Encode failure into your program.
This package contains a Result
type that represents either success (Ok
) or failure (Err
).
For asynchronous tasks, neverthrow
offers a ResultAsync
class which wraps a Promise<Result<T, E>>
and gives you the same level of expressivity and control as a regular Result<T, E>
.
ResultAsync
is thenable
meaning it behaves exactly like a native Promise<Result>
... except you have access to the same methods that Result
provides without having to await
or .then
the promise! Check out the wiki for examples and best practices.
Need to see real-life examples of how to leverage this package for error handling? See this repo: https://github.com/parlez-vous/server
- Installation
- Top-Level API
- API Documentation
- A note on the Package Name
> npm install neverthrow
neverthrow
exposes the following:
ok
convenience function to create anOk
variant ofResult
err
convenience function to create anErr
variant ofResult
Ok
class and typeErr
class and typeResult
Type as well as namespace / object from which to callResult.fromThrowable
ResultAsync
classokAsync
convenience function to create aResultAsync
containing anOk
typeResult
errAsync
convenience function to create aResultAsync
containing anErr
typeResult
combine
utility function that allows you to turnResult<T, E>[]
intoResult<T[], E>
, or aResultAsync<T, E>[]
intoResultAsync<T[], E>
(just likePromise.all
)
import {
ok,
Ok,
err,
Err,
Result,
okAsync,
errAsync,
ResultAsync,
combine,
fromThrowable,
fromPromise,
fromSafePromise,
} from 'neverthrow'
Check out the wiki for help on how to make the most of neverthrow
.
If you find this package useful, please consider sponsoring me or simply buying me a coffee!
Constructs an Ok
variant of Result
Signature:
ok<T, E>(value: T): Ok<T, E> { ... }
Example:
import { ok } from 'neverthrow'
const myResult = ok({ myData: 'test' }) // instance of `Ok`
myResult.isOk() // true
myResult.isErr() // false
Constructs an Err
variant of Result
Signature:
err<T, E>(err: E): Err<T, E> { ... }
Example:
import { err } from 'neverthrow'
const myResult = err('Oh noooo') // instance of `Err`
myResult.isOk() // false
myResult.isErr() // true
Returns true
if the result is an Ok
variant
Signature:
isOk(): boolean { ... }
Returns true
if the result is an Err
variant
Signature:
isErr(): boolean { ... }
Maps a Result<T, E>
to Result<U, E>
by applying a function to a contained Ok
value, leaving an Err
value untouched.
This function can be used to compose the results of two functions.
Signature:
type MapFunc = <T, U>(f: T) => U
map<U>(fn: MapFunc): Result<U, E> { ... }
Example:
const { getLines } from 'imaginary-parser'
// ^ assume getLines has the following signature:
// getLines(str: string): Result<Array<string>, Error>
// since the formatting is deemed correct by `getLines`
// then it means that `linesResult` is an Ok
// containing an Array of strings for each line of code
const linesResult = getLines('1\n2\n3\n4\n')
// this Result now has a Array<number> inside it
const newResult = linesResult.map(
(arr: Array<string>) => arr.map(parseInt)
)
newResult.isOk() // true
Maps a Result<T, E>
to Result<T, F>
by applying a function to a contained Err
value, leaving an Ok
value untouched.
This function can be used to pass through a successful result while handling an error.
Signature:
type MapFunc = <E>(e: E) => F
mapErr<U>(fn: MapFunc): Result<T, F> { ... }
Example:
import { parseHeaders } 'imaginary-http-parser'
// imagine that parseHeaders has the following signature:
// parseHeaders(raw: string): Result<SomeKeyValueMap, ParseError>
const rawHeaders = 'nonsensical gibberish and badly formatted stuff'
const parseResult = parseHeaders(rawHeaders)
parseResult.mapErr(parseError => {
res.status(400).json({
error: parseError
})
})
parseResult.isErr() // true
Unwrap the Ok
value, or return the default if there is an Err
Signature:
unwrapOr<T>(v: T): T { ... }
Example:
const myResult = err('Oh noooo')
const multiply = (val: number): number => val * 2
const unwrapped: number = myResult.map(multiply).unwrapOr(10)
Same idea as map
above. Except you must return a new Result
.
The returned value will be a Result
. As of v4.1.0-beta
, you are able to return distinct error types (see signature below). Prior to v4.1.0-beta
, the error type could not be distinct.
This is useful for when you need to do a subsequent computation using the inner T
value, but that computation might fail.
