupleConstructor on nested class

[chrylis]

Groovy-Eclipse sometimes drops spurious errors in the editor saying that a constructor added by @TupleConstructor can't be found, but these are generally cosmetic and don't result in actual compilation errors (i.e., no markers, code runs).

However, I've discovered that as of 3.0.7 having a static nested class (a record for a command type) annotated with @TupleConstructor causes actual errors using the Groovy-Eclipse compiler; the previous false alarm does generate a marker and a failing class file. A full clean and build, which usually handles mismatches because everything's available, doesn't help. mvn compile with groovyc works fine.

[eric-milles]

I would need an example project to take this further. In general, the transforms that add members to a class that are not already represented by an interface of Object (TupleConstructor vs ToString) cause problems. Example, class A has reference to class B. When compiling class A, the reference to B is resolved by processing class B as far as conversion. B is not fully processed because it may contain a direct or indirect reference to class A. This is discussed further in #704 and others.

[chrylis]

Replicable. In order to reproduce:

• nested class; inner or static does not matter
• the client class must be @CompileStatic (the error is actually produced by the STC)

class Outer {
    @TupleConstructor(defaults = false)
    static class Nested {
        String value
    }
}

@CompileStatic
class Client {
    void foo() {
        new Nested('123')
    }
}

(Note that in the case of a top-level class the editor still shows the error, but the class does compile correctly anyway.)

[eric-milles]

ready to test