JavaScript Functional Programming: Advanced Topics

In the previous articles we did a small introduction and history of functional programming and we talked about some basic concepts regarding Functional Programming.

Now, we are going to discuss some functional programming patterns that allows you to achieve immutability and composition. Also, we’ll discuss how to achieve loops without running into errors which is one of the main problems with functional programming.



Currying is the mean to transform a function of arity n to n functions of arity 1 - @michelre

That sound very scary, doesn’t it?

I assure you is not, but to understand currying, its better if we start with a basic pure function that adds two numbers.

const addValues = (a, b) => {
  return a + b
console.log(addValues(6, 4))

Pretty straight forward, right? The output in the console would be:

$ node currying.js

That function is pure, but is not reusable in any way. So lets add some re-usability to that code by converting it into a High Order Function like so:

const highOrderAdd = a => {
  return b => {
    return a + b

Lets dissect what we did:

This is Currying!

The idea is that instead of having a function that receives n arguments, you create n functions, each one receiving only 1 argument. Let’s do another example:

// Instead of this
const addValues = (a, b, c, d) => a + b + c + d

// Do this
const highOrderAdd = a => b => c => d => a + b + c + d

Why would be this useful???

Well, we can reuse code this way. For instance, we could create a function that takes one number and sums 6 to it by using as base the highOrderAdd function:

// Use case of highOrderAdd:
const addSix = highOrderAdd(6)

In a way addSix is a use case for the highOrderAdd function.

Don’t worry if you think that this is a waste of time and this just makes things more complicated. Read along about composition to see why this is useful.


If you read the first part of this series you can see that I already touched on composition. But I didn’t explain why it was useful.

Again, lets start with some conventional functional code:

// composition.js

// Functions
const safeCapitalize = str => {
  return str.toLowerCase().replace(/^.{2}/, c => c.toUpperCase())
const removeDoubleSpaces = str => {
  return str.replace(/ {2,}/g, " ").trim()
const addPoint = str => {
  return str.replace(/\.*$/, "") + "."

// Standard usage:
  addPoint(removeDoubleSpaces(safeCapitalize(" THIS        is a string....")))

This will output.

$ node composition.js
This is a string.

As you can see in the console.log section at the end, this can become problematic. Just making sure that all parenthesis are accounted for can be a headache!

Now, you might say: “Why don’t you just save the out put of each call on a variable?”.

There are 2 problems with that approach:

And this is where [Currying](#currying] starts to make sense. With the help of higher order functions, we can solve this issue with a new function that we’ll cleverly call compose:

// Functions do not change
const safeCapitalize = str => {
  return str.toLowerCase().replace(/^.{2}/, c => c.toUpperCase())
const removeDoubleSpaces = str => {
  return str.replace(/ {2,}/g, " ").trim()
const addPoint = str => {
  return str.replace(/\.*$/, "") + "."

// Composition function
const compose = (...fns) => arg => {
  return fns.reduce((composed, fn) => fn(composed), arg)

// Using composition with currying
)(" THIS        is  ANOTHER string....")

Isn’t this cleaner and declarative???

The end result is exactly the same. Only that now you can see very clearly what (not how) the code is doing just by reading the function names.

The compose function takes functions as arguments (we use here the spread operator for that), and returns a single function that receives one argument.

Also, here you can see that currying can be useful since all the functions (except compose) require that only one parameter be passed.

Loops with recursion

Note: Tail Call Optimization no longer works in Node. Right now long recursion calls will give you a RangeError: Maximum call stack size exceeded always

The last pattern we’ll discuss is recursion. Which is an essential part of Functional Programming since the use of loops like for and while is discouraged because they are not Declarative nor they are Immutable (the loop control variable has to be overwritten on each iteration).

The idea is to make use of Lazy Evaluation, which in JavaScript is represented as Tail Call Optimization, to prevent call stack range errors.

This time we’re going to take some code that is already Functional but we are going to point out an issue it has:

// recursion.js
const sum = n => {
  if (n === 0) return 0
  return n + sum(n - 1)

This is a simple recursion that sums the numbers from 1 to n. So if I where to call this function with n = 100, I would get in the console 5050.

So far so good.

The problem lies in the return statement: The function is written in a way where before returning the current value, the function sum has to call upon itself again with the next value.

So I where to call this function with the n = 100000 I would get this:

const sum = (n) => {

RangeError: Maximum call stack size exceeded

In plain English, what this error points out is that you filled the call stack of the interpreter before returning the last value. This issue can be solved by using Tail Call Optimization.

Tail Call Optimization is not supported by all node versions.

To solve this issue, we need to rewrite the function in a way where the function not only receives the next value, but the current accumulated result so it doesn’t have to wait for the next return to execute.

This sound complicated, but is really easy:

const sumAccum = (n, accum = 0) => {
  if (n === 0) return accum
  return sumAccum(n - 1, n + accum)

Two things to notice here:

This 2 changes prevents the call stack to get filled since were doing our computation before the next operation and not after.

Final toughs

Rémi Michelre wrote 2 great articles about currying and composition. I highly suggest you take a look at them if you still have any doubts on those 2 concepts.

Also, there is a great explanation on the problems with recursion in this Jeremy Fairbank talk in YouTube (timestamp in the link).