Additionally, andThen
is really useful as a tool to flatten a Result<Result<A, E2>, E1>
into a Result<A, E2>
(see example below).
Signature:
class Result<T, E> {
// Note that the latest version lets you return distinct errors as well.
// If the error types (E and F) are the same (like `string | string`)
// then they will be merged into one type (`string`)
andThen<U, F>(fn: (val: T) => Result<U, F>): Result<U, E | F> { ... }
}
Example 1: Chaining Results
import { err, ok } from 'neverthrow'
const sq = (n: number): Result<number, number> => ok(n ** 2)
ok(2)
.andThen(sq)
.andThen(sq) // Ok(16)
ok(2)
.andThen(sq)
.andThen(err) // Err(4)
ok(2)
.andThen(err)
.andThen(sq) // Err(2)
err(3)
.andThen(sq)
.andThen(sq) // Err(3)
Example 2: Flattening Nested Results
// It's common to have nested Results
const nested = ok(ok(1234))
// notNested is a Ok(1234)
const notNested = nested.andThen(innerResult => innerResult)
Same idea as andThen
above, except you must return a new ResultAsync
.
The returned value will be a ResultAsync
.
Signature:
class Result<T, E> {
asyncAndThen<U, F>(fn: (val: T) => ResultAsync<U, F>): ResultAsync<U, E | F> { ... }
}
Takes an Err
value and maps it to a Result<T, SomeNewType>
. This is useful for error recovery.
Signature:
type ErrorCallback = <A>(e: E) => Result<T, A>
orElse<A>(f: ErrorCallback<A>): Result<T, A> { ... }
Example:
enum DatabaseError {
PoolExhausted = 'PoolExhausted',
NotFound = 'NotFound',
}
const dbQueryResult: Result<string, DatabaseError> = err(DatabaseError.NotFound)
const updatedQueryResult = dbQueryResult.orElse(dbError =>
dbError === DatabaseError.NotFound
? ok('User does not exist') // error recovery branch: ok() must be called with a value of type string
//
//
// err() can be called with a value of any new type that you want
// it could also be called with the same error value
//
// err(dbError)
: err(500)
)
Given 2 functions (one for the Ok
variant and one for the Err
variant) execute the function that matches the Result
variant.
Match callbacks do not necessitate to return a Result
, however you can return a Result
if you want to.
Signature:
match<A>(
okFn: (t: T) => A,
errFn: (e: E) => A
): A => { ... }
match
is like chaining map
and mapErr
, with the distinction that with match
both functions must have the same return type.
Example:
const result = computationThatMightFail()
const successCallback = (someNumber: number) => {
console.log('> number is: ', someNumber)
}
const failureCallback = (someFailureValue: string) => {
console.log('> boooooo')
}
// method chaining api
// note that you DONT have to append mapErr
// after map which means that you are not required to do
// error handling
result.map(successCallback).mapErr(failureCallback)
// match api
// works exactly the same as above,
// except, now you HAVE to do error handling :)
myval.match(successCallback, failureCallback)
Similar to map
except for two things:
- the mapping function must return a
Promise
- asyncMap returns a
ResultAsync
You can then chain the result of asyncMap
using the ResultAsync
apis (like map
, mapErr
, andThen
, etc.)
Signature:
type MappingFunc = (t: T) => Promise<U>
asyncMap<U>(fn: MappingFunc): ResultAsync<U, E> { ... }
Example:
import { parseHeaders } 'imaginary-http-parser'
// imagine that parseHeaders has the following signature:
// parseHeaders(raw: string): Result<SomeKeyValueMap, ParseError>
const asyncRes = parseHeaders(rawHeader)
.map(headerKvMap => headerKvMap.Authorization)
.asyncMap(findUserInDatabase)
Note that in the above example if parseHeaders
returns an Err
then .map
and .asyncMap
will not be invoked, and asyncRes
variable will resolve to an Err
when turned into a Result
using await
or .then()
.
Although Result is not an actual JS class, the way that
fromThrowable
has been implemented requires that you callfromThrowable
as though it were a static method onResult
. See examples below.
The JavaScript community has agreed on the convention of throwing exceptions. As such, when interfacing with third party libraries it's imperative that you wrap third-party code in try / catch blocks.
This function will create a new function that returns an Err
when the original
function throws.
It is not possible to know the types of the errors thrown in the original
function, therefore it is recommended to use the second argument errorFn
to
map what is thrown to a known type.
