Runtypes

Safely bring untyped data into the fold

Runtypes allow you to take values about which you have no assurances and check that they conform to some type A. This is done by means of composable type validators of primitives, literals, arrays, tuples, records, unions, intersections and more.

Installation

npm install --save runtypes

Example

Suppose you have objects which represent asteroids, planets, ships and crew members. In TypeScript, you might write their types like so:

type Vector = [number, number, number];

type Asteroid = {
  type: 'asteroid';
  location: Vector;
  mass: number;
};

type Planet = {
  type: 'planet';
  location: Vector;
  mass: number;
  population: number;
  habitable: boolean;
};

type Rank = 'captain' | 'first mate' | 'officer' | 'ensign';

type CrewMember = {
  name: string;
  age: number;
  rank: Rank;
  home: Planet;
};

type Ship = {
  type: 'ship';
  location: Vector;
  mass: number;
  name: string;
  crew: CrewMember[];
};

type SpaceObject = Asteroid | Planet | Ship;

If the objects which are supposed to have these shapes are loaded from some external source, perhaps a JSON file, we need to validate that the objects conform to their specifications. We do so by building corresponding Runtypes in a very straightforward manner:

import { Boolean, Number, String, Literal, Array, Tuple, Record, Union } from 'runtypes';

const Vector = Tuple(Number, Number, Number);

const Asteroid = Record({
  type: Literal('asteroid'),
  location: Vector,
  mass: Number,
});

const Planet = Record({
  type: Literal('planet'),
  location: Vector,
  mass: Number,
  population: Number,
  habitable: Boolean,
});

const Rank = Union(
  Literal('captain'),
  Literal('first mate'),
  Literal('officer'),
  Literal('ensign'),
);

const CrewMember = Record({
  name: String,
  age: Number,
  rank: Rank,
  home: Planet,
});

const Ship = Record({
  type: Literal('ship'),
  location: Vector,
  mass: Number,
  name: String,
  crew: Array(CrewMember),
});

const SpaceObject = Union(Asteroid, Planet, Ship);

(See the examples directory for an expanded version of this.)

Now if we are given a putative SpaceObject we can validate it like so:

// spaceObject: SpaceObject
const spaceObject = SpaceObject.check(obj);

If the object doesn't conform to the type specification, check will throw an exception.

Static type inference

In TypeScript, the inferred type of Asteroid in the above example is

Runtype<{
  type: 'asteroid'
  location: [number, number, number]
  mass: number
}>

That is, it's a Runtype<Asteroid>, and you could annotate it as such. But we don't really have to define the Asteroid type in TypeScript at all now, because the inferred type is correct. Defining each of your types twice, once at the type level and then again at the value level, is a pain and not very DRY. Fortunately you can define a static Asteroid type which is an alias to the Runtype-derived type like so:

import { Static } from 'runtypes';

type Asteroid = Static<typeof Asteroid>;

which achieves the same result as

type Asteroid = {
  type: 'asteroid';
  location: [number, number, number];
  mass: number;
};

Type guards

In addition to providing a check method, runtypes can be used as type guards:

function disembark(obj: {}) {
  if (SpaceObject.guard(obj)) {
    // obj: SpaceObject
    if (obj.type === 'ship') {
      // obj: Ship
      obj.crew = [];
    }
  }
}

Pattern matching

The Union runtype offers the ability to do type-safe, exhaustive case analysis across its variants using the match method:

const isHabitable = SpaceObject.match(
  asteroid => false,
  planet => planet.habitable,
  ship => true,
);

if (isHabitable(spaceObject)) {
  // ...
}

There's also a top-level match function which allows testing an ad-hoc sequence of runtypes. You should use it along with when helper function to enable type inference of the parameters of the case functions:

const makeANumber = match(
  when(Number, n => n * 3),
  when(Boolean, b => (b ? 1 : 0)),
  when(String, s => s.length),
);

makeANumber(9); // = 27

To allow the function to be applied to anything and then handle match failures, simply use an Unknown case at the end:

const makeANumber = match(
  when(Number, n => n * 3),
  when(Boolean, b => (b ? 1 : 0)),
  when(String, s => s.length),
  when(Unknown, () => 42),
);

Constraint checking

Beyond mere type checking, we can add arbitrary runtime constraints to a Runtype:

const Positive = Number.withConstraint(n => n > 0);

Positive.check(-3); // Throws error: Failed constraint check

You can provide more descriptive error messages for failed constraints by returning a string instead of false:

const Positive = Number.withConstraint(n => n > 0 || `${n} is not positive`);

Positive.check(-3); // Throws error: -3 is not positive

You can set a custom name for your runtype, which will be used in default error messages and reflection, by using the name prop on the optional options parameter:

const C = Number.withConstraint(n => n > 0, { name: 'PositiveNumber' });

To change the type, there are two ways to do it: passing a type guard function to a new Runtype.withGuard() method, or using the familiar Runtype.withConstraint() method. (Both methods also accept an options parameter to optionally set the name.)

Using a type guard function is the easiest option to change the static type, because TS will infer the desired type from the return type of the guard function.

