NeverThrow 🙅
Description
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
Table Of Contents
- Installation
- Recommended: Use
eslint-plugin-neverthrow - Top-Level API
- API Documentation
- Synchronous API (
Result)okerrResult.isOk(method)Result.isErr(method)Result.map(method)Result.mapErr(method)Result.unwrapOr(method)Result.andThen(method)Result.asyncAndThen(method)Result.orElse(method)Result.match(method)Result.asyncMap(method)Result.andTee(method)Result.andThrough(method)Result.asyncAndThrough(method)Result.fromThrowable(static class method)Result.combine(static class method)Result.combineWithAllErrors(static class method)Result.safeUnwrap()
- Asynchronous API (
ResultAsync)okAsyncerrAsyncResultAsync.fromThrowable(static class method)ResultAsync.fromPromise(static class method)ResultAsync.fromSafePromise(static class method)ResultAsync.map(method)ResultAsync.mapErr(method)ResultAsync.unwrapOr(method)ResultAsync.andThen(method)ResultAsync.orElse(method)ResultAsync.match(method)ResultAsync.andTee(method)ResultAsync.andThrough(method)ResultAsync.combine(static class method)ResultAsync.combineWithAllErrors(static class method)ResultAsync.safeUnwrap()
- Utilities
- Testing
- Synchronous API (
- A note on the Package Name
Installation
> npm install neverthrowRecommended: Use eslint-plugin-neverthrow
As part of neverthrows bounty program, user mdbetancourt created eslint-plugin-neverthrow to ensure that errors are not gone unhandled.
Install by running:
> npm install eslint-plugin-neverthrowWith eslint-plugin-neverthrow, you are forced to consume the result in one of the following three ways:
- Calling
.match - Calling
.unwrapOr - Calling
._unsafeUnwrap
This ensures that you're explicitly handling the error of your Result.
This plugin is essentially a porting of Rust's must-use attribute.
Top-Level API
neverthrow exposes the following:
okconvenience function to create anOkvariant ofResulterrconvenience function to create anErrvariant ofResultOkclass and typeErrclass and typeResultType as well as namespace / object from which to callResult.fromThrowable, Result.combine.ResultAsyncclassokAsyncconvenience function to create aResultAsynccontaining anOktypeResulterrAsyncconvenience function to create aResultAsynccontaining anErrtypeResult
import {
ok,
Ok,
err,
Err,
Result,
okAsync,
errAsync,
ResultAsync,
fromAsyncThrowable,
fromThrowable,
fromPromise,
fromSafePromise,
safeTry,
} 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!
API Documentation
Synchronous API (Result)
ok
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() // falseerr
Constructs an Err variant of Result
Signature:
err<T, E>(error: E): Err<T, E> { ... }Example:
import { err } from 'neverthrow'
const myResult = err('Oh noooo') // instance of `Err`
myResult.isOk() // false
myResult.isErr() // trueResult.isOk (method)
Returns true if the result is an Ok variant
Signature:
isOk(): boolean { ... }Result.isErr (method)
Returns true if the result is an Err variant
Signature:
isErr(): boolean { ... }Result.map (method)
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:
class Result<T, E> {
map<U>(callback: (value: T) => U): Result<U, E> { ... }
}Example:
import { 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() // trueResult.mapErr (method)
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:
class Result<T, E> {
mapErr<F>(callback: (error: E) => F): Result<T, F> { ... }
}Example:
import { parseHeaders } from '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() // trueResult.unwrapOr (method)
Unwrap the Ok value, or return the default if there is an Err
Signature:
class Result<T, E> {
unwrapOr<T>(value: T): T { ... }
}Example:
const myResult = err('Oh noooo')
const multiply = (value: number): number => value * 2
const unwrapped: number = myResult.map(multiply).unwrapOr(10)Result.andThen (method)
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>(
callback: (value: 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)Result.asyncAndThen (method)
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>(
callback: (value: T) => ResultAsync<U, F>
): ResultAsync<U, E | F> { ... }
}Result.orElse (method)
Takes an Err value and maps it to a Result<T, SomeNewType>. This is useful for error recovery.
