Configuration

A great out-of-the-box experience is a core goal of ArkType, including safe defaults and helpful messages for complex errors.

However, it's equally important that when you need different behavior, you can easily configure it with the right granularity.

import { configure } from "arktype/config"
// use the "arktype/config" entrypoint
configure({ numberAllowsNaN: true })

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

type.number.allows(Number.NaN) // true

const myScope = scope(
	{ user: { age: "number < 100" } },
	{
		max: {
			actual: () => "unacceptably large"
		}
	}
)
const types = myScope.export()
// ArkErrors: age must be less than 100 (was unacceptably large)
types.user({ name: "Alice", age: 101 })
const parsedAfter = myScope.type({
	age: "number <= 100"
})
// ArkErrors: age must be at most 100 (was unacceptably large)
parsedAfter({ age: 101 })

// avoid logging "was xxx" for password
const password = type("string >= 8").configure({ actual: () => "" })
const user = type({
	email: "string.email",
	password
})
// ArkErrors: password must be at least length 8
const out = user({
	email: "david@arktype.io",
	password: "ez123"
})

Some options only apply at specific levels, as reflected in the corresponding input types.

Errors

To allow custom errors to be integrated seemlessly with builtin logic for composite errors (i.e. union and intersection), ArkType supports a set of composable options:

const password = type.string.atLeastLength(8).describe("a valid password")
// ArkErrors: must be a valid password
const out = password("ez123")

const password = type.string.atLeastLength(8).configure({
	expected: ctx =>
		ctx.code === "minLength" ? `${ctx.rule} characters or better` : "way better"
})
// ArkErrors: must be 8 characters or better (was 5)
const out1 = password("ez123").toString()
// ArkErrors: must be way better (was a number)
const out2 = password(12345678).toString()

const password = type("string >= 8").configure({ actual: () => "" })
// ArkErrors: password must be at least length 8
const out = password("ez123")

const password = type("string >= 8").configure({
	problem: ctx => `${ctx.actual} isn't ${ctx.expected}`
})
// ArkErrors: 5 isn't at least length 8
const out1 = password("ez123")
// ArkErrors: a number isn't a string
const out2 = password(12345678)

const user = type({
	password: "string >= 8"
}).configure({
	message: ctx =>
		`${ctx.propString || "(root)"}: ${ctx.actual} isn't ${ctx.expected}`
})
// ArkErrors: (root): a string isn't an object
const out1 = user("ez123")
// `.configure` only applies shallowly, so the nested error isn't changed!
// ArkErrors: password must be at least length 8 (was 5)
const out2 = user({ password: "ez123" })

By Code

Errors can also be configured by their associated code property at a scope or global level.

For example:

const mod = type.module(
	{ isEven: "number%2" },
	{
		divisor: {
			// the available `ctx` types will include data specific to your errors
			expected: ctx => `% ${ctx.rule} !== 0`,
			problem: ctx => `${ctx.actual} ${ctx.expected}`
		}
	}
)
// ArkErrors: 3 % 2 !== 0
mod.isEven(3)

ArkErrors

For use cases like i18n that fall outside the scope of this composable message config, the ArkErrors array returned on validation failure contains ArkError instances that can be discriminated via calls like .hasCode("divisor") and contain contextual data specific to that error type as well as getters for each composable error part.

These ArkError instances can be arbitrarily transformed and composed with an internationalization library. This is still a topic we're working on investigating and documenting, so please reach out with any questions or feedback!

Keywords

Builtin keywords like string.email can be globally configured.

This can be very helpful for customizing error messages without needing to create your own aliases or wrappers.

import { configure } from "arktype/config"

configure({
	keywords: {
		string: "shorthand description",
		"string.email": {
			actual: () => "definitely fake"
		}
	}
})

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

const user = type({
	name: "string",
	email: "string.email"
})

const out = user({
	// ArkErrors: name must be shorthand description (was a number)
	name: 5,
	// ArkErrors: email must be an email address (was definitely fake)
	email: "449 Canal St"
})

The options you can provide here are identical to those used to configure a Type directly, and can also be extended at a type-level to include custom metadata.

Clone

By default, before a morph is applied, ArkType will deeply clone the original input value with a builtin deepClone function that tries to make reasonable assumptions about preserving prototypes etc. The implementation of deepClone can be found here.

You can provide an alternate clone implementation to the clone config option.

import { configure } from "arktype/config"

configure({ clone: structuredClone })

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

// will now create a new object using structuredClone
const userForm = type({
	age: "string.numeric.parse"
})

To mutate the input object directly, you can set the clone config option to false.

import { configure } from "arktype/config"

configure({ clone: false })

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

const userForm = type({
	age: "string.numeric.parse"
})

const formData = {
	age: "42"
}

const out = userForm(formData)

// the original object's age key is now a number
console.log(formData.age)

onUndeclaredKey

Like TypeScript, ArkType defaults to ignoring undeclared keys during validation. However, it also supports two additional behaviors:

• "ignore" (default): Allow undeclared keys on input, preserve them on output
• "delete": Allow undeclared keys on input, delete them before returning output
• "reject": Reject input with undeclared keys

These behaviors can be associated with individual Types via the builtin "+" syntax (see those docs for more on how they work). You can also change the default globally:

import { configure } from "arktype/config"

configure({ onUndeclaredKey: "delete" })

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

const userForm = type({
	name: "string"
})

// out is now { name: "Alice" }
const out = userForm({
	name: "Alice",
	age: "42"
})

jitless

By default, when a Type is instantiated, ArkType will precompile optimized validation logic that will run when the type is invoked. This behavior is disabled by default in environments that don't support new Function, e.g. Cloudflare Workers.

If you'd like to opt out of it for another reason, you can set the jitless config option to true.

import { configure } from "arktype/config"

configure({ jitless: true })

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

// will not be precompiled
const myObject = type({
	foo: "string"
})

onFail

In some domains, you may always want to throw on failed validation or transform the result in some other way.

By specifying onFail in your global config, you can control what happens when you invoke a Type on invalid data:

import { configure } from "arktype/config"

const config = configure({
	onFail: errors => errors.throw()
})

// be sure to specify both the runtime and static configs

declare global {
	interface ArkEnv {
		onFail: typeof config.onFail
	}
}

import "./config.ts"
// import your config file before arktype
import { type } from "arktype"

// data is inferred as string- no need to discriminate!
const data = type.string("foo")

// now thrown instead of returned
// ArkErrors: must be a string (was number)
const bad = type.string(5)

metadata

Additional arbitrary metadata can also be associated with a Type.

It can even be made type-safe via an interface extension ArkType exposes for this purpose:

// add this anywhere in your project
declare global {
	interface ArkEnv {
		meta(): {
			// meta properties should always be optional
			secretIngredient?: string
		}
	}
}

// now types you define can specify and access your metadata
const mrPingsSecretIngredientSoup = type({
	broth: "'miso' | 'vegetable'",
	ingredients: "string[]"
}).configure({ secretIngredient: "nothing!" })

prototypes

When you .infer your Types, ArkType traverses them and extracts special values like morphs, e.g. (In: string) => Out<number>.

Though generally this is able to preserve the original type, it is inefficient and can accidentally expand certain object types.

You can use the type-level prototypes config to tell ArkType to treat those types as external:

declare global {
	interface ArkEnv {
		prototypes(): MySpecialClass
	}
}

class MySpecialClass {}

const 
const t: Type<MySpecialClass, {}>t = type.instanceOf(MySpecialClass)