Preface

This post has many awesome pictures which credits go to Aditya Bhargava. His original article Functors, Applicatives, And Monads In Pictures is extremely well written, with sample code in Haskell though.

In this post I will try to provide proof of concept of Functor/Applicatives/Monad in pure Swift, plus example for using Reader Monad for Dependency Injection(DI), and the idea of Try monad concept from Scala.

Maybe is Functor

We all know the Optional Type (the ? mark) in Swift. We can define our option type named Maybe using enum.

enum Maybe<T> {
	case Just(T)
	case Nothing
}

Simple enough! A Maybe type is a "box" which can contains the value or ... nothing

The interesting part come from here: we can define a fmap function which takes a normal function and a Maybe type, then return another Maybe

How does fmap look like? Well, its implement is not that hard

extension Maybe {
	func fmap<U>(f: T -> U) -> Maybe<U> {
		switch self {
			case .Just(let x): return .Just(f(x))
			case .Nothing: return .Nothing
		}
	}
}

And the "magic" above is actually not-so-magical. At this time, our Maybe type is already a Functor.

Maybe is Applicatives

Applicatives is a type which can define the function apply that

Take a function wrapped in that type
Take also a value wrapped in that type
Then return a new value which is wrapped also

I will define an apply function for Maybe

extension Maybe {
	func apply<U>(f: Maybe<T -> U>) -> Maybe<U> {
		switch f {
			case .Just(let JustF): return self.fmap(JustF)
			case .Nothing: return .Nothing
		}
	}
}

That is it! Our Maybe now is both Functor and Applicatives.

Maybe is Monad

How to learn about Monads:

Get a PhD in computer science.

Throw it away because you don’t need it for this section

Maybe can be considered as a monad if it can define a function flatmap that

Take a function which return type is Maybe
Take also a value wrapped in Maybe
Return another Maybe

Let's Swift! Here is our flatMap and bind operator >>=

extension Maybe {
	func flatMap<U>(f: T -> Maybe<U>) -> Maybe<U> {
		switch self {
			case .Just(let x): return (f(x))
			case .Nothing: return .Nothing
		}
	}
}
infix operator >>= { associativity left }
func >>=<T, U>(a: Maybe<T>, f: T -> Maybe<U>) -> Maybe<U> {
	return a.flatMap(f)
}

Suppose that we already have a half function that return an Maybe type

func half(a: Int) -> Maybe<Int> {
	return a % 2 == 0 ? Maybe.Just(a / 2) : Maybe.Nothing
}

Then with the >>= operator you can chain Maybe like:

Maybe.Just(20) >>= half >>= half >>= half

And this is how it is actually processed

Now our Maybe is Functor, Applicatives and also Monad as well.

A step further, the Reader monad

In this section I will introduce minimal version for one of three useful Monads: the Reader Monad

class Reader<E, A> {
	let g: E -> A
	init(g: E -> A) {
		self.g = g
	}
	func apply(e: E) -> A {
		return g(e)
	}
	func map<B>(f: A -> B) -> Reader<E, B> {
		return Reader<E, B>{ e in f(self.g(e)) }
	}
	func flatMap<B>(f: A -> Reader<E, B>) -> Reader<E, B> {
		return Reader<E, B>{ e in f(self.g(e)).g(e) }
	}
}

As you can see, we have map, and flatMap function here. This class type is both Functor and Monad at the same time. Very same with Maybe monad above, Reader can define infix operator and chain to whenever we want.

infix operator >>= { associativity left }
func >>=<E, A, B>(a: Reader<E, A>, f: A -> Reader<E, B>) -> Reader<E, B> {
	return a.flatMap(f)
}

func half(i: Float ) -> Reader<Float , Float> {
	return Reader{_ in i/2}
}
let f = Reader{i in i} >>= half >>= half >>= half
f.apply(20) // 2.5

Why Reader monad matter

Reader monad take g function in init time. By switching (or injecting) this function, we can create our own Dependency Injection(DI) framework easily. Let's see an example:

This is our model:

struct User {
	var name: String
	var age: Int
}
struct DB {
	var path: String
	func findUser(userName: String) -> User {
		// DB Select operation
		return User(name: userName, age: 29)
	}
	func updateUser(u: User) -> Void {
		// DB Update operation
		print(u.name + " in: " + path)
	}
}

and usage:

let dbPath = "path_to_db"
func update(userName: String, newName: String) -> Void {
	let db = DB(path: dbPath)
	var user = db.findUser(userName)
	user.name = newName
	db.updateUser(user)
}
update("dummy_id", newName: "Thor")
// Thor in: path_to_db

In real life DB may be compicated and seperated as a whole infrastructure layer. Assume that DB can find an user by his name and update his information to the Database.

The problem is update function now holding a reference to dbPath, which I want to switch during test or runtime. I will rewrite the update function to return only a Reader

struct Environment {
	var path: String
}
func updateF(userName: String, newName: String) -> Reader<Environment, Void> {
	return Reader<Environment, Void>{ env in
		let db = DB(path: env.path)
		var user = db.findUser(userName)
		user.name = newName
		db.updateUser(user)
	}
}

then call Reader.apply later base on what passed through Environment variable.

let test = Environment(path: "path_to_sqlite")
let production = Environment(path: "path_to_realm")
updateF("dummy_id", newName: "Thor").apply(test)
// Thor in: path_to_sqlite
updateF("dummy_id", newName: "Thor").apply(production)
// Thor in: path_to_realm

The Try Monad

Scala's Try type is a functional approach for error handling. Very likely to Optional (or Maybe), Try is a "box" that contains value or a Throwable if something has gone wrong. Try can be a Successful or a Failure.

enum Try<T> {
	case Successful(T)
	case Failure(ErrorType)
	init(f: () throws -> T) {
		do {
			self = .Successful(try f())
		} catch {
			self = .Failure(error)
		}
	}
}

To make Try a functor/monad, I will add map and flatMap function

extension Try {
	func map<U>(f: T -> U) -> Try<U> {
		switch self {
			case .Successful(let value): return .Successful(f(value))
			case .Failure(let error): return .Failure(error)
		}
	}
	func flatMap<U>(f: T -> Try<U>) -> Try<U> {
		switch self {
			case .Successful(let value): return f(value)
			case .Failure(let error): return .Failure(error)
		}
	}
}

With an operation which can throws some ErrorType, just wrap them inside a Try and chain(with map and flatMap) to whenever you want. At every step the result will be a Try type. When you want the real value inside that box, just do a pattern matching.

enum DoomsdayComing: ErrorType {
	case Boom
	case Bang
}
let endOfTheWorld = Try {
	throw DoomsdayComing.Bang
}
let result = Try {4/2}.flatMap { _ in endOfTheWorld}
switch result {
	case .Successful(let value): print(value)
	case .Failure(let error): print(error)
}
// Bang

Conclusion

A functor is a type that implements map.
An applicative is a type that implements apply.
A monad is a type that implements flatMap.

Maybe have map, apply and flatMap, so it is a functor, an applicative, and a monad.
Reader is a monad which can be used for DI(Dependency Injection)
Try is a monad, and so as Future, Signal or Observable (see their flatMap implement!)

Thanks for reading this article and feel free to give any feedback or suggestion. You can open a pull request or reach me out at @dtvd88. If you want to play around with above class, clone this repo and open the Playground.

Many thanks to Aditya Bhargava for his awesome blog.

orakaro/Swift-monad-Maybe-Reader-and-Try