// use Buffer.isBuffer, which is typed as: isBuffer(obj: any): obj is Buffer;
const B = Unknown.withGuard(Buffer.isBuffer);
type T = Static<typeof B>; // T is Buffer

However, if you want to return a custom error message from your constraint function, you can't do this with a type guard because these functions can only return boolean values. Instead, you can roll your own constraint function and use the withConstraint<T>() method. Remember to specify the type parameter for the Constraint because it can't be inferred from your check function!

const check = (o: any) => Buffer.isBuffer(o) || 'Dude, not a Buffer!';
const B = Unknown.withConstraint<Buffer>(check);
type T = Static<typeof B>; // T will have type of `Buffer`

One important choice when changing Constraint static types is choosing the correct underlying type. The implementation of Constraint will validate the underlying type before running your constraint function. So it's important to use a lowest-common-denominator type that will pass validation for all expected inputs of your constraint function or type guard. If there's no obvious lowest-common-denominator type, you can always use Unknown as the underlying type, as shown in the Buffer examples above.

Speaking of base types, if you're using a type guard function and your base type is Unknown, then there's a convenience runtype Guard available, which is a shorthand for Unknown.withGuard.

// use Buffer.isBuffer, which is typed as: isBuffer(obj: any): obj is Buffer;
const B = Guard(Buffer.isBuffer);
type T = Static<typeof B>; // T will have type of `Buffer`

Template literals

The Template runtype validates that a value is a string that conforms to the template.

You can use the familiar syntax to create a Template runtype:

const T = Template`foo${Literal('bar')}baz`;

But then the type inference won't work:

type T = Static<typeof T>; // inferred as string

Because TS doesn't provide the exact string literal type information (["foo", "baz"] in this case) to the underlying function. See the issue microsoft/TypeScript#33304, especially this comment microsoft/TypeScript#33304 (comment) we hope to be implemented.

If you want the type inference rather than the tagged syntax, you have to manually write a function call:

const T = Template(['foo', 'baz'] as const, Literal('bar'));
type T = Static<typeof T>; // inferred as "foobarbaz"

As a convenient solution for this, it also supports another style of passing arguments:

const T = Template('foo', Literal('bar'), 'baz');
type T = Static<typeof T>; // inferred as "foobarbaz"

// Equivalent runtypes
Template(Literal('42'));
Template(42);
Template(Template('42'));
Template(4, '2');
Template(Literal(4), '2');
Template(String.withConstraint(s => s === '42'));
Template(
  Intersect(
    Number.withConstraint(n => n === 42),
    String.withConstraint(s => s.length === 2),
    // `Number`s in `Template` accept alternative representations like `"0x2A"`,
    // thus we have to constraint the length of string, to accept only `"42"`
  ),
);

Trivial items such as bare literals, Literals, and single-element Unions and Intersects are all coerced into strings at the creation time of the runtype. Additionally, Unions of such runtypes are converted into RegExp patterns like (?:foo|bar|...), so we can assume Union of Literals is a fully supported runtype in Template.

Caveats

A Template internally constructs a RegExp to parse strings. This can lead to a problem if it contains multiple non-literal runtypes:

const UpperCaseString = Constraint(String, s => s === s.toUpperCase(), {
  name: 'UpperCaseString',
});
const LowerCaseString = Constraint(String, s => s === s.toLowerCase(), {
  name: 'LowerCaseString',
});
Template(UpperCaseString, LowerCaseString);

The only thing we can do for parsing such strings correctly is brute-forcing every single possible combination until it fulfills all the constraints, which must be hardly done. Actually Template treats String runtypes as the simplest RegExp pattern .* and the “greedy” strategy is always used, that is, the above runtype won't work expectedly because the entire pattern is just ^(.*)(.*)$ and the first .* always wins. You have to avoid using Constraint this way, and instead manually parse it using a single Constraint which covers the entire string.

Function contracts

Runtypes along with constraint checking are a natural fit for enforcing function contracts. You can construct a contract from Runtypes for the parameters and return type of the function:

const divide = Contract(
  // Parameters:
  Number,
  Number.withConstraint(n => n !== 0 || 'division by zero'),
  // Return type:
  Number,
).enforce((n, m) => n / m);

divide(10, 2); // 5

divide(10, 0); // Throws error: division by zero

Branded types

Branded types is a way to emphasize the uniqueness of a type. This is useful until we have nominal types:

const Username = String.withBrand('Username');
const Password = String.withBrand('Password').withConstraint(
  str => str.length >= 8 || 'Too short password',
);

const signIn = Contract(Username, Password, Unknown).enforce((username, password) => {
  /*...*/
});

const username = Username.check('someone@example.com');
const password = Password.check('12345678');

// Static type OK, runtime OK
signIn(username, password);

// Static type ERROR, runtime OK
signIn(password, username);

// Static type ERROR, runtime OK
signIn('someone@example.com', '12345678');