Example:
import { Result } from 'neverthrow'
type ParseError = { message: string }
const toParseError = (): ParseError => ({message: "Parse Error" })
const safeJsonParse = Result.fromThrowable(JSON.parse, toParseError)
// the function can now be used safely, if the function throws, the result will be an Err
const res = safeJsonParse("{");
Constructs an Ok
variant of ResultAsync
Signature:
okAsync<T, E>(value: T): ResultAsync<T, E>
Example:
import { okAsync } from 'neverthrow'
const myResultAsync = okAsync({ myData: 'test' }) // instance of `ResultAsync`
const myResult = await myResultAsync // instance of `Ok`
myResult.isOk() // true
myResult.isErr() // false
Constructs an Err
variant of ResultAsync
Signature:
errAsync<T, E>(err: E): ResultAsync<T, E>
Example:
import { errAsync } from 'neverthrow'
const myResultAsync = errAsync('Oh nooo') // instance of `ResultAsync`
const myResult = await myResultAsync // instance of `Err`
myResult.isOk() // false
myResult.isErr() // true
Transforms a Promise<T>
that may throw into a ResultAsync<T, E>
.
The second argument handles the rejection case of the promise and maps the error from unknown
into some type E
.
Signature:
fromPromise<U, E>(p: Promise<U>, f: (e: unknown) => E): ResultAsync<U, E> { ... }
Example:
import { ResultAsync } from 'neverthrow'
import { insertIntoDb } from 'imaginary-database'
// insertIntoDb(user: User): Promise<User>
const res = ResultAsync.fromPromise(insertIntoDb(myUser), () => new Error('Database error'))
// res has a type of ResultAsync<User, Error>
Same as ResultAsync.fromPromise
except that it does not handle the rejection of the promise. Ensure you know what you're doing, otherwise a thrown exception within this promise will cause ResultAsync to reject, instead of resolve to a Result.
Signature:
fromSafePromise<T, E>(p: Promise<T>): ResultAsync<T, E> { ... }
Example:
import { RouteError } from 'routes/error'
// simulate slow routes in an http server that works in a Result / ResultAsync context
// Adopted from https://github.com/parlez-vous/server/blob/2496bacf55a2acbebc30631b5562f34272794d76/src/routes/common/signup.ts
export const slowDown = <T>(ms: number) => (val: T) =>
ResultAsync.fromSafePromise<T, RouteError>(
new Promise((resolve) => {
setTimeout(() => {
resolve(val)
}, ms)
})
)
export const signupHandler = route<User>((req, sessionManager) =>
decode(userSignupDecoder, req.body, 'Invalid request body').map((parsed) => {
return createUser(parsed)
.andThen(slowDown(3000)) // slowdown by 3 seconds
.andThen(sessionManager.createSession)
.map(({ sessionToken, admin }) => AppData.init(admin, sessionToken))
})
)
Maps a ResultAsync<T, E>
to ResultAsync<U, E>
by applying a function to a contained Ok
value, leaving an Err
value untouched.
The applied function can be synchronous or asynchronous (returning a Promise<U>
) with no impact to the return type.
This function can be used to compose the results of two functions.
Signature:
type MapFunc = <T>(f: T | Promise<T>) => U
map<U>(fn: MapFunc): ResultAsync<U, E> { ... }
Example:
const { findUsersIn } from 'imaginary-database'
// ^ assume findUsersIn has the following signature:
// findUsersIn(country: string): ResultAsync<Array<User>, Error>
const usersInCanada = findUsersIn("Canada")
// Let's assume we only need their names
const namesInCanada = usersInCanada.map((users: Array<User>) => users.map(user => user.name))
// namesInCanada is of type ResultAsync<Array<string>, Error>
// We can extract the Result using .then() or await
namesInCanada.then((namesResult: Result<Array<string>, Error>) => {
if(namesResult.isErr()){
console.log("Couldn't get the users from the database", namesResult.error)
}
else{
console.log("Users in Canada are named: " + namesResult.value.join(','))
}
})
Maps a ResultAsync<T, E>
to ResultAsync<T, F>
by applying a function to a contained Err
value, leaving an Ok
value untouched.
The applied function can be synchronous or asynchronous (returning a Promise<F>
) with no impact to the return type.
This function can be used to pass through a successful result while handling an error.
