Functional design: Algebraic Data Types

gcanti profile image Giulio Canti Updated on ・5 min read

A good first step while building a new application is to define its domain model. TypeScript offers many tools to help you in this task. Algebraic Data Types (ADT for short) are one of these tools.

What is an ADT?

In computer programming, especially functional programming and type theory, an algebraic data type is a kind of composite type, i.e., a type formed by combining other types.

Two common classes of algebraic types are:

  • product types
  • sum types

Product types

A product type is a collection of types Ti indexed by a set I.

Two common members of this family are n-tuples, where I is a non empty interval of natural numbers...

type Tuple1 = [string] // I = [0]
type Tuple2 = [string, number] // I = [0, 1]
type Tuple3 = [string, number, boolean] // I = [0, 1, 2]

// Accessing by index
type Fst = Tuple2[0] // string
type Snd = Tuple2[1] // number

...and structs, where I is a set of labels.

// I = {"name", "age"}
interface Person {
  name: string
  age: number

// Accessing by label
type Name = Person['name'] // string
type Age = Person['age'] // number

Why "product" types?

If we write C(A) for the number of inhabitants of the type A (aka its cardinality) then the following equality holds

C([A, B]) = C(A) * C(B)

the cardinality of the product is the product of cardinalities


type Hour = 1 | 2 | 3 | 4 | 5 | 6 | 7 | 8 | 9 | 10 | 11 | 12
type Period = 'AM' | 'PM'
type Clock = [Hour, Period]

The Clock type has 12 * 2 = 24 inhabitants.

When should I use a product type?

Whenever its components are independent.

type Clock = [Hour, Period]

Here Hour and Period are independent, i.e. the value of Hour doesn't affect the value of Period and vice versa, all pairs are legal and meaningful.

Sum types

A sum type is a data structure used to hold a value that could take on several different, but fixed, types. Only one of the types can be in use at any one time, and a tag field explicitly indicates which one is in use.

In the TypeScript documentation they are named tagged union types.

Example (redux actions)

type Action =
  | {
      type: 'ADD_TODO'
      text: string
  | {
      type: 'UPDATE_TODO'
      id: number
      text: string
      completed: boolean
  | {
      type: 'DELETE_TODO'
      id: number

The type field is the tag and ensures that its members are disjoint.


A sum type with n members requires n constructors, one for each member

const add = (text: string): Action => ({
  type: 'ADD_TODO',

const update = (id: number, text: string, completed: boolean): Action => ({
  type: 'UPDATE_TODO',

const del = (id: number): Action => ({
  type: 'DELETE_TODO',

Sum types can be polymorphic and/or recursive.

Example (linked lists)

//        ↓ type parameter
type List<A> = { type: 'Nil' } | { type: 'Cons'; head: A; tail: List<A> }
//                                                              ↑ recursion

Pattern matching

JavaScript doesn't have pattern matching (and so TypeScript) however we can define a "poor man" pattern matching by defining a fold function

const fold = <A, R>(
  fa: List<A>,
  onNil: () => R,
  onCons: (head: A, tail: List<A>) => R
): R => (fa.type === 'Nil' ? onNil() : onCons(fa.head, fa.tail))

Example (calculate the length of a List recursively)

const length = <A>(fa: List<A>): number =>
  fold(fa, () => 0, (_, tail) => 1 + length(tail))

Why "sum" types?

The following equality holds

C(A | B) = C(A) + C(B)

the cardinality of the sum is the sum of cardinalities

Example (the Option type)

type Option<A> =
  | { type: 'None' }
  | {
      type: 'Some'
      value: A

From the general formula C(Option<A>) = 1 + C(A) we can derive for example the cardinality of Option<boolean>: 1 + 2 = 3 inhabitants.

When should I use a sum type?

When its components would be dependent if implemented as a product type.

Example (component props)

interface Props {
  editable: boolean
  onChange?: (text: string) => void

class Textbox extends React.Component<Props> {
  render() {
    if (this.props.editable) {
      // error: Cannot invoke an object which is possibly 'undefined' :(

The problem here is that Props is modelled as a product type but onChange depends on editable.

A sum type is a better choice

type Props =
  | {
      type: 'READONLY'
  | {
      type: 'EDITABLE'
      onChange: (text: string) => void

class Textbox extends React.Component<Props> {
  render() {
    switch (this.props.type) {
      case 'EDITABLE' :
        this.props.onChange(...) // :)

Example (node callbacks)

declare function readFile(
  path: string,
  //         ↓ ---------- ↓ CallbackArgs
  callback: (err?: Error, data?: string) => void
): void

The result is modelled as a product type

type CallbackArgs = [Error | undefined, string | undefined]

however, its components are dependent: you get either an error or a string

err data legal?
Error undefined
undefined string
Error string
undefined undefined

A sum type would be a better choice, but which one?

Functional error handling

Let's see how to deal with errors in a functional style.

