Leandro TK
21 Nov 2018
•
15 min read
After a long time learning and working with object-oriented programming, I took a step back to think about system complexity.
"Complexity is anything that makes software hard to understand or to modify." — John Outerhout
Doing some research, I found functional programming concepts like immutability and pure function. Those concepts are big advantages to build side-effect-free functions, so it is easier to maintain systems — with some other benefits.
In this post, I will tell you more about functional programming, and some important concepts, with a lot of code examples.
Functional programming is a programming paradigm — a style of building the structure and elements of computer programs — that treats computation as the evaluation of mathematical functions and avoids changing-state and mutable data — Wikipedia
The first fundamental concept we learn when we want to understand functional programming is pure functions. But what does that really mean? What makes a function pure?
So how do we know if a function is pure
or not? Here is a very strict definition of purity:
It returns the same result if given the same arguments (it is also referred as
deterministic
)
It does not cause any observable side effects
Imagine we want to implement a function that calculates the area of a circle. An impure function would receive radius
as the parameter, and then calculate radius * radius * PI
. In Clojure, the operator comes first, so radius * radius * PI
becomes (* radius radius PI)
:
(def PI 3.14)
(defn calculate-area
[radius]
(* radius radius PI))
(calculate-area 10) ;; returns 314.0
Why is this an impure function? Simply because it uses a global object that was not passed as a parameter to the function.
Now imagine some mathematicians argue that the PI
value is actually 42
and change the value of the global object.
Our impure function will now result in 10 * 10 * 42
= 4200
. For the same parameter (radius = 10
), we have a different result. Let's fix it!
(def PI 3.14)
(defn calculate-area
[radius, PI]
(* radius radius PI))
(calculate-area 10 PI) ;; returns 314.0
TA-DA 🎉! Now we’ll always pass thePI
value as a parameter to the function. So now we are just accessing parameters passed to the function. No external object
.
For the parameters radius = 10
& PI = 3.14
, we will always have the same the result: 314.0
For the parameters radius = 10
& PI = 42
, we will always have the same the result: 4200
If our function reads external files, it’s not a pure function — the file’s contents can change.
(defn characters-counter
[text]
(str "Character count: " (count text)))
(defn analyze-file
[filename]
(characters-counter (slurp filename)))
(analyze-file "test.txt")
Any function that relies on a random number generator cannot be pure.
(defn year-end-evaluation
[]
(if (> (rand) 0.5)
"You get a raise!"
"Better luck next year!"))
Examples of observable side effects include modifying a global object or a parameter passed by reference.
Now we want to implement a function to receive an integer value and return the value increased by 1.
(def counter 1)
(defn increase-counter
[value]
(def counter (inc value))) ;; please don't do this
(increase-counter counter) ;; 2
counter ;; 2
We have the counter
value. Our impure function receives that value and re-assigns the counter with the value increased by 1.
Observation: mutability is discouraged in functional programming.
We are modifying the global object. But how would we make it pure
? Just return the value increased by 1. Simple as that.
(def counter 1)
(defn increase-counter
[value]
(inc value))
(increase-counter counter) ;; 2
counter ;; 1
See that our pure function increase-counter
returns 2, but the counter
value is still the same. The function returns the incremented value without altering the value of the variable.
If we follow these two simple rules, it gets easier to understand our programs. Now every function is isolated and unable to impact other parts of our system.
Pure functions are stable, consistent, and predictable. Given the same parameters, pure functions will always return the same result. We don’t need to think of situations when the same parameter has different results — because it will never happen.
The code’s definitely easier to test. We don’t need to mock anything. So we can unit test pure functions with different contexts:
Given a parameter A
→ expect the function to return value B
Given a parameter C
→ expect the function to return value D
A simple example would be a function to receive a collection of numbers and expect it to increment each element of this collection.
(defn increment-numbers
[numbers]
(map inc numbers))
We receive the numbers
collection, use map
with the inc
function to increment each number, and return a new list of incremented numbers.
(= [2 3 4 5 6] (increment-numbers [1 2 3 4 5])) ;; true
For the input
[1 2 3 4 5]
, the expected output
would be [2 3 4 5 6]
.
Unchanging over time or unable to be changed.
When data is immutable, its state cannot change after it’s created. If you want to change an immutable object, you can’t. Instead, you create a new object with the new value.
In Javascript we commonly use the for
loop. This next for
statement has some mutable variables.
var values = [1, 2, 3, 4, 5];
var sumOfValues = 0;
for (var i = 0; i < values.length; i++) {
sumOfValues += values[i];
}
sumOfValues // 15
For each iteration, we are changing the i
and the sumOfValue
state. But how do we handle mutability in iteration? Recursion! Back to Clojure!
