dependent types

On Thu, Oct 27, 2011 at 3:33 PM, Paul Phillips wrote:
> Is it expected that  f1 works and f2 doesn't? Is there a way to
> structure f2 so it does work?
>
> trait Foo {
>  type Arg
>  type Prod
>  def makeProd(a: Arg): Prod
> }
>
> object Test {
>  def f1(x: Foo)(y: x.Arg) = x.makeProd(y)
>
>  case class f2(x: Foo) {
>    def apply(y: x.Arg) = x.makeProd(y)
>  }
>
>  val myFoo = new Foo {
>    type Arg = Int
>    type Prod = (Int, Int)
>    def makeProd(i: Int) = (i, i)
>  }
>
>  def main(args: Array[String]): Unit = {
>    println(f1(myFoo)(5))   // works
>    println(f2(myFoo)(10))  // nope
>    // test/files/run/t102.scala:22: error: type mismatch;
>    //  found   : Int(10)
>    //  required: _1.x.Arg where val _1: Test.f2
>    //     println(f2(myFoo)(10))
>    //                       ^
>    // one error found
>  }
> }

How about,

case class f2[T <: Foo](x: T) {
def apply(y: x.Arg) = x.makeProd(y)
}

Cheers,

Miles

that's probably tricky to fix, although the violation uniformity is quite the incentive to do just that
if we zoom in on the culprit, f2(myFoo)(10):
(f2.apply(myFoo))(10) // redundant parentheses are not that redundant
where the apply method has been synthesized as:
def apply(x: Foo): Test.f2 = new Test.this.f2(x)
breaking it down further
(f2.apply(myFoo)): f2
the type of this expression conveniently forgets about x == myFoo,so by the time we get to the apply that supplies the 10, we don't know we're looking for a myFoo.Arg, not a f2#Arg
Miles solution fixes exactly that problem. We could probably generate some kind of refinement type for the synthesized def, def apply(_x: Foo): Test.f2{val x = _x} = new Test.this.f2(x)so that the value can be propagated, but that seems pretty, errr, unlikely to happen (right?)
obviously, f1 is a totally different beast, where we get a shot at propagating actual arguments to the types in later argument lists
(it's the following line in doTypedApply:val restpe = mt.resultType(args1 map (arg => gen.stableTypeFor(arg) getOrElse arg.tpe)))
On Thu, Oct 27, 2011 at 4:33 PM, Paul Phillips <paulp@improving.org> wrote:

Is it expected that  f1 works and f2 doesn't? Is there a way to
structure f2 so it does work?

trait Foo {
 type Arg
 type Prod
 def makeProd(a: Arg): Prod
}

object Test {
 def f1(x: Foo)(y: x.Arg) = x.makeProd(y)

 case class f2(x: Foo) {
   def apply(y: x.Arg) = x.makeProd(y)
 }

 val myFoo = new Foo {
   type Arg = Int
   type Prod = (Int, Int)
   def makeProd(i: Int) = (i, i)
 }

 def main(args: Array[String]): Unit = {
   println(f1(myFoo)(5))   // works
   println(f2(myFoo)(10))  // nope
   // test/files/run/t102.scala:22: error: type mismatch;
   //  found   : Int(10)
   //  required: _1.x.Arg where val _1: Test.f2
   //     println(f2(myFoo)(10))
   //                       ^
   // one error found
 }
}