Branded types are like opaque types and work as expected, except it is impossible to use as a key of an object type:

const StringBranded = String.withBrand('StringBranded');
type StringBranded = Static<typeof StringBranded>;
// Then the type `StringBranded` is computed as:
// string & { [RuntypeName]: "StringBranded" }

// TS1023: An index signature parameter type must be either `string` or `number`.
type SomeObject1 = { [K: StringBranded]: number };

// Both of these result in empty object type i.e. `{}`
type SomeObject2 = { [K in StringBranded]: number };
type SomeObject3 = Record<StringBranded, number>;

// You can do like this, but...
const key = StringBranded.check('key');
const SomeRecord = Record({ [key]: Number });
// This type results in { [x: string]: number }
type SomeRecord = Static<typeof SomeRecord>;

// So you have to use `Map` to achieve strongly-typed branded keys
type SomeMap = Map<StringBranded, number>;

Optional values

Runtypes can be used to represent a variable that may be undefined.

// For variables that might be `string | undefined`
Union(String, Undefined);
String.Or(Undefined); // shorthand syntax for the above
Optional(String); // equivalent to the above two basically
String.optional(); // shorthand syntax for the above

The last syntax is not any shorter than writing Optional(String) when you import Optional directly from runtypes, but if you use scoped import i.e. import * as rt from 'runtypes', it would look better to write rt.String.optional() rather than rt.Optional(rt.String).

If a Record may or may not have some properties, we can declare the optional properties using Record({ x: Optional(String) }) (or formerly Partial({ x: String })). Optional properties validate successfully if they are absent or undefined or the type specified.

// Using `Ship` from above
const RegisteredShip = Ship.And(
  Record({
    // All registered ships must have this flag
    isRegistered: Literal(true),

    // We may or may not know the ship's classification
    shipClass: Optional(Union(Literal('military'), Literal('civilian'))),

    // We may not know the ship's rank (so we allow it to be undefined via `Optional`),
    // we may also know that a civilian ship doesn't have a rank (e.g. null)
    rank: Optional(Rank.Or(Null)),
  }),
);

There's a difference between Union(String, Undefined) and Optional(String) iff they are used within a Record; the former means "it must be present, and must be string or undefined", while the latter means "it can be present or missing, but must be string or undefined if present".

Prior to v5.2, Union(..., Undefined) in a Record was passing even if the property was missing. Although some users considered this behavior was a bug especially for the sake of mirroring TS behavior, it was a long-standing thing, and some other users have been surprised with this fix. So the v5.2 release has been marked deprecated on npm, due to the breaking change.

Note that null is a quite different thing than undefined in JS and TS, so Optional doesn't take care of it. If your Record has properties which can be null, then use the Null runtype explicitly.

const MilitaryShip = Ship.And(
  Record({
    shipClass: Literal('military'),

    // Must NOT be undefined, but can be null
    lastDeployedTimestamp: Number.Or(Null),
  }),
);

You can save an import by using nullable shorthand instead. All three below are equivalent things.

Union(Number, Null);
Number.Or(Null);
Number.nullable();

Readonly records and arrays

Array and Record runtypes have a special function .asReadonly(), that creates a new runtype where the values are readonly.

For example:

const Asteroid = Record({
  type: Literal('asteroid'),
  location: Vector,
  mass: Number,
}).asReadonly();
type Asteroid = Static<typeof Asteroid>;
// { readonly type: 'asteroid', readonly location: Vector, readonly mass: number }

const AsteroidArray = Array(Asteroid).asReadonly();
type AsteroidArray = Static<typeof AsteroidArray>;
// ReadonlyArray<Asteroid>

Helper functions for `Record`

Record runtype has the methods .pick() and .omit(), which will return a new Record with or without specified fields (see Example section for detailed definition of Rank and Planet):

const CrewMember = Record({
  name: String,
  age: Number,
  rank: Rank,
  home: Planet,
});

const Visitor = CrewMember.pick('name', 'home');
type Visitor = Static<typeof Visitor>; // { name: string; home: Planet; }

const Background = CrewMember.omit('name');
type Background = Static<typeof Background>; // { age: number; rank: Rank; home: Planet; }

Also you can use .extend() to get a new Record with extended fields:

const PetMember = CrewMember.extend({
  species: String,
});
type PetMember = Static<typeof PetMember>;
// { name: string; age: number; rank: Rank; home: Planet; species: string; }

It is capable of reporting compile-time errors if any field is not assignable to the base runtype. You can suppress this error by using @ts-ignore directive or .omit() before, and then you'll get an incompatible version from the base Record.

const WrongMember = CrewMember.extend({
  rank: Literal('wrong'),
  // Type '"wrong"' is not assignable to type '"captain" | "first mate" | "officer" | "ensign"'.
});

Related libraries

generate-runtypes Generates runtypes from structured data. Useful for code generators
json-to-runtypes Generates runtypes by parsing example JSON data
rest.ts Allows building type safe and runtime-checked APIs
runtypes-generate Generates random data by Runtype for property-based testing
runtyping Generate runtypes from static types & JSON schema
schemart Generate runtypes from your database schema.

patrickocoffeyo/runtypes