RunType is a runtime type system for TypeScript.
It was inspired by IO-TS, but I made some opinionated changes in the concept. IO-TS is mathematically correct and follows JavaScript ant TypeScript specifications to the letter. With RunType I wanted to create something more practical.
Some of the changes:
- I am not too familiar with functional programming concepts, so I don't use them in RunType.
- The struct combinator handles optional fields easier (without the partial + intersection things in IO-TS)
- number decoder does not accept NaN.
- Decoder accepts a config argument and supports type coercion and some other modifiers
- Validators
- Runtime type description generation (print() method)
npm install @symbion/runtype
This is a work in progress. I plan to write some documentation soon, in the meantime you can look at the test files (src/*.spec.ts) for usage information.
First create a type:
import T from '@symbion/runtype'
const tMyType = T.struct({
s: T.string,
n: T.optional(T.number)
})
You can extract a TypeScript type from it:
type MyType = T.TypeOf<typeof tMyType>
You can decode an unknown value:
const u: unknown = { s: 'string', n: 42 }
const decoded = T.decode(tMyType, u)
isOk(decoded)
// = true
const value: MyType = decoded.ok
// = { s: 'string', n: 42 }
Type | TypeScript | RunType |
---|---|---|
undefined | undefined | T.undefinedValue |
null | null | T.nullValue |
true | true | T.trueValue |
false | false | T.falseValue |
string | string | T.string |
number | number | T.number |
integer | X | T.integer |
boolean | boolean | T.boolean |
date | date | T.date |
any | any | T.any |
unknown | unknown | T.unknown |
literal | 'a' | 'b' | 3 |
T.literal('a', 'b', 3) |
optional | Type | undefined |
T.optional(tType) |
nullable | Type | null | undefined |
T.nullable(tType) |
array | Array<Type> |
T.array(tType) |
record | record<string, Type> |
T.record(tType) |
struct | { s: string, n: number } |
T.struct({ s: T.string, n: T.number }) |
key of | keyof { s: string, n: number } |
T.keyof(T.struct({ s: T.string, n: T.number })) |
tuple | [string, number, Type] |
T.tuple(T.string, T.number, tType) |
union | string | number | Type |
T.union(T.string, T.number, tType) |
intersect | boolean | true |
T.union(T.boolean, T.trueValue) |
intersect | { s: string } & { n: number } |
T.intersect(T.struct({ s: T.string }), T.struct({ n: T.number })) |
tagged union | { tag: 's', s: string } | { tag: 'n', n: number } |
T.taggedUnion('tag')({ tag: T.literal('s'), s: T.string }, { tag: T.literal('n'), n: T.number }) |
Recursive types can be created with T.lazy() and manual TypeScript types (because TypeScript can't infer recursive types):
interface Recursive {
name: string
children: MyType[]
}
const tRecursive: T.Type<Recursive> = T.lazy(() => T.struct({
name: T.string,
children: T.array(tRecursive)
}))
The T.partial() type modifier takes a Struct type and converts all fields to optional:
const tStruct = T.struct({
s: T.string,
n: T.optional(T.number)
})
// = { s: string, n?: number }
const tPartialType = T.partial(tStruct)
// = { s?: string, n?: number }
The T.patch() type modifier takes a Struct type and converts all optional fields to nullable and all requires fields to optional. It is useful for update APIs, where undefined or missing fields mean not to update and null value means to clear that field.
const tStruct = T.struct({
s: T.string,
n: T.optional(T.number)
})
// = { s: string, n?: number }
const tPatchType = T.patch(tStruct)
// = { s?: string, n?: number | null }
The T.pick() type modifier takes a Struct type and picks the specified fields.
const tStruct = T.struct({
s: T.string,
n: T.optional(T.number),
b: T.boolean
})
// = { s: string, n?: number, b: boolean }
const tPickType = T.pick(tStruct, ['s', 'n'])
// = { s: string, n?: number }
The T.omit() type modifier takes a Struct type and omits the specified fields.
