typed-validators

Complex type validators that generate TypeScript or Flow types for you. The validation errors are detailed. Adapted from the brilliant work in flow-runtime.

Table of Contents

• typed-validators
• Table of Contents
• Introduction
• Limitations
• Generating validators from type defs
  • Before
  • Command
  • After
• API
  • Type creators
    • t.any()
    • t.unknown()
    • t.boolean()
    • t.boolean(true)
    • t.string()
    • t.string('foo')
    • t.number()
    • t.number(3)
    • t.symbol()
    • t.symbol(MySymbol)
    • t.null() / t.nullLiteral()
    • t.nullOr(t.string())
    • t.undefined() / t.undefinedLiteral()
    • t.nullish()
    • t.nullishOr(t.string())
    • t.array(t.number())
    • t.readonlyArray(t.number())
    • t.object(properties)
    • t.object({ required?, optional?, exact? })
    • t.opaque<DateString>(() => t.string())
    • t.readonly(objectType)
    • t.merge(...objectTypes)
    • t.mergeInexact(...objectTypes)
    • t.record(t.string(), t.number())
    • t.instanceOf(() => Date)
    • t.tuple(t.string(), t.number())
    • t.allOf(A, B)
    • t.oneOf(t.string(), t.number())
    • t.alias(name, type)
    • t.ref(() => typeAlias)
  • t.Type<T>
    • accepts(input: any): boolean
    • acceptsSomeCompositeTypes: boolean (getter)
    • assert<V extends T>(input: any, prefix = '', path?: (string | number | symbol)[]): V
    • validate(input: any, prefix = '', path?: (string | number | symbol)[]): Validation<T>
    • warn(input: any, prefix = '', path?: (string | number | symbol)[]): void
    • toString(): string
  • t.ExtractType<T extends Type<any>>
  • t.TypeAlias<T>
    • readonly name: string
    • addConstraint(...constraints: TypeConstraint<T>[]): this
  • Custom Constraints
  • Recursive Types

Introduction

When you need to validate the inputs to a TypeScript or Flow API, a problem arises. How do you ensure that a value that passes validation matches your declared TypeScript type? Someone might modify one and forget to modify the other:

1type Post = {
2  author: {
3    name: string
4    username: string
5  }
6  content: string
7  // newly added by developer
8  tags: string[]
9}
10
11// hypothetical syntax
12const validator = requireObject({
13  author: requireObject({
14    name: requireString(),
15    username: requireString(),
16  }),
17  content: requireString(),
18  // uhoh!! developer forgot to add tags here
19})

typed-validators solves this by generating TypeScript or Flow types from your validators:

1import * as t from 'typed-validators'
2
3const PostValidator = t.object({
4  author: t.object({
5    name: t.string(),
6    username: t.string(),
7  }),
8  content: t.string(),
9  tags: t.array(t.string()),
10})
11
12type Post = t.ExtractType<typeof PostValidator>
13
14const example: Post = PostValidator.assert({
15  author: {
16    name: 'MC Hammer',
17    username: 'hammertime',
18  },
19  content: "Can't touch this",
20  tags: ['mc-hammer', 'hammertime'],
21})

Hover over Post in VSCode and you'll see, voilà:

1type Post = {
2  author: {
3    name: string
4    username: string
5  }
6  content: string
7  tags: string[]
8}

Example error message:

1PostValidator.assert({
2  author: {
3    name: 'MC Hammer',
4    usernme: 'hammertime',
5  },
6  content: 1,
7  tags: ['mc-hammer', { tag: 'hammertime' }],
8})

RuntimeTypeError: input.author is missing required property username, which must be a string

Actual Value: {
  name: "MC Hammer",
  usernme: "hammertime",
}

-------------------------------------------------

input.author has unknown property: usernme

Actual Value: {
  name: "MC Hammer",
  usernme: "hammertime",
}

-------------------------------------------------

input.content must be a string

Actual Value: 1

-------------------------------------------------

input.tags[1] must be a string

Actual Value: {
  tag: "hammertime",
}

Limitations

• Flow seems to suck at fully resolving t.ExtractType<...> for deeply nested object types. Past a certain level of complexity it seems to give up and use any for some object-valued properties. That's why I created gen-typed-validators, so that you can control the type definitions and generate typed-validators from them.
• Generic types aren't supported. I may add support for it in the future if I'm confident I can make a robust implementation.
• Function types aren't supported. You can use t.instanceOf(() => Function), but Flow treats the Function type as any. I may add t.function() in the future, but it won't validate argument or return types, because those can't be determined from function instances at runtime.
• The goal is to support a subset of types common to TS and Flow well, rather than support every possible complex derived type you can make. (That's what babel-plugin-flow-runtime basically tried to do, and it was too ambitious. I created this so that I could stop using it.)

Generating validators from type defs

This is now possible with gen-typed-validators!

