Anonymous Functions in Scala

Anonymous Functions in Scala

This post is part 2 of the Functional Programming in Scala series. You can view all the posts in the series here.


The problem with function types is that they are cumbersome and still heavily object oriented. Instead, we can use lambda functions - (x, y) => x + y.

The anatomy of a lambda looks like this:

val lambdaFunction = (name: String, age: Int) => name + " is " + age + " years old"
  • First you have the parameters - (name: String, age: Int) - each parameter has a name and a type
  • Then the implementation name + " is " + age + " years old" - which is a single expression
  • The return type is always inferred
  • And the parentheses for the parameters are mandatory if there is more than one parameter
  • This is heavy syntactic sugar for instantiating anonymous FunctionX types

We also use further syntactic sugar in lambda functions with underscores:

val add: (Int, Int) => Int = _ + _

Function Types are Object Oriented

We now know about function types from the previous blog post. There is a problem with the approach outlined in that blog post though. It is still very much tied to object orientation.

Take a look at the function below:

val doubler = new Function1[Int, Int] {
  override def apply(x: Int) = x * 2

This function simply doubles an Int. To implement it, we need to create a new Function1[Int, Int] and then override def apply and return x * 2. This is quite long winded. It follows the objected oriented approach of defining an anonymous function and instantiating it on the spot.

Lambda Functions

In Scala, we can do all of the above like this (this is the exact equivalent of the code above):

val doubler = (x: Int) => x * 2

It instantiates a new Function1, with the override def apply which takes an Int x, and returns x * 2.

This is syntactic sugar for the function type code. It is called an anonymous function, or a lambda.

The lambda is a value which is an instance of Function1. The type of our val doubler we can see in the IDE by hovering over it is Int => Int. So we could actually explicitly declare it as such:

val doubler: Int => Int = (x: Int) => x * 2

Given that we have declared doubler as the type Int => Int we can actually remove the (x: Int), and let it be matched by the compiler:

val doubler: Int => Int = x => x * 2

If the compiler already knows that doubler should be an Int => Int , it will assume that the expression will match the types. So x will be an Int, and the return value will be x * 2 , and all the types will match.

Lambda Functions with Multiple Parameters

What happens if you have multiple parameters in a lambda? Say we had a val called adder that adds 2 numbers, a and b. We give it a type of taking two Ints and returning an Int:

val adder: (Int, Int) => Int = (a: Int, b: Int) => a + b

Notice that we must put the (Int, Int) in between parentheses. When we had a single parameter (as before) we didn’t need to do that, but for multiple parameters we do.

We could also omit the type of the adder function, and let the compiler infer it for us:

val adder = (a: Int, b: Int) => a + b

Lambda Functions with no Parameters

What about if we have no parameters? The syntax for a no param lambda is () => and the value that you want to return (e.g 3):

val justDoSomething = () => 3

The type of this lambda is () => Int - we could pass that to the compiler if we so choose:

val justDoSomething: () => Int = () => 3

Now look at the two println statements below:


If we run this code, the first println returns something like scalaBasics.functionalProgramming.AnonymousFunctions$$$Lambda$11/991505714@2acf57e3.

The second statement returns 3, which is probably what we wanted!

You might be used to using justDoSomething when calling methods in Scala. But for calling lambdas, you must use the parentheses, i.e. justDoSomething().

Using Curly Braces with Lambdas

Another way of writing lambda functions is to use curly braces. The common style looks like this:

val stringToInt = { (str: String) =>

At the level of the open curly brace we put (str: String) => , and right below we put the implementation of the lambda.

Syntactic Sugar for Lambdas with Underscore (_)

Lets look at more syntactic sugar. Say we had a function of type Int => Int, which just increments the parameter. Normally we would write something like (x: Int) => x + 1, for example:

val niceIncrementer: Int => Int = (x: Int) => x + 1

An even shorter way to write the above is:

val niceIncrementer: Int => Int = _ + 1

The above is equivalent to x => x + 1.

Or if you had a two parameter function like this:

val niceAdder: (Int, Int) => Int = (a, b) => a + b

Instead of the above, this could be rewritten as:

val niceAdder: (Int, Int) => Int = _ + _

Where each of the underscores are a different parameter. This is equivalent to (a,b) => a + b.

This notation is extremely useful in practice when you want to chain higher order function calls.

But do note that the underscore notation is extremely contextual. If you miss the type, the compiler won’t know what each underscore means. For example if we wrote it like this:

val niceAdder = _ + _

This won’t compile, the compiler doesn’t know which underscore means what. It doesn’t even know how many parameters the function should have.

Replace FunctionX calls in MyList Application with Lambdas

We are going to replace all the FunctionX calls in MyList with lambdas. For reference, we first started writing the MyList application in the Covariant Generic List in Scala post.


The first FunctionX call that we are going to replace is the map call:

println( Function1[Int, Int] {
  override def apply(elem: Int): Int = elem * 2

This can be simply changed to:

println( => elem * 2).toString)

Or we can make an even shorter notation:

println( * 2).toString)


Next for the filter method, we are going to replace this:

println(listOfIntegers.filter(new Function1[Int, Boolean] {
  override def apply(elem: Int): Boolean = elem % 2 == 0

This we can change to:

println(listOfIntegers.filter(elem => elem % 2 == 0).toString)

Or for the even shorter notation:

println(listOfIntegers.filter(_ % 2 == 0).toString)


Finally for the flatMap, we can change this:

println(listOfIntegers.flatMap(new Function1[Int, MyList[Int]] {
  override def apply(elem: Int): MyList[Int] = new Cons(elem, new Cons(elem + 1, Empty))

into this:

println(listOfIntegers.flatMap(elem => new Cons(elem, new Cons(elem + 1, Empty))).toString)

Note that the underscore notation doesn’t work for this lambda, because we use the elem two times in the implementation. As we saw earlier, each underscore stands for a different parameter. You can’t use an underscore multiple times.

Lambda Version of SuperAdder

Next we will rewrite the superAdder function that we wrote in the previous post as an anonymous function. As a reminder, this is what the superAdder looked like previously:

val superAdder: Function1[Int, Function1[Int, Int]] = new Function1[Int, Function1[Int, Int]] {
  override def apply(x: Int): Function1[Int, Int] = new Function1[Int, Int] {
    override def apply(y: Int): Int = x + y

Rewritten in a lambda, it looks like this:

val superAdder = (x: Int) => (y: Int) => x + y

So this reads as (x: Int) then arrow and this pipes into another function, then (y: Int), and finally arrow into the implementation x + y.

This is the lambda version of the curried function. This is a much more eloquent and shorter way to write it.

We can call it in the same way as before:

println(superAdder(3)(4)) // prints 7

Curried functions will be discussed more in upcoming posts.

Source Code

As always, the source code for this post is available on Github.