Signature:
class Result<T, E> {
orElse<U, A>(
callback: (error: E) => Result<U, A>
): Result<U | 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)
)Result.match (method)
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:
class Result<T, E> {
match<A, B = A>(
okCallback: (value: T) => A,
errorCallback: (error: E) => B
): A | B => { ... }
}match is like chaining map and mapErr, with the distinction that with match both functions must have the same return type.
The differences between match and chaining map and mapErr are that:
- with
matchboth functions must have the same return typeA matchunwraps theResult<T, E>into anA(the match functions' return type)- This makes no difference if you are performing side effects only
Example:
// map/mapErr api
// note that you DON'T have to append mapErr
// after map which means that you are not required to do
// error handling
computationThatMightFail().map(console.log).mapErr(console.error)
// match api
// works exactly the same as above since both callbacks
// only perform side effects,
// except, now you HAVE to do error handling :)
computationThatMightFail().match(console.log, console.error)
// Returning values
const attempt = computationThatMightFail()
.map((str) => str.toUpperCase())
.mapErr((err) => `Error: ${err}`)
// `attempt` is of type `Result<string, string>`
const answer = computationThatMightFail().match(
(str) => str.toUpperCase(),
(err) => `Error: ${err}`
)
// `answer` is of type `string`If you don't use the error parameter in your match callback then match is equivalent to chaining map with unwrapOr:
const answer = computationThatMightFail().match(
(str) => str.toUpperCase(),
() => 'ComputationError'
)
// `answer` is of type `string`
const answer = computationThatMightFail()
.map((str) => str.toUpperCase())
.unwrapOr('ComputationError')Result.asyncMap (method)
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:
class Result<T, E> {
asyncMap<U>(
callback: (value: T) => Promise<U>
): ResultAsync<U, E> { ... }
}Example:
import { parseHeaders } from '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().
Result.andTee (method)
Takes a Result<T, E> and lets the original Result<T, E> pass through regardless the result of the passed-in function.
This is a handy way to handle side effects whose failure or success should not affect your main logics such as logging.
Signature:
class Result<T, E> {
andTee(
callback: (value: T) => unknown
): Result<T, E> { ... }
}Example:
import { parseUserInput } from 'imaginary-parser'
import { logUser } from 'imaginary-logger'
import { insertUser } from 'imaginary-database'
// ^ assume parseUserInput, logUser and insertUser have the following signatures:
// parseUserInput(input: RequestData): Result<User, ParseError>
// logUser(user: User): Result<void, LogError>
// insertUser(user: User): ResultAsync<void, InsertError>
// Note logUser returns void upon success but insertUser takes User type.
const resAsync = parseUserInput(userInput)
.andTee(logUser)
.asyncAndThen(insertUser)
// Note no LogError shows up in the Result type
resAsync.then((res: Result<void, ParseError | InsertError>) => {e
if(res.isErr()){
console.log("Oops, at least one step failed", res.error)
}
else{
console.log("User input has been parsed and inserted successfully.")
}
}))Result.andThrough (method)
Similar to andTee except for:
- when there is an error from the passed-in function, that error will be passed along.
Signature:
class Result<T, E> {
andThrough<F>(
callback: (value: T) => Result<unknown, F>
): Result<T, E | F> { ... }
}Example:
import { parseUserInput } from 'imaginary-parser'
import { validateUser } from 'imaginary-validator'
import { insertUser } from 'imaginary-database'
// ^ assume parseUseInput, validateUser and insertUser have the following signatures:
// parseUserInput(input: RequestData): Result<User, ParseError>
// validateUser(user: User): Result<void, ValidateError>
// insertUser(user: User): ResultAsync<void, InsertError>
// Note validateUser returns void upon success but insertUser takes User type.
const resAsync = parseUserInput(userInput)
.andThrough(validateUser)
.asyncAndThen(insertUser)
resAsync.then((res: Result<void, ParseErro | ValidateError | InsertError>) => {e
if(res.isErr()){
console.log("Oops, at least one step failed", res.error)
}
else{
console.log("User input has been parsed, validated, inserted successfully.")