It creates or replaces validators anywhere you declare a variable of type t.TypeAlias:

Before

1// Post.ts
2import * as t from 'typed-validators'
3
4type Author = {
5  name: string
6  username: string
7}
8
9export type Post = {
10  author: Author
11  content: string
12  tags: string[]
13}
14
15export const PostType: t.TypeAlias<Post> = null

Command

1$ gen-typed-validators Post.ts

After

1// Post.ts
2import * as t from 'typed-validators'
3
4export type Author = {
5  name: string
6  username: string
7}
8
9const AuthorType: t.TypeAlias<Author> = t.alias(
10  'Author',
11  t.object({
12    name: t.string(),
13    username: t.string(),
14  })
15)
16
17export type Post = {
18  author: Author
19  content: string
20  tags: string[]
21}
22
23export const PostType: t.TypeAlias<Post> = t.alias(
24  'Post',
25  t.object({
26    author: t.ref(() => AuthorType),
27    content: t.string(),
28    tags: t.array(t.string()),
29  })
30)

API

I recommend importing like this:

1import * as t from 'typed-validators'

Type creators

All of the following methods return an instance of t.Type<T>.

t.any()

A validator that accepts any value.

t.unknown()

A validator that accepts any value but has TS unknown type/Flow mixed type.

t.boolean()

A validator that requires the value to be a boolean.

t.boolean(true)

A validator that requires the value to be true.

Note: to get the proper Flow types, you'll unforunately have to do t.boolean<true>(true).

t.string()

A validator that requires the value to be a string.

t.string('foo')

A validator that requires the value to be 'foo'.

Note: to get the proper Flow types, you'll unfortunately have to do t.string<'foo'>('foo').

t.number()

A validator that requires the value to be a number.

t.number(3)

A validator that requires the value to be 3.

Note: to get the proper Flow types, you'll unfortunately have to do t.number<3>(3).

t.symbol()

A validator that requires the value to be a symbol.

t.symbol(MySymbol)

A validator that requires the value to be MySymbol.

t.null() / t.nullLiteral()

A validator that requires the value to be null.

t.nullOr(t.string())

A validator that requires the value to be string | null

t.undefined() / t.undefinedLiteral()

A validator that requires the value to be undefined.

t.nullish()

A validator that requires the value to be null | undefined.

t.nullishOr(t.string())

A validator that requires the value to be string | null | undefined.

t.array(t.number())

A validator that requires the value to be number[].

t.readonlyArray(t.number())

A validator that requires the value to be number[]. Doesn't require the value to be frozen; just allows the extracted type to be ReadonlyArray.

t.object(properties)

A validator that requires the value to be an object with all of the given required properties an no additional properties.

For example:

1const PersonType = t.object({
2  name: t.string(),
3  age: t.number(),
4})
5
6PersonType.assert({ name: 'dude', age: 100 }) // ok
7PersonType.assert({ name: 'dude' }) // error
8PersonType.assert({ name: 1, age: 100 }) // error
9PersonType.assert({ name: 'dude', age: 100, powerLevel: 9000 }) // error

t.object({ required?, optional?, exact? })

A validator that requires the value to be an object with given properties. Additional properties won't be allowed unless exact is false.

For example:

1const PersonType = t.object({
2  required: {
3    name: t.string(),
4  },
5  optional: {
6    age: t.number(),
7  },
8})
9
10PersonType.assert({ name: 'dude' }) // ok
11PersonType.assert({ name: 'dude', age: 100 }) // ok
12PersonType.assert({ name: 1 }) // error
13PersonType.assert({ name: 'dude', age: 'old' }) // error

t.opaque<DateString>(() => t.string())

A validator that requires the value to be a string, but presents the type as DateString (for instance with export opaque type DateString = string)

t.readonly(objectType)

Use t.readonly(t.object(...)) or t.readonly(t.merge(...)) etc. Doesn't require the object to be frozen, just allows the extracted type to be readonly.

t.merge(...objectTypes)

Merges the properties of multiple object validators together into an exact object validator (no additional properties are allowed).

Note: merging t.aliases and t.refs that resolve to object validators is supported, but any constraints on the referenced aliases won't be applied.

For example:

1const PersonType = t.object({
2  required: {
3    name: t.string(),
4  },
5  optional: {
6    age: t.number(),
7  },
8})
9const AddressType = t.object({
10  street: t.string(),
11  city: t.string(),
12  state: t.string(),
13  zip: t.string(),
14})
15
16const PersonWithAddressType = t.merge(PersonType, AddressType)
17
18PersonWithAddressType.assert({
19  // ok
20  name: 'dude',
21  age: 100,
22  street: 'Bourbon Street',
23  city: 'New Orleans',
24  zip: '77777',
25})

t.mergeInexact(...objectTypes)

Merges the properties of multiple object validators together into an inexact object validator (additional properties are allowed).

