1.0.5 • Public • Published


A simple library for complex logic.

  • 🐭 4KB Gzipped!!!
  • 💾 Serializable (perfect for localStorage)
  • 🎁 Powerful OOTB (Functor, Monad, Bifunctor and more!)
  • 🛠 Extensible ( Open specification, data first )
  • 🚀 Makes UI and API code safer, cleaner and more fun.

Live Example

Live Demo

Quick Start

npm install sum-type
// let T = SumType

import * as T from 'sum-type'

const Loaded = 

const loaded = 
	Loaded.Y("Hello World")

const loading = 

const render = 
		x => `Loading: ${x}%`
		, x => `Loaded: ${x}`

const transform =
		x => x.toUpperCase()
[ render( transform( loaded ) ) //=> 'Loaded: HELLO WORLD'
, render( transform( loading ) ) //=> 'Loading: 55%'
, render( transform( loading ) ) //=> 'LOADING: 55%'
.forEach( x => document.write(`<p>${x}</p>`))

Live Demo



Recommended usage is to simply npm install sum-type and import * as T from 'sum-type

Browser / Playground

Checkout all the distributions here:


If you require('sum-type') it will use the .cjs distribution out of the box


ESM is the native format for this library. Simply import * as T from 'sum-type'

What is it

Freedom from booleans.

Scenario: You've solving a moderately difficult problem, and there's a degree of data modelling involved. You've got several booleans for tracking loading states, save states, modified states, selected states and on and on.

Oh! and you're tracking all those states for every item in a list separately.

Depending on a specific combination of these boolean flags you need to render differently, talk to the server differently, persist state differently.

It very quicky becomes a mess. We reach for complex tools to help us manage the mess. But instead all we needed was to make it impossible to not traverse every single state.

First step. Create a sum-type for every state we want to track.

import * as S from 'sum-type'

const Data = S.tags('Data', 
	[ 'Deselected'
	, 'Loading'
	, 'Modified'
	, 'Saved'

Our type Data has been generated with 4 tags, each has a constructor:

//=> { type: 'Data', tag: 'Y', value: data }

//=> { type: 'Deselected', tag: 'N' }

Our constructors just tag our data with a label that allows us to build logic on top of it. Normally this label would be stored separately to the data.

Something like:

var selected = false
var data = ...

Which is fine at first, until it looks like this:

var selected = true
var loaded = true
var modified = true
var saved = false
var data = x

We'd model that in sum-type like so:

import * as S from 'sum-type'

const Data = S.tags('Data', 
	[ 'Deselected'
	, 'Loading'
	, 'Modified'
	, 'Saved'

const data = Data.Modified(x)

When we want to transform the value of data we can use any number of useful helpers. Like map<Tag>, chain<Tag>, get<Tag>With or get<Tag>Or where <Tag> is one of your tag names. E.g. mapSaved.

const f = Data.mapSaved(
	x => x * 2

f( Data.Saved(2) )
// => Data.Saved(4)

f( Data.Modified(1) )

const g = Data.getModifiedOr(0)

g( Data.Saved(2) )
//=> 0

g( Data.Modified(3) )
//=> 3

🤓 Types generated by sum-type are decorated with a lot of functions for free. Check out lib/decorate.js to see how it all works.

You can also fold all the cases, as every type generated by sum-type is a discrimated union. Which means (in layperson's terms), you'll get a helpful error if you don't account for every possibly tag.

const NoData = S.otherwise(['Selected', 'Loading'])

const f = 
		... NoData( () => 'Nothing' )
		,   Saved: x => 'Saved: ' + x
		,   Modified: x => 'Modified: ' + x

f( Data.Loading() )
//=> 'Nothing'

f( Data.Saved('cool') )
//=> 'Saved: Cool'

There's loads of other helpful utilities, and sum-type will guide you with helpful error messages if you make a mistake. That's one of our design goals!

Helpful Errors

If we pass the wrong data structure into our composition, we will get a specific, helpful error message explaining what your type looked like and what that particular method was expecting.

Your stack trace is going to be particularly legible because sum-type internally avoids point free composition and auto currying, and will report errors at the first incorrect invocation.

So even though fold requires 3 calls:

fold (Type) ({ Y: () => 'hi', N: () => 'bye' }) ( Type.Y() )

sum-type will error if any one of those invocations didn't satisfy the required constraints.

🤓 Every error that sum-type yields, is itself a transformation of a sum-type. All the error types are documented in the Errors section


sum-type differentiates itself from other sum type library by documenting the internal structure used for types and instances of types. This allows you to create your own constructors/transformers in userland. You can store the exact output of a sum-type constructor in a redux-store, localStorage or even a json column in your favourite database.

sum-type does not care where your data came from, just that it adheres to a particular structure.


Each module listed here adheres to the sum-type specification. That specification is defined at docs/

  • superouter A Router that both exposes and internally uses sum-type to model route definitions, validation and more.

Project Goals and Motivations

  • Serializable
  • 0 Dependencies
  • Tiny for frontend usage
  • Avoid pitfalls found in other sum type libraries
  • Helpful Error Messages designed for makers

How does sum-type differ from other libraries in the ecosystem?

sum-type removes the following features because we believe they lead to brittle codebases.

  • placeholder cases
    • replaced by otherwise which is slightly more verbose but way safer and more powerful
  • auto spreading of values in cata/fold
    • sum-type just have 1 value, and that's what get's passed in to a fold
    • use a list, or an object to have an instance carry more values
  • auto curried constructors
    • Functions are manually curried and report errors per invocation
  • prototypes (reference equality checks / instanceof)
    • Makes it hard to share type checkings for data across serialization boundaries and realms (e.g. Electron apps)

sum-type is technically 0KB, it's an idea. You can use sum-type in your codebase without ever running npm install. But this library is only 4kb gzipped, so even the non-idea part is pretty small.