}
}))Result.asyncAndThrough (method)
Similar to andThrough except you must return a ResultAsync.
You can then chain the result of asyncAndThrough using the ResultAsync apis (like map, mapErr, andThen, etc.)
Signature:
import { parseUserInput } from 'imaginary-parser'
import { insertUser } from 'imaginary-database'
import { sendNotification } from 'imaginary-service'
// ^ assume parseUserInput, insertUser and sendNotification have the following signatures:
// parseUserInput(input: RequestData): Result<User, ParseError>
// insertUser(user: User): ResultAsync<void, InsertError>
// sendNotification(user: User): ResultAsync<void, NotificationError>
// Note insertUser returns void upon success but sendNotification takes User type.
const resAsync = parseUserInput(userInput)
.asyncAndThrough(insertUser)
.andThen(sendNotification)
resAsync.then((res: Result<void, ParseError | InsertError | NotificationError>) => {e
if(res.isErr()){
console.log("Oops, at least one step failed", res.error)
}
else{
console.log("User has been parsed, inserted and notified successfully.")
}
}))Result.fromThrowable (static class method)
Although Result is not an actual JS class, the way that
fromThrowablehas been implemented requires that you callfromThrowableas 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("{");Result.combine (static class method)
Although Result is not an actual JS class, the way that
combinehas been implemented requires that you callcombineas though it were a static method onResult. See examples below.
Combine lists of Results.
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 Results and still be able to combine them. Note that you cannot combine lists that contain both Results and ResultAsyncs.
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 infinitumExample:
const resultList: Result<number, never>[] =
[ok(1), ok(2)]
const combinedList: Result<number[], unknown> =
Result.combine(resultList)Example with tuples:
/** @example tuple(1, 2, 3) === [1, 2, 3] // with type [number, number, number] */
const tuple = <T extends any[]>(...args: T): T => args
const resultTuple: [Result<string, never>, Result<string, never>] =
tuple(ok('a'), ok('b'))
const combinedTuple: Result<[string, string], unknown> =
Result.combine(resultTuple)Result.combineWithAllErrors (static class method)
Although Result is not an actual JS class, the way that
combineWithAllErrorshas been implemented requires that you callcombineWithAllErrorsas though it were a static method onResult. See examples below.
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.
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 infinitumExample usage:
const resultList: Result<number, string>[] = [
ok(123),
err('boooom!'),
ok(456),
err('ahhhhh!'),
]
const result = Result.combineWithAllErrors(resultList)
// result is Err(['boooom!', 'ahhhhh!'])Result.safeUnwrap()
.safeUnwrap in a generator context with safeTry. Please see safeTry.
Allows for unwrapping a Result or returning an Err implicitly, thereby reducing boilerplate.
Asynchronous API (ResultAsync)
okAsync
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() // falseerrAsync
Constructs an Err variant of ResultAsync
Signature:
errAsync<T, E>(error: 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() // trueResultAsync.fromThrowable (static class method)
Similar to Result.fromThrowable, but for functions that return a Promise.
Example:
import { ResultAsync } from 'neverthrow'
import { insertIntoDb } from 'imaginary-database'
// insertIntoDb(user: User): Promise<User>
const insertUser = ResultAsync.fromThrowable(insertIntoDb, () => new Error('Database error'))
// `res` has a type of (user: User) => ResultAsync<User, Error>Note that this can be safer than using ResultAsync.fromPromise with
the result of a function call, because not all functions that return a Promise are async, and thus they can throw
errors synchronously rather than returning a rejected Promise. For example:
// NOT SAFE !!
