Question

def foo(x: Int, f: Unit => Int) = println(f())

foo(2, { Unit => 3 + 4 })

// case 1
def loop: Int = 7
foo(2, loop) // does not compile

changing loop to 
// case 2
def loop(): Int = 7
foo(2, loop) // does not compile

changing loop to
// case 3
def loop(x: Unit): Int = 7 // changing according to Don's Comments
foo(2, loop) // compiles and works fine

Shouldn’t case 1 and case 2 also work? Why are they not working?

Defining foo as

def foo(x: Int, y: () => Int)

then case 2 works but not case 1.

Arent they all supposed to work, defining the functions either way?

Also I think () => Int in foo is a bad style, y:=> Int does not work. Comments?

Answer 1

Scala distinguishes between the following things:

• Functions/methods with no parameter lists (“by-name parameter” if a function)
• Functions with one empty parameter list
• Functions with one parameter of type Unit

None of these are equivalent, although as a convenience Scala allows you to elide empty parameter lists. (Incidentally, two empty parameter lists are also not the same.)

So, even though Unit is written (), this is not the same as the function argument parens () for a function or method. Instead, think of () as a Tuple0.

So, if you say f: Unit => Int, what you mean is “f takes one parameter, but it’s a really boring parameter because it is Unit, which must always be the same boring Tuple0 value ()“. What you’re writing is really short for f: (Unit) => Int.

If you say f: () => Int, then you mean that “f takes no parameters and produces an Int“.

If you say f: => Int, then you mean that “delay the execution of whatever statement produces an Int value until we use it in this code (and re-evaluate it each time)”. Functionally, this ends up being basically the same as f: () => Int (and internally is converted into the same Function0 class), but it has a different usage, presumably to allow for a more compact form of closures (you always omit the => in the calling code).