tsum
Typescript sum types with pattern matching, using multimethods
Install
npm install --save tsum
yarn add tsum
Motivation
Being able to simply represent, manipulate, and pattern match on immutable values is one of the most important things a programming language can provide.
Typescript offers two built-in solutions:
- Add a 'type' field to all your data, and switch on it. This is verbose, requires explicit serialize/deserialize, and error prone when specifying literal data. Additionally, switch offers no exhaustiveness guarantee, and is a statement rather than an expression.
- Put your data into classes. This is verbose, requires explicit serialize/deserialize and lacks named fields:
new Person('Garrett')
vs{name: 'Garrett'}
Most importantly, you lose the ability to easily make new immutable values from previous values.
Using plain data structures solves many of these problems, and improves notation:
- Values are trivial to persist or send over a wire, and trivial to read when receiving data from the wire or disk
- Data definitions are as terse and straightforward as possible in Typescript
- Fields are named
- Open system, as long as a piece of data matches one of these interfaces, it can participate in this polymorphism (unlike classes)
- Matching every possible case is enforced by the compiler, and matches are expressions which return a value
- Critically, you gain the ability to use spread notation to flexibly produce new values:
const olderPerson = {...person, age: person.age + 1}
Classes offer no equivalent to this
Example
import Sum from 'tsum'
Describe each of your types:
interface Dog {
name: string
color: string
}
interface Cat {
name: string
dogFriendly: boolean
}
interface Chicken {
eggsLaid: number
}
interface Cow {
mooVolume: number
}
Put them together:
interface Animals {
Dog: Dog
Cat: Cat
Chicken: Chicken
Cow: Cow
}
Provide the Animals interface, along with a set of predicates to determine the type of an arbitrary Animal:
const Animal = Sum<Animals>({
predicates: ({isString, isBoolean, isNumber}) => ({
Dog: _ => isString(_.color),
Cat: _ => isBoolean(_.dogFriendly),
Chicken: _ => isNumber(_.eggsLaid),
Cow: _ => isNumber(_.mooVolume)
})
})
Create functions that act on any Animal:
// The type of describe is: (animal: Animal) => string
const describe = Animal.f({
// Typescript knows the type in each branch, and makes sure you cover all cases
Dog: dog => `a ${dog.color} dog named ${dog.name}`,
Cat: cat => cat.dogFriendly ? `a cat named ${cat.name} (dog-friendly)` : `a cat named ${cat.name} (not dog-friendly)`,
Chicken: chicken => `a chicken who has laid ${chicken.eggsLaid} eggs`,
Cow: cow => cow.mooVolume > 5 ? 'a loud cow' : 'a quiet cow'
})
const dog = {name: 'Zorro', color: 'grey'}
const cat = {name: 'Oberyn', dogFriendly: false}
const chicken = {eggsLaid: 13}
const cow = {mooVolume: 6}
console.log(describe(dog)) // a grey dog named Zorro
console.log(describe(cat)) // a cat named Oberyn (not dog-friendly)
console.log(describe(chicken)) // a chicken who has laid 13 eggs
console.log(describe(cow)) // a loud cow
Or match inline:
const cat = {name: 'Oberyn', dogFriendly: false}
const cost = Animal.match(cat, {
Dog: dog => dog.name === 'Zorro' ? 99999 : 300,
Cat: cat => cat.name === 'Oberyn' ? 99999 : 800,
Chicken: chicken => chicken.eggsLaid * 10,
Cow: cow => cow.mooVolume > 5 ? 5000 : 6000
})
console.log(cost) // 99999
Convenience to avoid repeating yourself:
// Evaluates to Dog | Cat | Chicken | Cow
type Animal = typeof Animal.types