views:

185

answers:

3

The scala type Nothing represents (as I understand it) the bottom of the type hierarchy, also denoted by the symbol ⊥. That is, Nothing is a sub-type of any given type. The requirement for a Nothing type is explained well by James Iry for those of us without a theoretical background in type theory!

So my question is, if Nothing is a subtype of every type, why can I not call any type's methods on Nothing? Obviously, I cannot instantiate Nothing but why doesn't the following compile?

var n: Nothing = _

def main(args: Array[String]) {
  println(n.length) //compile error: value length is not a member of Nothing
}

Surely as Nothing is a subtype of String this should be OK? Note that the following compiles just fine!

var n: Nothing = _

def foo(s: String) : Int =  s.length

def main(args: Array[String]) {
  println(foo(n))
}

as does:

def main(args: Array[String]) {
  println(n.asInstanceOf[String].length) 
}
+2  A: 

I suppose Nothing could accept any method, and perform a standard operation on all of them (throwing an exception). That wouldn't be very useful, though.

By presenting a compile error, the compiler is warning the programmer that a type he most likely didn't want, Nothing, got inferred somehow at a certain point in the code.

Daniel
But I can call `toString` on nothing, I was just wondering if there's some ugliness which treats Nothing as a special case from the compiler's perspective. Or whether my type-theory was awry and the methods a type supports *and the type itself* are not necessarily correlated
oxbow_lakes
I suppose the idea was to make a compile-time verification where possible. It is not ugly, it is good, when instead of a runtime error you get a compile-time one.
Pavel Shved
A: 

You can call toString on Nothing variable because of it s defintion:
final trait Nothing extends Any
And toString is member of Any. I think scala compiler treat's Nothing in type bounds only and treats it like any other trait on all over cases. Letting invoke any method on variable with type of Nothing will be very strange I think.

Nikolay Ivanov
But there's still some special treatment going on. From the type *signature* of `Nothing`, I can't tell that it extends every other type.
oxbow_lakes
Ah so it's a trait, not a class. So you can only call the methods that are explicitly defined by the trait. That would explain a lot.
jqno
@oxbow_lakes: Well special treatment described in specification so compiler treats it specially in terms of type bounds , e.g. for each type T following is true: Nothing < T < Any, but beside of that why compiler should treat Nothing trait any more special? No references of that in spec.
Nikolay Ivanov
+6  A: 

While Nothing is a subtype of everything, it does not inherit any method except for those in Any. This is because Nothing is more geared toward the functional end of the language. It's necessary for things like Option and List, but only as a type, not as a class.

The distinction here is a bit weird for those coming from an object-oriented background, but the fact is that subtyping as a concept is very distinct from OOP. Granted, object-oriented really implies subtyping in some form, but the reverse is not true. Benjamin Pierce's Types and Programming Languages does a good job of presenting the language F_< (pronounced "F sub"), which serves as a minimal example of a language with subtyping (but not OO).

Now, with all that said, I do agree that the fact that Nothing is immune from the normal inheritance rules does seem a bit inconsistent. However, from a theoretical standpoint, it makes perfect sense.

Daniel Spiewak