const tStruct = T.struct({
s: T.string,
n: T.optional(T.number),
b: T.boolean
})
// = { s: string, n?: number, b: boolean }
const tOmitType = T.omit(tStruct, ['b'])
// = { s: string, n?: number }
The decoder() function accepts an optional config argument. It can be used for type coercion:
T.decode(T.number, '42')
// = { _tag: 'Err', err: [ { path: [], error: 'expected number' } ] }
T.decode(T.number, '42', { coerceStringToNumber: true })
// = isOk(decoded)
All available coercion options:
Option name | Type | |
---|---|---|
coerceNumberToString | boolean | Coerce numbers to string |
coerceNumberToBoolean | boolean | Coerce numbers to boolean |
coerceStringToNumber | boolean | Coerce string to number |
coerceScalar | boolean | = coerceNumberToString, coerceNumberToBoolean, coerceStringToNumber |
coerceStringToDate | boolean | Coerce string to Date |
coerceNumberToDate | boolean | Coerce numbers to Date |
coerceDate | boolean | = coerceStringToDate, coerceNumberToDate |
coerceAll | boolean | All the above coercion |
acceptNaN | boolean | Make T.number accept NaN as number |
unknownFields | 'reject' | 'drop' | 'discard' |
How to treat unknown fields. (reject: error, drop: drops unknown fields from the output, discard: leaves in output as is) |
The decode() function does type decoding, which is a synchron function. Runtype also handles data validation, what is defined as an asynchron function. The type constructors define some validator methods and user defined functions can also be used.
const tMyType = struct({
s: T.string.minLength(2)
})
Validation works like decoding:
T.validate(T.string.minLength(2), 'abc')
// = { _tag: 'Ok', ok: 'abc' }
T.decode(T.string.minLength(2), 'a')
// = { _tag: 'Ok', ok: 'a' }
T.validate(T.string.minLength(2), 'a')
// = { _tag: 'Err', err: [ { path: [], error: 'length must be at least 2' } ] }
Awailable validators:
Validator | Description |
---|---|
in(value1, value2, ...) | Value is one of value1, value2, .... |
minLength(length) | Length is at least length |
maxLength(length) | Length is at most length |
matches(pattern) | Value matches pattern |
email() | Value is an email address |
Validator | Description |
---|---|
in(value1, value2, ...) | Value is one of value1, value2, .... |
integer() | Value is an integer |
min(minValue) | Value is at least minValue |
max(maxValue) | Value is at most maxValue |
between(minValue, maxValue) | Value is between minValue and maxValue |
Validator | Description |
---|---|
true() | Value is true |
false() | Value is false |
Validator | Description |
---|---|
in(value1, value2, ...) | Value is one of value1, value2, .... |
function max42(v: number | undefined) {
return (v || 0) <= 42 ? T.ok(v) : T.error("Max 42 is allowed!")
}
await T.validate(T.number.addValidator(max42), 43)
// = { _tag: 'Err', err: [ { path: [], error: "Max 42 is allowed!" } ] }
TypeScript (because of JavaScript) differentiates missing properties and properties with undefined value. This is sometimes useful, however it makes it more difficult to handle this in runtime type systems. Take the following simple TypeScript type:
interface Person {
name: string
age?: number
}
In IO-TS you can create it like this:
const tPerson = T.intersection([
T.type({
name: T.string
}),
T.partial({
age: T.number
})
type Person = T.TypeOf<typeof tPerson>
])
RunType uses complex TypeScript mechanisms to achieve a simpler and readable syntax:
const tPerson = T.struct({
name: T.string
age: T.optional(T.number)
])
type Person = T.TypeOf<typeof tPerson>
Under the hood RunType generates the same intersection type beacause of limitations in TypeScipt, but it works the same as the original type:
type Person = { name: string } & { age?: number }
If you want to boost your TypeScript knowledge to the next level I highly recommend to write a runtime type system. I guarantee it will be fun! :)