Progress towards full SIP-23 support: correct inference for scala.Singleton, support for symbol literals

[smarter]

No description provided.

[smarter]

@milessabin The narrowing trick from scalac does not work in Dotty:

scala> def narrow[T <: Singleton](x: T): T {} = x 
def narrow[T <: Singleton](x: T): T
scala> narrow(1) 
val res0: Int = 1

However, the following works in both scalac and dotty:

scala> type Identity[T] = T 
// defined alias type Identity = [T] => T
scala> def narrow[T <: Singleton](x: T): Identity[T] = x 
def narrow[T <: Singleton](x: T): ([T] => T)[T]
scala> narrow(1) 
val res1: ([T] => T)[1] = 1
scala> val b = res51
val b: ([T] => T)[1] = 1
scala> val c: 1 = b 
val c: 1 = 1

(The fact that the dotty repl prints ([T] => T)[T] instead of Identity[T] seems to be a pretty-printer issue, it shouldn't matter).

It might be useful to add a scala.Identity[T] as well as a definition of narrow to the Predef at some point.

[odersky]

Otherwise LGTM

[odersky]

Why add Scala? I'd prefer we just called it SYMBOLconst.

[odersky]

I don't think you need the duplication. SYMBOLconsts are simple types, so they should just occur in readSimpleType. I believe the duplication of CLASSconst and ENUMconst is also wrong and should be removed.

Comments addressed.