Note: merging t.aliases and t.refs that resolve to object validators is supported, but any constraints on the referenced aliases won't be applied.

Accepts a variable number of arguments, though type generation is only overloaded up to 8 arguments. Accepts a variable number of arguments, though type generation is only overloaded up to 8 arguments.

t.record(t.string(), t.number())

A validator that requires the value to be Record<string, number>.

t.instanceOf(() => Date)

A validator that requires the value to be an instance of Date.

t.tuple(t.string(), t.number())

A validator that requires the value to be [string, number]. Accepts a variable number of arguments, though type generation for Flow is only overloaded up to 8 arguments.

t.allOf(A, B)

A validator that requires the value to be A & B. Accepts a variable number of arguments, though type generation is only overloaded up to 8 arguments. For example:

1const ThingType = t.object({ name: t.string() })
2const CommentedType = t.object({ comment: t.string() })
3
4const CommentedThingType = t.allOf(ThingType, CommentedType)
5
6CommentedThingType.assert({ name: 'foo', comment: 'sweet' })

t.oneOf(t.string(), t.number())

A validator that requires the value to be string | number. Accepts a variable number of arguments, though type generation is only overloaded up to 32 arguments.

t.alias(name, type)

Creates a TypeAlias with the given name and type.

Type aliases serve two purposes:

• They allow you to create recursive type validators with t.ref()
• You can add custom constraints to them

t.ref(() => typeAlias)

Creates a reference to the given TypeAlias. See Recursive Types for examples.

t.Type<T>

The base class for all validator types.

T is the type of values it accepts.

accepts(input: any): boolean

Returns true if and only if input is the correct type.

acceptsSomeCompositeTypes: boolean (getter)

Returns true if the validator accepts some values that are not primitives, null or undefined.

assert<V extends T>(input: any, prefix = '', path?: (string | number | symbol)[]): V

Throws an error if input isn't the correct type.

prefix will be prepended to thrown error messages.

path will be prepended to validation error paths. If you are validating a function parameter named foo, pass ['foo'] for path to get clear error messages.

validate(input: any, prefix = '', path?: (string | number | symbol)[]): Validation<T>

Validates input, returning any errors in the Validation.

prefix and path are the same as in assert.

warn(input: any, prefix = '', path?: (string | number | symbol)[]): void

Logs a warning to the console if input isn't the correct type.

toString(): string

Returns a string representation of this type (using TS type syntax in most cases).

t.ExtractType<T extends Type<any>>

Gets the TypeScript type that a validator type accepts. For example:

1import * as t from 'typed-validators'
2
3const PostValidator = t.object({
4  author: t.object({
5    name: t.string(),
6    username: t.string(),
7  }),
8  content: t.string(),
9  tags: t.array(t.string()),
10})
11
12type Post = t.ExtractType<typeof PostValidator>

Hover over Post in the IDE and you'll see, voilà:

1type Post = {
2  author: {
3    name: string
4    username: string
5  }
6  content: string
7  tags: string[]
8}

t.TypeAlias<T>

readonly name: string

The name of the alias.

addConstraint(...constraints: TypeConstraint<T>[]): this

Adds custom constraints. TypeConstraint<T> is a function (value: T) => string | null | undefined which returns nullish if value is valid, or otherwise a string describing why value is invalid.

Custom Constraints

It's nice to be able to validate that something is a number, but what if we want to make sure it's positive? We can do this by creating a type alias for number and adding a custom constraint to it:

1const PositiveNumberType = t
2  .alias('PositiveNumber', t.number())
3  .addConstraint((value: number) => (value > 0 ? undefined : 'must be > 0'))
4
5PositiveNumberType.assert(-1)

The assertion will throw a t.RuntimeTypeError with the following message:

input must be > 0

Actual Value: -1

Recursive Types

Creating validators for recursive types takes a bit of extra effort. Naively, we would want to do this:

1const NodeType = t.object({
2  required: {
3    value: t.any(),
4  },
5  optional: {
6    left: NodeType,
7    right: NodeType,
8  },
9})

But left: NodeTYpe causes the error Block-scoped variable 'NodeType' referenced before its declaration.

To work around this, we can create a TypeAlias and a reference to it:

1const NodeType: t.TypeAlias<{
2  value: any
3  left?: Node
4  right?: Node
5}> = t.alias(
6  'Node',
7  t.object({
8    required: {
9      value: t.any(),
10    },
11    optional: {
12      left: t.ref(() => NodeType),
13      right: t.ref(() => NodeType),
14    },
15  })
16)
17
18type Node = t.ExtractType<typeof NodeType>
19
20NodeType.assert({
21  value: 'foo',
22  left: {
23    value: 2,
24    right: {
25      value: 3,
26    },
27  },
28  right: {
29    value: 6,
30  },
31})

Notice how we use a thunk function in t.ref(() => NodeType) to avoid referencing NodeType before its declaration.