Fix #3938: adapt prefix to ease prefix inference

[liufengyun]

Fix #3938: adapt prefix to ease prefix inference

[liufengyun]

@odersky : we are doing prefix inference in exhaustivity check, this PR does some magic in order for inference to work smoothly. Could you have a look and see if it makes sense?

[odersky]

I could not correlate the claim Foo(Test.this.foo) <:< Foo(Foo.this) in i3939.scala with the code here. In i3938, foo is not a module, so how does the removeThisType affect it?

Also, note that if we do have a

 object Foo

then we have already Foo.type <:< Foo.this and Foo.this <:< Foo.type by the rules for ThisType in TypeComparer's firstTry and secondTry methods. So I am not sure why the code would be necessesary.

[liufengyun]

Thanks for the clarification. The correlation is that in the else branch, it throws away of the ThisType. For the module case, I'll try if I can throw it away.

[liufengyun]

I just checked, the patch for module cannot be removed, as the subtyping for Foo.type <:< Foo.this only works if it's static owner:

          case tp1: NamedType if cls2.is(Module) && cls2.eq(tp1.widen.typeSymbol) =>
            cls2.isStaticOwner ||
            isSubType(tp1.prefix, cls2.owner.thisType) ||
            secondTry(tp1, tp2)

The test case tests/patmat/3455.scala shows that we need the patch:

trait AxisCompanion {
   sealed trait Format
   object Format {
      case object Decimal extends Format
      case object Integer extends Format
   }
}
object Axis extends AxisCompanion
class Axis {
   import Axis._
   def test( f: Format ) = f match {
      case Format.Integer => "Int"
   }
}

[odersky]

Thanks for explaining @liufengyun.