Signature:
type MapFunc = <E>(e: E) => F | Promise<F>
mapErr<U>(fn: MapFunc): ResultAsync<T, F> { ... }
Example:
const { findUsersIn } from 'imaginary-database'
// ^ assume findUsersIn has the following signature:
// findUsersIn(country: string): ResultAsync<Array<User>, Error>
// Let's say we need to low-level errors from findUsersIn to be more readable
const usersInCanada = findUsersIn("Canada").mapErr((e: Error) => {
// The only error we want to pass to the user is "Unknown country"
if(e.message === "Unknown country"){
return e.message
}
// All other errors will be labelled as a system error
return "System error, please contact an administrator."
})
// usersInCanada is of type ResultAsync<Array<User>, string>
usersInCanada.then((usersResult: Result<Array<User>, string>) => {
if(usersResult.isErr()){
res.status(400).json({
error: usersResult.error
})
}
else{
res.status(200).json({
users: usersResult.value
})
}
})
Unwrap the Ok
value, or return the default if there is an Err
.
Works just like Result.unwrapOr
but returns a Promise<T>
instead of T
.
Signature:
unwrapOr<T>(v: T): Promise<T> { ... }
Example:
const unwrapped: number = await errAsync(0).unwrapOr(10)
// unwrapped = 10
Same idea as map
above. Except the applied function must return a Result
or ResultAsync
.
ResultAsync.andThen
always returns a ResultAsync
no matter the return type of the applied function.
This is useful for when you need to do a subsequent computation using the inner T
value, but that computation might fail.
andThen
is really useful as a tool to flatten a ResultAsync<ResultAsync<A, E2>, E1>
into a ResultAsync<A, E2>
(see example below).
Signature:
type AndThenFunc = (t: T) => ResultAsync<U, E> | Result<U, E>
andThen<U>(f: AndThenFunc): ResultAsync<U, E> { ... }
class ResultAsync<T, E> {
// Note that the latest version (v4.1.0-beta) lets you return distinct errors as well.
// If the error types (E and F) are the same (like `string | string`)
// then they will be merged into one type (`string`)
andThen<U, F>(f: (t: T) => Result<U, F> | ResultAsync<U, F>): ResultAsync<U, E | F> { ... }
}
Example
const { validateUser } from 'imaginary-validator'
const { insertUser } from 'imaginary-database'
const { sendNotification } from 'imaginary-service'
// ^ assume validateUser, insertUser and sendNotification have the following signatures:
// validateUser(user: User): Result<User, Error>
// insertUser(user): ResultAsync<User, Error>
// sendNotification(user): ResultAsync<void, Error>
const resAsync = validateUser(user)
.andThen(insertUser)
.andThen(sendNotification)
// resAsync is a ResultAsync<void, Error>
resAsync.then((res: Result<void, Error>) => {
if(res.isErr()){
console.log("Oops, at least one step failed", res.error)
}
else{
console.log("User has been validated, inserted and notified successfully.")
}
})
Takes an Err
value and maps it to a ResultAsync<T, SomeNewType>
. This is useful for error recovery.
Signature:
type ErrorCallback = <A>(e: E) => Result<T, A> | ResultAsync<T, A>
orElse<A>(f: ErrorCallback<A>): ResultAsync<T, A> { ... }
Given 2 functions (one for the Ok
variant and one for the Err
variant) execute the function that matches the ResultAsync
variant.
The difference with Result.match
is that it always returns a Promise
because of the asynchronous nature of the ResultAsync
.
Signature:
match<A>(
okFn: (t: T) => A,
errFn: (e: E) => A
): Promise<A> => { ... }
Example:
const { validateUser } from 'imaginary-validator'
const { insertUser } from 'imaginary-database'
// ^ assume validateUser and insertUser have the following signatures:
// validateUser(user: User): Result<User, Error>
// insertUser(user): ResultAsync<User, Error>
// Handle both cases at the end of the chain using match
const resultMessage = await validateUser(user)
.andThen(insertUser)
.match(
(user: User) => `User ${user.name} has been successfully created`,
(e: Error) => `User could not be created because ${e.message}`
)
// resultMessage is a string
Combine lists of Result
s or lists of ResultAsync
s.
If you're familiar with Promise.all
, the combine function works conceptually the same.
combine
works on both heterogeneous and homogeneous lists. This means that you can have lists that contain different kinds of Result
s and still be able to combine them. Note that you cannot combine lists that contain both Result
s and ResultAsync
s.
