Imported From: https://issues.scala-lang.org/browse/SI-2609?orig=1
Reporter: @jorgeortiz85

@jorgeortiz85 said:
Sorry, meant to say "since the type T is correctly inferred to be String".

@harrah said:
T is inferred to be Nothing, not String. Nothing is a subtype of String, so the second statement is allowed. More directly,

scala> val s: String = null.asInstanceOf[Nothing]
s: String = null

Of course, Nothing is not supposed to have any instances, so having an actual instance assigned to a val of type Nothing is presumably the cause of the following:

scala> val s = null.asInstanceOf[Nothing]
java.lang.NullPointerException                   
        at RequestResult$$.<init>(<console>:9)    
        at RequestResult$$.<clinit>(<console>)    
        at RequestResult$$result(<console>)
        ...

@odersky said:
I think I agree with Mark's assessment. The problem is that the expected return type String only gives an upper bound. So it does not constrain the parameter type of the Manifest. So this is in fact as it's supposed to be (given the current type inference algorithm).

@adriaanm said (edited on Jul 8, 2011 9:31:43 AM UTC):
Here's what -Xprint:typer tells us for variations on this theme. Pretty consistently Nothing.

object Test extends App {
  def inject0[T]: T = null.asInstanceOf[T]
  def m0: String = inject0 // inject0[Nothing] is inferred

  def inject1[T](implicit m: Manifest[T]): T = {println(m.toString); null.asInstanceOf[T]}
  def m1: String = inject1 // inject1[Nothing] is inferred

  class M[T]; implicit def M[T] = new M[T]
  def inject2[T](implicit m: M[T]): T = {println(m.toString); null.asInstanceOf[T]}
  def m2: String = inject2 // inject2[Nothing] is inferred

  m0
  m1
  m2
}