An NPE in TypeTree.setOriginal

[scabug]

import scala.language.experimental.macros
import scala.reflect.macros.Macro

object TypeOf {
  type TypeOf[T](s: T) = macro TypeOfMacro.TypeOf[T]
}
trait TypeOfMacro extends Macro {
  import c.universe._

  def TypeOf[T: WeakTypeTag](s: Expr[T]): Tree = {
    c.echo(NoPosition, "T " + weakTypeOf[T])
    tq"${weakTypeOf[T]}"
  }
}

import TypeOf._

trait A {
  val i: Int
  val j: TypeOf(i)
}

[scabug]

Imported From: https://issues.scala-lang.org/browse/SI-7284?orig=1
Reporter: @xeno-by

[scabug]

@xeno-by said (edited on Mar 22, 2013 9:20:16 AM UTC):
The TypeTree produced by the macro expansion has a null original. Therefore when the result of its typedType, a TypeTree, calls setOriginal passing that tree as an argument, the following code NPEs:

def setOriginal(tree: Tree): this.type = {
  def followOriginal(t: Tree): Tree = t match {
    case tt: TypeTree => followOriginal(tt.original)
    case t => t
  }

  orig = followOriginal(tree); setPos(tree.pos)
  this
}

[scabug]

@paulp said:
Can you offer a reproduction which doesn't use code which isn't in master? I can easily fix the code but I am not interested in reverse engineering a test.

[scabug]

@xeno-by said:
Will try to find one.

[scabug]

@xeno-by said:
Don't think this particular issue is reproducible if we don't have macro types.

The problematic scenario is caused by the weird combination of:

1. Input of typedTypeTree having a non-null original (application of a type macro)
2. The result of typedType on the input having a null original (the result of the expansion).

Probably the correct fix here would be setting the type macro application tree as original for the result of macro expansion, which would prevent #2 from happening, making the bug impossible.

That said, I think we still should fix the NPE in setOriginal. It's just we don't have an integration test triggering that NPE.

[scabug]

@paulp said:
I think we should fix it also, and I already did the patch, except I'll be hassled for a test.

[scabug]

@xeno-by said:
I'm closing this bug since type macros have been discontinued.