final val in object is never called

[iphydf]

object Foo {
  final val spartiates = 300
  val leonidasses = 1
}

object Bar {
  def spartiates: Int = Foo.spartiates
  def leonidasses: Int = Foo.leonidasses
}

Produces in Bar.class:

  public int spartiates();
    Code:
       0: sipush        300
       3: ireturn       

  public int leonidasses();
    Code:
       0: getstatic     #21                 // Field im/tox/tox4j/crypto/Foo$.MODULE$:Lim/tox/tox4j/crypto/Foo$;
       3: invokevirtual #23                 // Method im/tox/tox4j/crypto/Foo$.leonidasses:()I
       6: ireturn       

and in Foo.class:

  public final int spartiates();
    Code:
       0: sipush        300
       3: ireturn       

  public int leonidasses();
    Code:
       0: aload_0       
       1: getfield      #19                 // Field leonidasses:I
       4: ireturn       

The (private final) field leonidasses is initialised to 1 in the constructor.

The problem here is that for the "final val" case, it is a compile time constant and it's immediately constant-folded into the call site, causing 0% coverage on this code, even when explicitly called from Java (which does not do constant folding for these calls).

[sksamuel]

Yeah the problem was that the final val's get inlined, so they're not around to be instrumented. I could solve this perhaps if I can plug into the compiler before that happens.

[mirkof]

One of the solution we found is that if you put a type at that final, scoverage picks it up ok. :)

[dsilvasc]

Putting a type at that final will make it no longer a constant, so you won't be able to use it in annotations.

For constants, it seems like the current workaround is to turn off coverage for those statements:

object Foo {
  // disable coverage to work around https://github.com/scoverage/scalac-scoverage-plugin/issues/125
  // $COVERAGE-OFF$
  final val spartiates = 300
  // $COVERAGE-ON$

  val leonidasses = 1
}

[bufistov]

Hi, I have similar issue with 1.3.1
The coverage on/off directives eventually make code ugly. Is there any way to exclude final val fields within plugin? I guess it is always safe, right?

This issue seems also be related.

[som-snytt]

The constant value definition has coverage if every use site has coverage. It has an attachment for what was inlined. It might be interesting to know that it is used twice but tested once. (Perhaps that is also interesting for regular vals.)