(defn sum
[values]
(loop [vals values
total 0]
(if (empty? vals)
total
(recur (rest vals) (+ (first vals) total)))))
(sum [1 2 3 4 5]) ;; 15
So here we have the sum
function that receives a vector of numerical values. The recur
jumps back into the loop
until we get the vector empty (our recursion base case
). For each "iteration" we will add the value to the total
accumulator.
With recursion, we keep our variables immutable.
Observation: Yes! We can use reduce
to implement this function. We will see this in the Higher Order Functions
topic.
It is also very common to build up the final state of an object. Imagine we have a string, and we want to transform this string into a url slug
.
class UrlSlugify
attr_reader :text
def initialize(text)
@text = text
end
def slugify!
text.downcase!
text.strip!
text.gsub!(' ', '-')
end
end
UrlSlugify.new(' I will be a url slug ').slugify! # "i-will-be-a-url-slug"
In OOP in Ruby, we would create a class, let’s say, UrlSlugify
. And this class will have a slugify!
method to transform the string input into a url slug
.
Beautiful! It’s implemented! Here we have imperative programming saying exactly what we want to do in each slugify
process — first lower case, then remove useless white spaces and, finally, replace remaining white spaces with hyphens.
But we are mutating the input state in this process.
We can handle this mutation by doing function composition, or function chaining. In other words, the result of a function will be used as an input for the next function, without modifying the original input string.
(defn slugify
[string]
(clojure.string/replace
(clojure.string/lower-case
(clojure.string/trim string)) #" " "-"))
(slugify " I will be a url slug ")
Here we have:
trim
: removes whitespace from both ends of a string
lower-case
: converts the string to all lower-case
replace
: replaces all instances of match with replacement in a given string
We combine all three functions and we can "slugify"
our string.
Speaking of combining functions, we can use the comp
function to compose all three functions. Let's take a look:
(defn slugify
[string]
((comp #(clojure.string/replace % #" " "-")
clojure.string/lower-case
clojure.string/trim)
string))
(slugify " I will be a url slug ") ;; "i-will-be-a-url-slug"
Let’s implement a square function
:
(defn square
[n]
(* n n))
This (pure) function will always have the same output, given the same input.
(square 2) ;; 4
(square 2) ;; 4
(square 2) ;; 4
;; ...
Passing “2” as a parameter of the square function
will always returns 4. So now we can replace the (square 2)
with 4. That's it! Our function is referentially transparent
.
Basically, if a function consistently yields the same result for the same input, it is referentially transparent.
pure functions + immutable data = referential transparency
With this concept, a cool thing we can do is to memoize the function. Imagine we have this function:
(+ 3 (+ 5 8))
The (+ 5 8)
equals 13
. This function will always result in 13
. So we can do this:
(+ 3 13)
And this expression will always result in 16
. We can replace the entire expression with a numerical constant and memoize it.
The idea of functions as first-class entities is that functions are also treated as values and used as data.
In Clojure it’s common to use defn
to define functions, but this is just syntactic sugar for (def foo (fn ...))
. fn
returns the function itself. defn
returns a var
which points to a function object.
Functions as first-class entities can:
refer to it from constants and variables
pass it as a parameter to other functions
return it as result from other functions
The idea is to treat functions as values and pass functions like data. This way we can combine different functions to create new functions with new behavior.
Imagine we have a function that sums two values and then doubles the value. Something like this:
(defn double-sum
[a b]
(* 2 (+ a b)))
Now a function that subtracts values and the returns the double:
(defn double-subtraction
[a b]
(* 2 (- a b)))
These functions have similar logic, but the difference is the operators functions. If we can treat functions as values and pass these as arguments, we can build a function that receives the operator function and use it inside our function. Let’s build it!
(defn double-operator
[f a b]
(* 2 (f a b)))
(double-operator + 3 1) ;; 8
(double-operator - 3 1) ;; 4
Done! Now we have an f
argument, and use it to process a
and b
. We passed the +
and -
functions to compose with the double-operator
function and create a new behavior.
When we talk about higher-order functions, we mean a function that either:
takes one or more functions as arguments, or
returns a function as its result
The double-operator
function we implemented above is a higher-order function because it takes an operator function as an argument and uses it.
You’ve probably already heard about filter
, map
, and reduce
. Let's take a look at these.
Given a collection, we want to filter by an attribute. The filter function expects a true
or false
value to determine if the element should or should not be included in the result collection. Basically,if the callback expression is true
, the filter function will include the element in the result collection. Otherwise, it will not.
A simple example is when we have a collection of integers and we want only the even numbers.