The Option type

The Option type represents the effect of a computation that may fail or yield a value of type A

type Option<A> =
  | { type: 'None' } // represents a failure
  | { type: 'Some'; value: A } // represents a success

Constructors and pattern matching:

// a nullary constructor can be implemented as a constant
const none: Option<never> = { type: 'None' }

const some = <A>(value: A): Option<A> => ({ type: 'Some', value })

const fold = <A, R>(fa: Option<A>, onNone: () => R, onSome: (a: A) => R): R =>
  fa.type === 'None' ? onNone() : onSome(fa.value)

The Option type can be used to avoid throwing exceptions and/or to represent optional values, so we can go from...

//                this is a lie ↓
const head = <A>(as: Array<A>): A => {
  if (as.length === 0) {
    throw new Error('Empty array')
  return as[0]

let s: string
try {
  s = String(head([]))
} catch (e) {
  s = e.message

...where the type system is unaware of possible failures, to...

//                              ↓ the type system "knows" that this computation may fail
const head = <A>(as: Array<A>): Option<A> => {
  return as.length === 0 ? none : some(as[0])

const s = fold(head([]), () => 'Empty array', a => String(a))

...where the possibility of failures is lifted to the type system.

The Either type

A common use of Either is as an alternative to Option for dealing with possible missing values. In this usage, None is replaced with a Left which can contain useful information. Right takes the place of Some. Convention dictates that Left is used for failure and Right is used for success.

type Either<L, A> =
  | { type: 'Left'; left: L } // represents a failure
  | { type: 'Right'; right: A } // represents a success

Constructors and pattern matching:

const left = <L, A>(left: L): Either<L, A> => ({ type: 'Left', left })

const right = <L, A>(right: A): Either<L, A> => ({ type: 'Right', right })

const fold = <L, A, R>(
  fa: Either<L, A>,
  onLeft: (left: L) => R,
  onRight: (right: A) => R
): R => (fa.type === 'Left' ? onLeft(fa.left) : onRight(fa.right))

Getting back to our callback example

declare function readFile(
  path: string,
  callback: (err?: Error, data?: string) => void
): void

readFile('./myfile', (err, data) => {
  let message: string
  if (err !== undefined) {
    message = `Error: ${err.message}`
  } else if (data !== undefined) {
    message = `Data: ${data.trim()}`
  } else {
    // should never happen
    message = 'The impossible happened'

we can change its signature to

declare function readFile(
  path: string,
  callback: (result: Either<Error, string>) => void
): void

and then consume the API like this

readFile('./myfile', e => {
  const message = fold(e, err => `Error: ${err.message}`, data => `Data: ${data.trim()}`)


In this post we've seen product types and sum types, and how reasoning in terms of the number of states they represent can greatly affect the design of our domain models.

A common pitfall of many real-world APIs is to misuse product types that, in addition to all the legal states, also model many illegal ones.

Sum types are an incredibly useful and fundamental language feature, they are the key to design excellent domain models by allowing to make illegal states irrepresentable.

Posted on by:


markdown guide

Great article !
What do you think about this approach to pattern matching ?
I think it looks a bit more declarative then checking the type manually.

type Tagged = {_tag: string}

type Tags<TUnionType extends Tagged> = TUnionType['_tag']

type TagTypeMap<TTags extends string, TUnionType extends Tagged> = {
  [TTag in TTags]: TUnionType extends {_tag: TTag} ? TUnionType : never

type Pattern<TResult, TTagTypeMap> = {
  [K in keyof TTagTypeMap]: (item: TTagTypeMap[K]) => TResult

type Patt<TResult, TUnionType extends Tagged> =
  Pattern<TResult, TagTypeMap<Tags<TUnionType>, TUnionType>> 

const match = <TUnionType extends Tagged, TObject extends TUnionType, TResult>
  (pattern: Patt<TResult, TUnionType>) =>
    (object: TObject) =>
      pattern[object._tag][object] as TResult

// ------------------Sample------------------
type Square = {len: number, _tag: 'Square'}
type Circle = {rad: number, _tag: 'Circle'}
type Rectangle = {xlen: number, ylen: number, _tag: 'Rectangle'}

type Shape = Square | Circle | Rectangle

const shapes = [
  {len: 1} as Square,
  {rad: 2} as Circle,
  {xlen: 3, ylen: 4} as Rectangle

const shapeDescription: Patt<string, Shape> = {
  Square: ({len}) => `I am square with length ${len}`,
  Circle: ({rad}) => `I am circle with radius ${rad}`,
  Rectangle: ({xlen, ylen}) => `I am rectangle with sides ${xlen} and ${ylen}`

const descriptions = shapes.map(match(shapeDescription))

There is a small typo and don't know if one can submit a change to a post.

const fold = <A, R>(fa: Option<A>, onNone: () => R, onSome: (a: A) => R): R =>
  fa.type === 'None' ? onNone() : onSome(fa.value)
onNone: () => R

The following code block declare:

const s = fold(head([]), 'Empty array', a => String(a))

Its should have been:

const s = fold(head([]), () => 'Empty array', a => String(a))

Are you aware of ReasonML, which have real support for ADTs and pattern matching? 🙂