RFR: 8334248: Invalid error for early construction local class constructor method reference [v2]

[Archie Cobbs]

On Fri, 14 Jun 2024 21:26:40 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

> As far as I can tell, that additional check for method reference is there "only" to generate a better error message. And, in fact, the check in Resolve dominates the one in Attr: even if the check you tweaked in this PR is commented out completely, no regression test fails.

That may have been true before "flexible constructors", but it's not now - when that check is commented out, this program compiles successfully for me (erroneously), leading to a verifier error:

class Q {
    class Inner { }
    public Q() {
        this(Inner::new);
    }
    public Q(Runnable r) {
    }
    public static void main(String[] args) {
        new Q();
    }
}

> Ideally we should only keep the check in Resolve (as that generates a better error), but augment it so that it detects the issue described in 8173456.

I agree that the remaining problem seems to be within `Resolve.resolveImplicitThis()`.

See commit e0f5d183651 to this PR where I tried to fix it... I'm curious if it that change looks correct to you?

In any case, when that commit is combined with all the other patches floating around (yours & mine), all of the `langtools_javac` regression tests (including the test for JDK-8173456) are passing, and the `Q.java` example above now correctly yields:

Q.java:4: error: cannot reference this before supertype constructor has been called
        this(Inner::new);
             ^
1 error

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

PR Comment: https://git.openjdk.org/jdk/pull/19705#issuecomment-2169920987