Imperative approach
An imperative way to do it with Javascript is to:
create an empty vector evenNumbers
iterate over the numbers
vector
push the even numbers to the evenNumbers
vector
var numbers = [0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10];
var evenNumbers = [];
for (var i = 0; i < numbers.length; i++) {
if (numbers[i] % 2 == 0) {
evenNumbers.push(numbers[i]);
}
}
console.log(evenNumbers); // (6) [0, 2, 4, 6, 8, 10]
We can use the filter
higher order function to receive the even?
function, and return a list of even numbers:
(defn even-numbers
[coll]
(filter even? coll))
(even-numbers [0 1 2 3 4 5 6 7 8 9 10]) ;; (0 2 4 6 8 10)
One interesting problem I solved on Hacker Rank FP Path was the Filter Array problem. The problem idea is to filter a given array of integers and output only those values that are less than a specified value X
.
An imperative Javascript solution to this problem is something like:
var filterArray = function(x, coll) {
var resultArray = [];
for (var i = 0; i < coll.length; i++) {
if (coll[i] < x) {
resultArray.push(coll[i]);
}
}
return resultArray;
}
console.log(filterArray(3, [10, 9, 8, 2, 7, 5, 1, 3, 0])); // (3) [2, 1, 0]
We say exactly what our function needs to do — iterate over the collection, compare the collection current item with x
, and push this element to the resultArray
if it pass the condition.
Declarative approach
But we want a more declarative way to solve this problem, and using the filter
higher order function as well.
A declarative Clojure solution would be something like this:
(defn filter-array
[x coll]
(filter #(> x %) coll))
(filter-array 3 [10 9 8 2 7 5 1 3 0]) ;; (2 1 0)
This syntax seems a bit strange in the first place, but is easy to understand.
#(> x %)
is just a anonymous function that receives x
and compares it with each element in the collection. %
represents the parameter of the anonymous function — in this case the current element inside the filter
.
We can also do this with maps. Imagine we have a map of people with their name
and age
. And we want to filter only people over a specified value of age, in this example people who are more than 21 years old.
(def people [{:name "TK" :age 26}
{:name "Kaio" :age 10}
{:name "Kazumi" :age 30}])
(defn over-age
[people]
(filter
#(< 21 (:age %))
people))
(over-age people) ;; ({:name "TK", :age 26} {:name "Kazumi", :age 30})
Summary of code:
we have a list of people (with name
and age
).
we have the anonymous function #(< 21 (:age %))
. Remember that the %
represents the current element from the collection? Well, the element of the collection is a people map. If we do (:age {:name "TK" :age 26})
, it returns the age value, 26
in this case.
we filter all people based on this anonymous function.
The idea of map is to transform a collection.
The
map
method transforms a collection by applying a function to all of its > elements and building a new collection from the returned values.
Let’s get the same people
collection above. We don't want to filter by “over age” now. We just want a list of strings, something like TK is 26 years old
. So the final string might be :name is :age years old
where :name
and :age
are attributes from each element in the people
collection.
In a imperative Javascript way, it would be:
var people = [
{ name: "TK", age: 26 },
{ name: "Kaio", age: 10 },
{ name: "Kazumi", age: 30 }
];
var peopleSentences = [];
for (var i = 0; i < people.length; i++) {
var sentence = people[i].name + " is " + people[i].age + " years old";
peopleSentences.push(sentence);
}
console.log(peopleSentences); // ['TK is 26 years old', 'Kaio is 10 years old', 'Kazumi is 30 years old']
In a declarative Clojure way, it would be:
(def people [{:name "TK" :age 26}
{:name "Kaio" :age 10}
{:name "Kazumi" :age 30}])
(defn people-sentences
[people]
(map
#(str (:name %) " is " (:age %) " years old")
people))
(people-sentences people) ;; ("TK is 26 years old" "Kaio is 10 years old" "Kazumi is 30 years old")
The whole idea is to transform a given collection into a new collection.
Another interesting Hacker Rank problem was the update list problem. We just want to update the values of a given collection with their absolute values.
For example, the input [1 2 3 -4 5]
needs the output to be [1 2 3 4 5]
. The absolute value of -4
is 4
.
A simple solution would be an in-place update for each collection value.
var values = [1, 2, 3, -4, 5];
for (var i = 0; i < values.length; i++) {
values[i] = Math.abs(values[i]);
}
console.log(values); // [1, 2, 3, 4, 5]
We use the Math.abs
function to transform the value into its absolute value, and do the in-place update.
This is not a functional way to implement this solution.
First, we learned about immutability. We know how immutability is important to make our functions more consistent and predictable. The idea is to build a new collection with all absolute values.
