Functional Programming Review

Basic functional programming technique in Elixir and JavaScript.

Other Resources

Medium post on Functional Programming in JS


In Web Development we’re using a functional language (Elixir) and some JavaScript libraries that expect us to manipulate data in a functional style (React, Redux).

This document provides some simple examples of what that looks like.

Functional Concepts

Functional programming is about values, and functions that create new values based on existing ones.

A good example of this is comparing the default “sort” methods in Python and Ruby.

# Python
>>> xs = [1,4,3,2]
>>> xs.sort()
>>> xs
[1, 2, 3, 4]

This is common to imperative style proramming. The sort method mutates the array, sorting the elements in place. Once the sort method has been called, the original unsorted array value is no longer available.

# Ruby
irb> xs = [1,4,3,2]
=> [1, 4, 3, 2]
irb> ys = xs.sort
=> [1, 2, 3, 4]
irb> xs
=> [1, 4, 3, 2]
irb> ys
=> [1, 2, 3, 4]

This variant is required in functional style programming. The sort method creates a new array containing the same element values as the original in a different order. The original unsorted array value is unchanged.

Because data isn’t mutated in place, various techniques can be used that rely on data values not changing. This can be used for performance in some cases, but primarily serves as a way to make programs easier to reason about.

Immutable data, either as a language feature or as a design pattern, allows for a couple of really useful optimizations:

Immutable data also has an amazing feature in the presence of concurrency (e.g. threads) - it completely prevents data races. A data race requires that data be both shared and modified. This can allow for some styles of programming to completely avoid locks while still gaining the benefits of concurrent (and parallel) execution.

To generalize, mutation is just a specific example of a more general concept that functional style programs try to avoid / isolate: Side effects. A function without side effects is called a “pure” function - for a given set of input values it will always produce the same output value. This lets you do some neat tricks:

Simple Data

Most languages have a concept of “primitive” data types that are innately treated as immutable values. Numbers are the simplest example: The number 4 always has the value 4. You can assign a different number to a variable or object field, but you can’t change the value itself.

Java handles strings this way. You can’t mutate a Java string, instead the StringBuffer class exists for cases where in-place string manipulation is desired. In a functional style, you don’t get StringBuffer.

Writing functions on simple data values works the same in a functional style as in any language.


  function square(a) {
    return a * a; 
  function pluralize(word) {
   return word + "s"; 


  def square(a) do
    a * a 
  function pluralize(word) do
    word <> "s" 


Here “struct” is a general term for an object with a fixed set of fields. This is a built-in concept in Elixir, and a design pattern in JavaScript.

In Java, the idea of an immutable struct is referred to as a “value object”.

For example, we might have a type of object we call a “person”, with three fields:

Name and birth date don’t change much, but sometimes occupation changes. To hande this, we produce a new struct with the same fields:


// In JS, a struct is just an object where we've decided on a fixed
// set of fields.

function set_job(person, job) {
  return {
    dob: person.dob,
    job: job,

// Shorthand:
function set_job1(person, job) {
  return {...person, job: job};


defmodule Person do
  # Set defaults
  defstruct name: "Bob", dob: ~D[2000-01-02], job: "Fisherman"

  def set_job(person, job) do
       dob: person.dob,
       job: job,

  # Shorthand:
  def set_job1(person, job) do
    %{ person | job: job }

# We can also just use a Map with a fixed set of fields, like in JS.

In Elixir, an common alternative for structs is the Tuple. It has multiple fields distingushed by position and is convenient to use with pattern matching.

# Compute the distance between two points
def distance({x0, y0}, {x1, y1}) do
  dx = x1 - x0
  dy = y1 - y0
  :math.sqrt(dx*dx + dy*dy)

iex> origin = {0, 0}
iex> dest = {10, 0}
iex> distance(origin, dest)

Arrays and Lists

We need a data type to represent sequences of values. Generally each value in a sequence is of the same type (e.g. “List of Number”, “List of Person”, “List of List of Number”).

The native functional sequence structure is the linked list because it allows structural sharing, but in JavaScript the default sequence type is the variable length array (C++ “vector”, Java “ArrayList”).

