value class restrictions should honor self type

[scabug]

trait Foo {
  self: bar.type =>
  
  final class V(val x: Int) extends AnyVal
  // ./a.scala:4: error: value class may not be a member of another class
  //   final class V(val x: Int) extends AnyVal
  //               ^
  // one error found
}

object bar extends Foo { }

This should compile, else we are robbed of all the mechanisms by which one normally can partition a large object into its constituent parts.

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-5962?orig=1
Reporter: @paulp

[scabug]

@adriaanm said (edited on Jul 3, 2012 4:07:00 PM UTC):
This would be hard to spec: "a class with a self type that's a singleton type referring to an accessible symbol does not need an outer pointer, and may thus serve as an outer class of a value class (which cannot accommodate an outer pointer)".

Instead, it seems more likely that we'll at one point support multi-field value classes,
only to then realize that the outer pointer is in fact not needed here and fall back to single-field value class.

(I hope I correctly paraphrased Martin's thoughts from today's meeting.)

[scabug]

@paulp said:
Right now the spec says "C is either a toplevel class or is a member of a statically accessible object." In what sense is V not a member of a statically accessible object? If there is a V, is V not a member of bar, the statically accessible object?

[scabug]

@paulp said:
I'm not sure by what route this ended up on the blockers for 2.10.0 list, but as the opener, I say it can certainly wait. (Nor does it look especially easy to fix, at least not in the face of separate compilation.)

[scabug]

@adriaanm said:
it ended up here so we could discuss it -- i agree it's not a blocker