Second, why not use map
here to "transform" all data?
My first idea was to build a to-absolute
function to handle only one value.
(defn to-absolute
[n]
(if (neg? n)
(* n -1)
n))
(to-absolute -1) ;; 1
(to-absolute 1) ;; 1
(to-absolute -2) ;; 2
(to-absolute 0) ;; 0
If it is negative, we want to transform it in a positive value (the absolute value). Otherwise, we don’t need to transform it.
Now that we know how to do absolute
for one value, we can use this function to pass as an argument to the map
function. Do you remember that a higher order function
can receive a function as an argument and use it? Yes, map can do it!
(defn update-list-map
[coll]
(map to-absolute coll))
(update-list-map []) ;; ()
(update-list-map [1 2 3 4 5]) ;; (1 2 3 4 5)
(update-list-map [-1 -2 -3 -4 -5]) ;; (1 2 3 4 5)
(update-list-map [1 -2 3 -4 5]) ;; (1 2 3 4 5)
Wow. So beautiful! 😍
The idea of reduce is to receive a function and a collection, and return a value created by combining the items.
A common example people talk about is to get the total amount of an order. Imagine you were at a shopping website. You’ve added Product 1
, Product 2
, Product 3
, and Product 4
to your shopping cart (order). Now we want to calculate the total amount of the shopping cart.
In imperative way, we would iterate the order list and sum each product amount to the total amount.
var orders = [
{ productTitle: "Product 1", amount: 10 },
{ productTitle: "Product 2", amount: 30 },
{ productTitle: "Product 3", amount: 20 },
{ productTitle: "Product 4", amount: 60 }
];
var totalAmount = 0;
for (var i = 0; i < orders.length; i++) {
totalAmount += orders[i].amount;
}
console.log(totalAmount); // 120
Using reduce
, we can build a function to handle the amount sum
and pass it as an argument to the reduce
function.
(def shopping-cart
[{ :product-title "Product 1" :amount 10 },
{ :product-title "Product 2" :amount 30 },
{ :product-title "Product 3" :amount 20 },
{ :product-title "Product 4" :amount 60 }])
(defn sum-amount
[total-amount current-product]
(+ (:amount current-product) total-amount))
(defn get-total-amount
[shopping-cart]
(reduce sum-amount 0 shopping-cart))
(get-total-amount shopping-cart) ;; 120
Here we have shopping-cart
, the function sum-amount
that receives the current total-amount
, and the current-product
object to sum
them.
The get-total-amount
function is used to reduce
the shopping-cart
by using the sum-amount
and starting from 0
.
Another way to get the total amount is to compose map
and reduce
. What do I mean by that? We can use map
to transform the shopping-cart
into a collection of amount
values, and then just use the reduce
function with +
function.
(def shopping-cart
[{ :product-title "Product 1" :amount 10 },
{ :product-title "Product 2" :amount 30 },
{ :product-title "Product 3" :amount 20 },
{ :product-title "Product 4" :amount 60 }])
(defn get-amount
[product]
(:amount product))
(defn get-total-amount
[shopping-cart]
(reduce + (map get-amount shopping-cart)))
(get-total-amount shopping-cart) ;; 120
The get-amount
receives the product object and returns only the amount
value. So what we have here is [10 30 20 60]
. And then the reduce
combines all items by adding up. Beautiful!
We took a look at how each higher-order function works. I want to show you an example of how we can compose all three functions in a simple example.
Talking about shopping cart
, imagine we have this list of products in our order:
(def shopping-cart
[{ :product-title "Functional Programming" :type "books" :amount 10 },
{ :product-title "Kindle" :type "eletronics" :amount 30 },
{ :product-title "Shoes" :type "fashion" :amount 20 },
{ :product-title "Clean Code" :type "books" :amount 60 }])
We want the total amount of all books in our shopping cart. Simple as that. The algorithm?
filter by book type
transform the shopping cart into a collection of amount using map
combine all items by adding them up with reduce
Done! 🎉
I’ve organised some resources I read and studied. I’m sharing the ones that I found really interesting. For more resources, visit my Functional Programming Github repository.
Hey people, I hope you had fun reading this post, and I hope you learned a lot here! This was my attempt to share what I’m learning.
Here is the repository with all codes from this article.
Come learn with me. I’m sharing resources and my code in this Learning Functional Programming repository.
I hope you saw something useful to you here. And see you next time! :)
My Twitter, Github, and Medium. ☺
TK.
Ground Floor, Verse Building, 18 Brunswick Place, London, N1 6DZ
108 E 16th Street, New York, NY 10003
Join over 111,000 others and get access to exclusive content, job opportunities and more!