import { ResultAsync } from 'neverthrow'
import { db } from 'imaginary-database'
// db.insert<T>(table: string, value: T): Promise<T>
const insertUser = (user: User): Promise<User> => {
if (!user.id) {
// this throws synchronously!
throw new TypeError('missing user id')
}
return db.insert('users', user)
}
// this will throw, NOT return a `ResultAsync`
const res = ResultAsync.fromPromise(insertIntoDb(myUser), () => new Error('Database error'))ResultAsync.fromPromise (static class method)
Transforms a PromiseLike<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 is a static class method
// also available as a standalone function
// import { fromPromise } from 'neverthrow'
ResultAsync.fromPromise<T, E>(
promise: PromiseLike<T>,
errorHandler: (unknownError: unknown) => E)
): ResultAsync<T, E> { ... }If you are working with PromiseLike objects that you know for a fact will not throw, then use fromSafePromise in order to avoid having to pass a redundant errorHandler argument.
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>ResultAsync.fromSafePromise (static class method)
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:
// fromPromise is a static class method
// also available as a standalone function
// import { fromPromise } from 'neverthrow'
ResultAsync.fromSafePromise<T, E>(
promise: PromiseLike<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) => (value: T) =>
ResultAsync.fromSafePromise<T, RouteError>(
new Promise((resolve) => {
setTimeout(() => {
resolve(value)
}, 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))
})
)ResultAsync.map (method)
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:
class ResultAsync<T, E> {
map<U>(
callback: (value: T) => U | Promise<U>
): ResultAsync<U, E> { ... }
}Example:
import { 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(','))
}
})ResultAsync.mapErr (method)
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:
class ResultAsync<T, E> {
mapErr<F>(
callback: (error: E) => F | Promise<F>
): ResultAsync<T, F> { ... }
}Example:
import { 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((error: Error) => {
// The only error we want to pass to the user is "Unknown country"
if(error.message === "Unknown country"){
return error.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
})
}
})ResultAsync.unwrapOr (method)
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:
class ResultAsync<T, E> {
unwrapOr<T>(value: T): Promise<T> { ... }
}Example:
const unwrapped: number = await errAsync(0).unwrapOr(10)
// unwrapped = 10ResultAsync.andThen (method)
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:
// 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`)
class ResultAsync<T, E> {
andThen<U, F>(
callback: (value: T) => Result<U, F> | ResultAsync<U, F>
): ResultAsync<U, E | F> { ... }
}Example
import { validateUser } from 'imaginary-validator'
import { insertUser } from 'imaginary-database'
import { 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.")
}
})ResultAsync.orElse (method)
Takes an Err value and maps it to a ResultAsync<T, SomeNewType>. This is useful for error recovery.
Signature:
class ResultAsync<T, E> {
orElse<U, A>(
callback: (error: E) => Result<U, A> | ResultAsync<U, A>
): ResultAsync<U | T, A> { ... }
}ResultAsync.match (method)
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:
class ResultAsync<T, E> {
match<A, B = A>(
okCallback: (value: T) => A,
errorCallback: (error: E) => B
): Promise<A | B> => { ... }
}Example:
import { validateUser } from 'imaginary-validator'
import { 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`,
(error: Error) => `User could not be created because ${error.message}`
)
// resultMessage is a stringResultAsync.andTee (method)
Takes a ResultAsync<T, E> and lets the original ResultAsync<T, E> pass through regardless
the result of the passed-in function.
This is a handy way to handle side effects whose failure or success should not affect your main logics such as logging.
Signature:
class ResultAsync<T, E> {
andTee(
callback: (value: T) => unknown
): ResultAsync<T, E> => { ... }
}Example:
import { insertUser } from 'imaginary-database'
import { logUser } from 'imaginary-logger'
import { sendNotification } from 'imaginary-service'
// ^ assume insertUser, logUser and sendNotification have the following signatures:
// insertUser(user: User): ResultAsync<User, InsertError>
// logUser(user: User): Result<void, LogError>
// sendNotification(user: User): ResultAsync<void, NotificationError>
// Note logUser returns void on success but sendNotification takes User type.
const resAsync = insertUser(user)
.andTee(logUser)
.andThen(sendNotification)
// Note there is no LogError in the types below
resAsync.then((res: Result<void, InsertError | NotificationError>) => {e
if(res.isErr()){
console.log("Oops, at least one step failed", res.error)
}
else{
console.log("User has been inserted and notified successfully.")