The combine function takes a list of results and returns a single result. If all the results in the list are Ok
, then the return value will be a Ok
containing a list of all the individual Ok
values.
If just one of the results in the list is an Err
then the combine function returns that Err value (it short circuits and returns the first Err that it finds).
Formally speaking:
// homogeneous lists
function combine<T, E>(resultList: Result<T, E>[]): Result<T[], E>
// heterogeneous lists
function combine<T1, T2, E1, E2>(resultList: [ Result<T1, E1>, Result<T2, E2> ]): Result<[ T1, T2 ], E1 | E2>
function combine<T1, T2, T3, E1, E2, E3> => Result<[ T1, T2, T3 ], E1 | E2 | E3>
function combine<T1, T2, T3, T4, E1, E2, E3, E4> => Result<[ T1, T2, T3, T4 ], E1 | E2 | E3 | E4>
// ... etc etc ad infinitum
Additionally, this same function also works for ResultAsync
. And thanks to typescript function overloading, the types can be distinguished.
function combine<T, E>(asyncResultList: ResultAsync<T, E>[]): ResultAsync<T[], E>
Like combine
but without short-circuiting. Instead of just the first error value, you get a list of all error values of the input result list.
If only some results fail, the new combined error list will only contain the error value of the failed results, meaning that there is no guarantee of the length of the new error list.
Like combine
, it works for both Result
and ResultAsync
.
Function signature:
// homogeneous lists
function combineWithAllErrors<T, E>(resultList: Result<T, E>[]): Result<T[], E[]>
// heterogeneous lists
function combineWithAllErrors<T1, T2, E1, E2>(resultList: [ Result<T1, E1>, Result<T2, E2> ]): Result<[ T1, T2 ], (E1 | E2)[]>
function combineWithAllErrors<T1, T2, T3, E1, E2, E3> => Result<[ T1, T2, T3 ], (E1 | E2 | E3)[]>
function combineWithAllErrors<T1, T2, T3, T4, E1, E2, E3, E4> => Result<[ T1, T2, T3, T4 ], (E1 | E2 | E3 | E4)[]>
// ... etc etc ad infinitum
Example usage:
const resultList: Result<number, string>[] = [
ok(123),
err('boooom!'),
ok(456),
err('ahhhhh!'),
]
const result = combineWithAllErrors(resultList)
// result is Err(['boooom!', 'ahhhhh!'])
Top level export of Result.fromThrowable
.
Please find documentation at Result.fromThrowable
Top level export of ResultAsync.fromPromise
.
Please find documentation at ResultAsync.fromPromise
Top level export of ResultAsync.fromSafePromise
.
Please find documentation at ResultAsync.fromSafePromise
Result
instances have two unsafe methods, aptly called _unsafeUnwrap
and _unsafeUnwrapErr
which should only be used in a test environment.
_unsafeUnwrap
takes a Result<T, E>
and returns a T
when the result is an Ok
, otherwise it throws a custom object.
_unsafeUnwrapErr
takes a Result<T, E>
and returns a E
when the result is an Err
, otherwise it throws a custom object.
That way you can do something like:
expect(myResult._unsafeUnwrap()).toBe(someExpectation)
However, do note that Result
instances are comparable. So you don't necessarily need to unwrap them in order to assert expectations in your tests. So you could also do something like this:
import { ok } from 'neverthrow'
// ...
expect(callSomeFunctionThatReturnsAResult("with", "some", "args")).toEqual(ok(someExpectation));
By default, the thrown value does not contain a stack trace. This is because stack trace generation makes error messages in Jest harder to understand. If you want stack traces to be generated, call _unsafeUnwrap
and / or _unsafeUnwrapErr
with a config object:
_unsafeUnwrapErr({
withStackTrace: true,
})
// ^ Now the error object will have a `.stack` property containing the current stack
If you find this package useful, please consider sponsoring me or simply buying me a coffee!
Although the package is called neverthrow
, please don't take this literally. I am simply encouraging the developer to think a bit more about the ergonomics and usage of whatever software they are writing.
Throw
ing and catching
is very similar to using goto
statements - in other words; it makes reasoning about your programs harder. Secondly, by using throw
you make the assumption that the caller of your function is implementing catch
. This is a known source of errors. Example: One dev throw
s and another dev uses the function without prior knowledge that the function will throw. Thus, and edge case has been left unhandled and now you have unhappy users, bosses, cats, etc.
With all that said, there are definitely good use cases for throwing in your program. But much less than you might think.