The semantics of constant folding are a source of endless confusion, ignorance, and eventually consternation.

• that one signals it should take place with "final", a word already triple-overloaded in meaning by java
• that there can be no type annotation, a requirement which seems utterly arbitrary to someone not versed in the implementation

And even if you know all the rules, you have a good chance of being out of luck:

• since you can't annotate the type, there is no way to see a Byte, Short, or Char inlined (as those types are never inferred for literals)
• you are required to know implementation details to know what will actually be inlined, because superficially identical sources fall out differently, as here:

object Test {
  final val s1 = "abc"
  final val s2 = s"${s1}def"
  final val s3 = s1 + "def"
  def s4 = s2
  def s5 = s3
}

Is it reasonable to expect people to intuit that s5 has only a reference to string constant "abcdef", but s4 calls s2? I don't think so.

There is an obvious solution, which will also be the right kind of obvious to the scala programmer. If you want a constant inlined, mark it @inline.

We can do this immediately, keeping the final semantics as well for as long as desired, but ideally deprecating it in favor of @inline now.