}
})) ResultAsync.andThrough (method)
Similar to andTee except for:
- when there is an error from the passed-in function, that error will be passed along.
Signature:
class ResultAsync<T, E> {
andThrough<F>(
callback: (value: T) => Result<unknown, F> | ResultAsync<unknown, F>,
): ResultAsync<T, E | F> => { ... }
}Example:
import { buildUser } from 'imaginary-builder'
import { insertUser } from 'imaginary-database'
import { sendNotification } from 'imaginary-service'
// ^ assume buildUser, insertUser and sendNotification have the following signatures:
// buildUser(userRaw: UserRaw): ResultAsync<User, BuildError>
// insertUser(user: User): ResultAsync<void, InsertError>
// sendNotification(user: User): ResultAsync<void, NotificationError>
// Note insertUser returns void upon success but sendNotification takes User type.
const resAsync = buildUser(userRaw)
.andThrough(insertUser)
.andThen(sendNotification)
resAsync.then((res: Result<void, BuildError | InsertError | NotificationError>) => {e
if(res.isErr()){
console.log("Oops, at least one step failed", res.error)
}
else{
console.log("User data has been built, inserted and notified successfully.")
}
})) ResultAsync.combine (static class method)
Combine lists of ResultAsyncs.
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 ResultAsyncs and still be able to combine them. Note that you cannot combine lists that contain both Results and ResultAsyncs.
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: ResultAsync<T, E>[]): ResultAsync<T[], E>
// heterogeneous lists
function combine<T1, T2, E1, E2>(resultList: [ ResultAsync<T1, E1>, ResultAsync<T2, E2> ]): ResultAsync<[ T1, T2 ], E1 | E2>
function combine<T1, T2, T3, E1, E2, E3> => ResultAsync<[ T1, T2, T3 ], E1 | E2 | E3>
function combine<T1, T2, T3, T4, E1, E2, E3, E4> => ResultAsync<[ T1, T2, T3, T4 ], E1 | E2 | E3 | E4>
// ... etc etc ad infinitumExample:
const resultList: ResultAsync<number, never>[] =
[okAsync(1), okAsync(2)]
const combinedList: ResultAsync<number[], unknown> =
ResultAsync.combine(resultList)Example with tuples:
/** @example tuple(1, 2, 3) === [1, 2, 3] // with type [number, number, number] */
const tuple = <T extends any[]>(...args: T): T => args
const resultTuple: [ResultAsync<string, never>, ResultAsync<string, never>] =
tuple(okAsync('a'), okAsync('b'))
const combinedTuple: ResultAsync<[string, string], unknown> =
ResultAsync.combine(resultTuple)ResultAsync.combineWithAllErrors (static class method)
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.
Function signature:
// homogeneous lists
function combineWithAllErrors<T, E>(resultList: ResultAsync<T, E>[]): ResultAsync<T[], E[]>
// heterogeneous lists
function combineWithAllErrors<T1, T2, E1, E2>(resultList: [ ResultAsync<T1, E1>, ResultAsync<T2, E2> ]): ResultAsync<[ T1, T2 ], (E1 | E2)[]>
function combineWithAllErrors<T1, T2, T3, E1, E2, E3> => ResultAsync<[ T1, T2, T3 ], (E1 | E2 | E3)[]>
function combineWithAllErrors<T1, T2, T3, T4, E1, E2, E3, E4> => ResultAsync<[ T1, T2, T3, T4 ], (E1 | E2 | E3 | E4)[]>
// ... etc etc ad infinitumExample usage:
const resultList: ResultAsync<number, string>[] = [
okAsync(123),
errAsync('boooom!'),
okAsync(456),
errAsync('ahhhhh!'),
]
const result = ResultAsync.combineWithAllErrors(resultList)
// result is Err(['boooom!', 'ahhhhh!'])ResultAsync.safeUnwrap()
.safeUnwrap in a generator context with safeTry. Please see safeTry.
