valueOf does not work in inlined code

[odersky]

[odersky]

object Test extends App {
  inline def v[T] = valueOf[T]

  println(v[String])
}

leads to

Unexpected tree in genLoad: TypeTree[TypeRef(ThisType(TypeRef(NoPrefix,module class lang)),class String)]/class dotty.tools.dotc.ast.Trees$TypeTree at: <56..56>
java.lang.RuntimeException: Unexpected tree in genLoad: TypeTree[TypeRef(ThisType(TypeRef(NoPrefix,module class lang)),class String)]/class dotty.tools.dotc.ast.Trees$TypeTree at: <56..56>
	at dotty.tools.backend.jvm.DottyBackendInterface.abort(DottyBackendInterface.scala:390)
	at dotty.tools.backend.jvm.BCodeBodyBuilder$PlainBodyBuilder.genLoad(BCodeBodyBuilder.scala:431)

The culprit is that synthesizeValueOf will succeed also if the argument is a non-singleton TypeRef but of course that can't work. This problem is fixed in #6218.

With the fix we get:

2 |  inline def v[T] = valueOf[T]  // error
  |                              ^
  |no implicit argument of type ValueOf[T] was found for parameter vt of method valueOf in object DottyPredef

That's still unsatisfactory. valueOf should work in inlined code. Its sibling constValue does, but we should have the functionality of both in one method.

The underlying problem is that valueOf demands an implicit unconditionally, so it can't work in inlined code, where the implicit is not yet available since the type is not yet specialized.

[odersky]

Here's a workaround:

      inline def valueOf2 = implicit match {
        case ev: ValueOf[T] => ev.value
      }

Question: Should valueOf be defined like this?

[milessabin]

@odersky That looks OK to me ... I'm happy to test and PR that change.

[milessabin]

No problem with that change, though it might be better to roll it in to #6218 so that a neg test can be added.

[odersky]

@milessabin OK, please go ahead!

[milessabin]

@odersky as a separate PR?