The standard pattern for traversing an array is a loop, while the standard pattern for traversing a list is recursion.

Arrays are innately constructed by mutation, so to program with them in a functional style we want to hide that mutation by doing it only locally within the fuction constructing the array. The efficient way to add single items to an array is with “push”.

We can completely hide this mutation by using standard higher order sequence processing functions like “map”, “reduce”, and “filter”.

Reverse a sequence of numbers.


function reverse(xs) {
    // Result is a new array.
    // Mutation only occurs on an object in the function where that
    // object is constructed. No references to the partially constructed
    // object are allowed to leak.
    let ys = [];
    for (let ii = xs.length - 1; ii >= 0; --ii) {
    return ys;


# This is inefficient, using O(n^2) time and O(n) extra stack space.
def reverse1(xs) do
  if xs == [] do
    reverse(tl(xs)) ++ [hd(xs)]

# The same with pattern matching.
def reverse2([]), do: []
def reverse2([x|xs]), do: reverse(xs) ++ [x]

# This is linear time and no extra space, using an accumulator.
# Algorithmically equivilent to the JS code above.
def reverse3(xs), do: reverse3(xs, [])
def reverse3([], ys), do: ys
def reverse3([x|xs], ys), do: reverse(xs, [y|ys])

# The standard list function "reduce" generalizes the
# above accumulator pattern.
def reverse4(xs) do
  Enum.reduce xs, [], fn (x, ys) ->
    [x | ys] 

# One line, "&(&1)" is shorthand for "fn (x) -> x end"
def reverse5(xs), do: Enum.reduce xs, [], &([&1 | &2])

Given a sequence of names, generate a sequence of “Hello, {name}”


function hello_all(names) {
  let ys = []
  for (name of names) {
    ys.push("Hello, " + name) 
  return ys;

# Use the standard "map" method and template strings
function hello_all2(names) {
  return (name) {
    return `Hello, ${name}`

# Using the lodash helper, usually because you forgot about built-in
# name method on array.
let hello_all3 = (names) =>, (name) => ("Hello, " + name));


def hello_all([]), do: []
def hello_all([name|rest]) do
  hello = "Hello, #{name}"
  [hello | hello_all(rest)]

def hello_all1(names) do names, fn (name) ->
    "Hello, #{name}" 

Find all the people in a list who were born before the year 2000.


function find_old(people) {
  let cutoff = new Date("2000-01-01");
  let ys = [];
  for (person of people) {
    if (person.dob < cutoff) {
  return ys;

function find_old1(people) {
  let cutoff = new Date("2000-01-01");
  return people.filter((person) => person.dob < cutoff));


def find_old([]), do: []
def find_old([person | rest]) do
  cutoff = ~D[2000-01-01]
  if, cutoff) == :lt do
    [person | find_old(rest)] 

def find_old1(people) do
  cutoff = ~D[2000-01-01]
  Enum.filter people, fn (person) ->, cutoff) == :lt 


In addition to sequences, the other super-common data structure is the key-value map. In imperative languages, this tends to be a hash table (HashMap in Java), while in functional languages it tends to be a tree to take advantage of structural sharing.

In JavaScript, the standard “object” type is frequently used as a map, but it has the constraint that strings must be keys. Recent versions of JavaScript have a dedicated Map type (make one with “new Map”) that properly handles non-string keys.

Elixir also has two map types. It has the standard map type constructed with “%{ … }”, and also uses association lists, or lists of key-value tuples. Association lists are less efficient for random access when they get big, but are extremely useful as an intermediate structure for constructing maps from lists.

Like with arrays, JavaScript objects and maps are innately constructed by mutation. The same pattern as with arrays should be followed: limit mutation to the function that constructs the object.


Some of the methods on arrays and maps in JavaScript follow the functional style (e.g. .map and .filter), while others mutate their inputs. Lodash provides a set of common functions that consistently produce new values rather than mutating.

Immutable collections in JavaScript

Following a functional style in JavaScript works great when using the built in data types with small collections, as long as you’re consistent about following the no-mutation constraint.

For larger data, or to avoid mutating collections by mistake, it can be useful to have proper immutable collection types in JavaScript. These are available in a library called immutable.js.