Reading Types

In the Core Language section of this book, we ran a bunch of code in the REPL. Well, we are going to do it again, but now with an emphasis on the types that are getting spit out. So type elm repl in your terminal again. You should see this:

---- elm repl 0.18.0 -----------------------------------------------------------
 :help for help, :exit to exit, more at <>

Primitives and Lists

Let's enter some simple expressions and see what happens:

> "hello"
"hello" : String

> not True
False : Bool

> round 3.1415
3 : Int

In these three examples, the REPL tells us the resulting value along with what type of value it happens to be. The value "hello" is a String. The value 3 is an Int. Nothing too crazy here.

Let's see what happens with lists holding different types of values:

> [ "Alice", "Bob" ]
[ "Alice", "Bob" ] : List String

> [ 1.0, 8.6, 42.1 ]
[ 1.0, 8.6, 42.1 ] : List Float

> []
[] : List a

In the first case, we have a List filled with String values. In the second, the List is filled with Float values. In the third case the list is empty, so we do not actually know what kind of values are in the list. So the type List a is saying "I know I have a list, but it could be filled with anything". The lower-case a is called a type variable, meaning that there are no constraints in our program that pin this down to some specific type. In other words, the type can vary based on how it is used.


Let's see the type of some functions:

> import String
> String.length
<function> : String -> Int

The function String.length has type String -> Int. This means it must take in a String argument, and it will definitely return an integer result. So let's try giving it an argument:

> String.length "Supercalifragilisticexpialidocious"
34 : Int

The important thing to understand here is how the type of the result Int is built up from the initial expression. We have a String -> Int function and give it a String argument. This results in an Int.

What happens when you do not give a String though?

> String.length [1,2,3]
-- error!

> String.length True
-- error!

A String -> Int function must get a String argument!

Anonymous Functions

Elm has a feature called anonymous functions. Basically, you can create a function without naming it, like this:

> \n -> n / 2
<function> : Float -> Float

Between the backslash and the arrow, you list the arguments of the function, and on the right of the arrow, you say what to do with those arguments. In this example, it is saying: I take in some argument I will call n and then I am going to divide it by two.

We can use anonymous functions directly. Here is us using our anonymous function with 128 as the argument:

> (\n -> n / 2) 128
64 : Float

We start with a Float -> Float function and give it a Float argument. The result is another Float.

Notes: The backslash that starts an anonymous function is supposed to look like a lambda λ if you squint. This is a possibly ill-conceived wink to the intellectual history that led to languages like Elm.

Also, when we wrote the expression (\n -> n / 2) 128, it is important that we put parentheses around the anonymous function. After the arrow, Elm is just going to keep reading code as long as it can. The parentheses put bounds on this, indicating where the function body ends.

Named Functions

In the same way that we can name a value, we can name an anonymous function. So rebellious!

> oneHundredAndTwentyEight = 128.0
128 : Float

> half = \n -> n / 2
<function> : Float -> Float

> half oneHundredAndTwentyEight
64 : Float

In the end, it works just like when nothing was named. You have a Float -> Float function, you give it a Float, and you end up with another Float.

Here is the crazy secret though: this is how all functions are defined! You are just giving a name to an anonymous function. So when you see things like this:

> half n = n / 2
<function> : Float -> Float

You can think of it as a convenient shorthand for:

> half = \n -> n / 2
<function> : Float -> Float

This is true for all functions, no matter how many arguments they have. So now let's take that a step farther and think about what it means for functions with multiple arguments:

> divide x y = x / y
<function> : Float -> Float -> Float

> divide 3 2
1.5 : Float

That seems fine, but why are there two arrows in the type for divide?! To start out, it is fine to think that "all the arguments are separated by arrows, and whatever is last is the result of the function". So divide takes two arguments and returns a Float.

To really understand why there are two arrows in the type of divide, it helps to convert the definition to use anonymous functions.

> divide x y = x / y
<function> : Float -> Float -> Float

> divide x = \y -> x / y
<function> : Float -> Float -> Float

> divide = \x -> (\y -> x / y)
<function> : Float -> Float -> Float

All of these are totally equivalent. We just moved the arguments over, turning them into anonymous functions one at a time. So when we run an expression like divide 3 2 we are actually doing a bunch of evaluation steps:

  divide 3 2
  (divide 3) 2                 -- Step 1 - Add the implicit parentheses
  ((\x -> (\y -> x / y)) 3) 2  -- Step 2 - Expand `divide`
  (\y -> 3 / y) 2              -- Step 3 - Replace x with 3
  3 / 2                        -- Step 4 - Replace y with 2
  1.5                          -- Step 5 - Do the math

After you expand divide, you actually provide the arguments one at a time. Replacing x and y are actually two different steps.

Let's break that down a bit more to see how the types work. In evaluation step #3 we saw the following function:

> (\y -> 3 / y)
<function> : Float -> Float

It is a Float -> Float function, just like half. Now in step #2 we saw a fancier function:

> (\x -> (\y -> x / y))
<function> : Float -> Float -> Float

Well, we are starting with \x -> ... so we know the type is going to be something like Float -> .... We also know that (\y -> x / y) has type Float -> Float.

So if you actually wrote down all the parentheses in the type, it would instead say Float -> (Float -> Float). You provide arguments one at a time. So when you replace x, the result is actually another function.

It is the same with all functions in Elm:

> import String
> String.repeat
<function> : Int -> String -> String

This is really Int -> (String -> String) because you are providing the arguments one at a time.

Because all functions in Elm work this way, you do not need to give all the arguments at once. It is possible to say things like this:

> divide 128
<function> : Float -> Float

> String.repeat 3
<function> : String -> String

This is called partial application. It lets us use the |> operator to chain functions together in a nice way, and it is why function types have so many arrows!

Type Annotations

So far we have just let Elm figure out the types, but it also lets you write a type annotation on the line above a definition if you want. So when you are writing code, you can say things like this:

half : Float -> Float
half n =
  n / 2

divide : Float -> Float -> Float
divide x y =
  x / y

askVegeta : Int -> String
askVegeta powerLevel =
  if powerLevel > 9000 then
    "It's over 9000!!!"

    "It is " ++ toString powerLevel ++ "."

People can make mistakes in type annotations, so what happens if they say the wrong thing? Well, the compiler does not make mistakes, so it still figures out the type on its own. It then checks that your annotation matches the real answer. In other words, the compiler will always verify that all the annotations you add are correct.

Note: Some folks feel that it is odd that the type annotation goes on the line above the actual definition. The reasoning is that it should be easy and noninvasive to add a type annotation later. This way you can turn a sloppy prototype into higher-quality code just by adding lines.

results matching ""

    No results matching ""