Java 1.8.0_131 breaks some super calls inside 2.12 lambdas

[SethTisue]

as reported by @milessabin, and reproduced by myself and @dwijnand, test/files/run/t4300.scala (#4300) and test/files/run/t8803.scala (#8803) are failing now on Java 8 Update 131 with e.g.

java.lang.IllegalAccessError: Receiver class Test$$anon$1 must be the current class or a subtype of interface A

I haven't isolated the exact circumstances, but both tests involve super within lambdas

[SethTisue]

Update 131 release notes: http://www.oracle.com/technetwork/java/javase/8u131-relnotes-3565278.html, and the full fix list is at http://www.oracle.com/technetwork/java/javase/2col/8u131-bugfixes-3565760.html. surprisingly, I see nothing relevant in either place (nor did Miles)

[SethTisue]

[info] >>> The dbuild result is------------: SUCCESS (project rebuilt ok)

community build is all green, so it seems this won't cause widespread breakage. whew

[lrytz]

Minimized:

trait A {
  def f() = println("A")
}

class B extends A {
  class C {
    def c() = B.super[A].f()
  }
}

object Test {
  def main(args : Array[String]) : Unit = {
    val b = new B()
    new b.C().c()
  }
}

[lrytz]

It seems that check was added here: http://hg.openjdk.java.net/jdk8u/jdk8u/hotspot/rev/0b85ccd62409#l11.66

[lrytz]

We have

public abstract interface A {
  public static synthetic f$(LA;)V
    // parameter final synthetic  $this
   L0
    LINENUMBER 2 L0
    ALOAD 0
    INVOKESPECIAL A.f ()V
}

public class B$C {
  public c()V
   L0
    LINENUMBER 7 L0
    ALOAD 0
    INVOKESTATIC A.f$ (LA;)V
}

I wonder why this worked before: B$C doesn't conform to A, so passing this in B$C.c to A.f$ should have produced a verifier error?

[lrytz]

Indeed, on the older JVMs, a slightly modified version leads to an IncompatibleClassChangeError:

trait A {
  def bar = println("A")
  def f() = bar
}
// otherwise the same as above

java.lang.IncompatibleClassChangeError: Class B$C does not implement the requested interface A
	at A.f(Test.scala:3)
	at A.f$(Test.scala:3)
	at B$C.c(Test.scala:8)

[lrytz]

So in a way that's good news: the code we generate was wrong and would only not crash in the most trivial cases (on old JVMs). The difference on 1.8.0_131 is only for those trivial cases (now they crash as well).

[lrytz]

Also to note: it's a regression in 2.12, 2.11 correctly uses the outer when invoking A.f:

public class B$C {
  public c()V
   L0
    LINENUMBER 7 L0
    ALOAD 0
    INVOKEVIRTUAL B$C.B$C$$$outer ()LB;
    INVOKESTATIC A$class.f (LA;)V

[milessabin]

I'm still seeing this on 2.13.x ... is the forward port of the fix still pending?

[lrytz]

Yes, it's here: scala/scala#5997