These docs are a bit stale as sum-type goes through some API churn.


import { either } from 'sum-type'

const Loaded = 


a -> Either Y a | N b


b -> Either Y a | N b


( a -> c ) -> Either Y c | N b


(( a -> c ), ( b -> d )) -> Either Y c | N d


(( a -> c ), ( b -> c )) -> Either Y a | N b -> c


( c , ( b -> c )) -> Either Y a | N b -> c


c -> Either Y a | N b -> c


Type -> { Y: a -> c, N: b -> c } -> tag -> c


( a -> Either Y c | N b ) -> Either Y c | N b


Either Y a | N -> boolean

You should almost always avoid coercing a sum type to a boolean. If you are checking for Y, try .map. If you are checking for N try getOr. Booleans have no context, no associated data, but there's almost always associated data in your model so toBoolean is much like moving from a lossless format to a lossy format.


( a -> b ) -> Either Y b | N Error

Takes a potentially unsafe function and decorates it to return an Either where non thrown values are encased in Either.Y and thrown values are encased in Either.N.


import { maybe } from 'sum-type'

const Selected = 


a -> Maybe Y a | N


() -> Maybe Y a | N


( a -> c ) -> Maybe Y c | N


(( () -> b ), ( a -> b )) -> Maybe Y b | N


(( () -> b ), ( a -> b )) -> Maybe Y a | N -> b


( b , ( a -> b )) -> Maybe Y a | N -> b


b -> Maybe Y a | N -> b


Type -> { Y: a -> b, N: () -> b } -> tag -> b


( a -> Maybe Y b | N ) -> Maybe Y b | N


Maybe Y a | N -> boolean

You should almost always avoid coercing a sum type to a boolean. If you are checking for Y, try .map. If you are checking for N try getOr. Booleans have no context, no associated data, but there's almost always associated data in your model so toBoolean is much like moving from a lossless format to a lossy format.


( a -> b ) -> Maybe Y b | N Error

Takes a potentially unsafe function and decorates it to return an Maybe where non thrown values are encased in Maybe.Y and thrown values are represented as Maybe.N(). If the specific error value is relevant try Either.encase instead as it will return your error object in an Either.N structure.

Canonical Either

In the future some functions will return optional values. This library encourages you to define your own but this library exports a pregenerated Either type that can be used canonically as the "real" Either which can be helpful when doing natural transformations and conversions between types and safe and unsafe data.

import { Y, N, getOr } from sum-type

const yes = Y(100)
const no = N()

const f = getOr(0)

// => 100

// => 0


import { tags } from 'sum-type'

const Geom = 
	tags ('Geom') (
		['Point' // : {x, y},
		,'Line' // [p1, p2],
		,'Poly' // [p1, p2, rest]

const p1 = Geom.Point({ x:0, y: 0 })

const p2 = p1

const line = Geom.Line([p1, p2])

const poly = Geom.Poly([p1, p2, [p3]])


Type -> { [tag]: a -> b } -> tag -> b


Type -> { [tag]: a -> b } -> tag -> Type b

Both map and chain will skip executing tags when their instance has no .value property (usually determined by their type constructor).


Type -> { [tag]: a -> Type b } -> tag -> Type b

Both map and chain will skip executing tags when their instance has no .value property (usually determined by their type constructor).

But when using map and chain you are still required to pass in a handler for every tag.

It's recommended to use otherwise with map and chain to prefill values that are not relevant to the fold.


string[] -> f -> { [key:string]: f }

A helper function for generating folds that are versioned separately to the type definition. It's useful when you want to avoid specifying each clause in a fold without losing type safety or introducing other modelling problems

Read more about otherwise here

const { Y, N } = T.Maybe
const Platform = T.tags ('Platform') (

const rest = T.otherwise([ // renamed

const windowsGUI = T.otherwise([

const foldWindowsGUI = f => T.mapAll(Platform) ({
	... rest2(N),
	... windowsGUI( () => Y(f()) )

const winPing =
		( () => 'ping \\t' )

winPing( Platform.Darwin() )
// => T.Maybe.N()

winPing( Platform.XP() )
// => T.Maybe.Y('ping \t')


These functions are likely to change at any moment.


tagName -> tagName:string

Extract the name of a tag from an instance of a type.


T -> (tag:string)[]

Returns a list of the tag's for a given type T.


Below is the source code definition for the internal errors this library throws.

const StaticSumTypeError =
	tags('StaticSumTypeError', [
		, 'ExtraTags' // {extraKeys}
		, 'MissingTags' // {missingKeys}
		, 'InstanceNull' // {T}
		, 'InstanceWrongType' // {T, x}
		, 'InstanceShapeInvalid' // {T, x}
		, 'tag' // {context}
		, 'VisitorNotAFunction' // {context, visitor}
		, 'NotAType' // {context, T}
Error Throws ...
ExtraTags when a fold specifies a visitor for tags that are not of the type.
MissingTags when a fold does not specify a visitor for each tag of the type.
InstanceNull when an argument was expected to be an instance of a sum type but was instead null.
InstanceWrongType when an instance is a valid sum-type but not the specifically expected type for that function.
InstanceShapeInvalid when an instance has the correct type property but an unknown tag property.
VisitorNotAFunction when a function was expected a visitor function but received anything else.
NotAType when a function expected a sum-type type but received anything else.

Package Sidebar


npm i sum-type

Weekly Downloads






Unpacked Size

304 kB

Total Files


Last publish


  • jaforbes