Allows for unwrapping a Result or returning an Err implicitly, thereby reducing boilerplate.
Utilities
fromThrowable
Top level export of Result.fromThrowable.
Please find documentation at Result.fromThrowable
fromAsyncThrowable
Top level export of ResultAsync.fromSafePromise.
Please find documentation at ResultAsync.fromThrowable
fromPromise
Top level export of ResultAsync.fromPromise.
Please find documentation at ResultAsync.fromPromise
fromSafePromise
Top level export of ResultAsync.fromSafePromise.
Please find documentation at ResultAsync.fromSafePromise
safeTry
Used to implicitly return errors and reduce boilerplate.
Let's say we are writing a function that returns a Result, and in that function we call some functions which also return Results and we check those results to see whether we should keep going or abort. Usually, we will write like the following.
declare function mayFail1(): Result<number, string>;
declare function mayFail2(): Result<number, string>;
function myFunc(): Result<number, string> {
// We have to define a constant to hold the result to check and unwrap its value.
const result1 = mayFail1();
if (result1.isErr()) {
return err(`aborted by an error from 1st function, ${result1.error}`);
}
const value1 = result1.value
// Again, we need to define a constant and then check and unwrap.
const result2 = mayFail2();
if (result2.isErr()) {
return err(`aborted by an error from 2nd function, ${result2.error}`);
}
const value2 = result2.value
// And finally we return what we want to calculate
return ok(value1 + value2);
}Basically, we need to define a constant for each result to check whether it's a Ok and read its .value or .error.
With safeTry, we can state 'Return here if its an Err, otherwise unwrap it here and keep going.' in just one expression.
declare function mayFail1(): Result<number, string>;
declare function mayFail2(): Result<number, string>;
function myFunc(): Result<number, string> {
return safeTry<number, string>(function*() {
return ok(
// If the result of mayFail1().mapErr() is an `Err`, the evaluation is
// aborted here and the enclosing `safeTry` block is evaluated to that `Err`.
// Otherwise, this `(yield* ...)` is evaluated to its `.value`.
(yield* mayFail1()
.mapErr(e => `aborted by an error from 1st function, ${e}`)
.safeUnwrap())
+
// The same as above.
(yield* mayFail2()
.mapErr(e => `aborted by an error from 2nd function, ${e}`)
.safeUnwrap())
)
})
}To use safeTry, the points are as follows.
- Wrap the entire block in a generator function
- In that block, you can use
yield* <RESULT>to state 'Return<RESULT>if it's anErr, otherwise evaluate to its.value' - Pass the generator function to
safeTry
You can also use async generator function to pass an async block to safeTry.
// You can use either Promise<Result> or ResultAsync.
declare function mayFail1(): Promise<Result<number, string>>;
declare function mayFail2(): ResultAsync<number, string>;
function myFunc(): Promise<Result<number, string>> {
return safeTry<number, string>(async function*() {
return ok(
// You have to await if the expression is Promise<Result>
(yield* (await mayFail1())
.mapErr(e => `aborted by an error from 1st function, ${e}`)
.safeUnwrap())
+
// You can call `safeUnwrap` directly if its ResultAsync
(yield* mayFail2()
.mapErr(e => `aborted by an error from 2nd function, ${e}`)
.safeUnwrap())
)
})
}For more information, see #448 and #444
Testing
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 stackIf you find this package useful, please consider sponsoring me or simply buying me a coffee!
A note on the Package Name
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.
Throwing 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 throws 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.
License
The neverthrow project is available as open source under the terms of the MIT license.