Usage With TypeScript
Redux Toolkit is written in TypeScript, and its API is designed to enable great integration with TypeScript applications.
This page is intended to give an overview of all common usecases and the most probable pitfalls you might encounter when using RTK with TypeScript.
If you encounter any problems with the types that are not described on this page, please open an issue for discussion.
configureStore
with TypeScript
Using Using configureStore should not need any additional typings. You might, however, want to extract the RootState
type and the Dispatch
type.
State
type
Getting the The easiest way of getting the State
type is to define the root reducer in advance and extract its ReturnType
.
It is recommend to give the type a different name like RootState
to prevent confusion, as the type name State
is usually overused.
import { combineReducers } from '@reduxjs/toolkit'const rootReducer = combineReducers({})export type RootState = ReturnType<typeof rootReducer>
Dispatch
type
Getting the If you want to get the Dispatch
type from your store, you can extract it after creating the store.
It is recommend to give the type a different name like AppDispatch
to prevent confusion, as the type name Dispatch
is usually overused.
import { configureStore } from '@reduxjs/toolkit'import rootReducer from './rootReducer'const store = configureStore({ reducer: rootReducer})export type AppDispatch = typeof store.dispatch
Dispatch
type
Correct typings for the The type of the dispatch
function type will be directly inferred from the middleware
option. So if you add correctly typed middlewares, dispatch
should already be correctly typed.
There might however be cases, where TypeScript decides to simplify your provided middleware array down to just Array<Middleware>
. In that case, you have to either specify the array type manually as a tuple, or in TS versions >= 3.4, just add as const
to your definition.
Please note that when calling getDefaultMiddleware
in TypeScript, you have to provide the state type as a generic argument.
import { configureStore } from '@reduxjs/toolkit'import additionalMiddleware from 'additional-middleware'// @ts-ignoreimport untypedMiddleware from 'untyped-middleware'import rootReducer from './rootReducer'
type RootState = ReturnType<typeof rootReducer>const store = configureStore({ reducer: rootReducer, middleware: [ // getDefaultMiddleware needs to be called with the state type ...getDefaultMiddleware<RootState>(), // correctly typed middlewares can just be used additionalMiddleware, // you can also manually type middlewares manually untypedMiddleware as Middleware< (action: Action<'specialAction'>) => number, RootState > ] as const // prevent this from becoming just `Array<Middleware>`})
type AppDispatch = typeof store.dispatch
If you need any additional reference or examples, the type tests for configureStore
contain many different scenarios on how to type this.
Dispatch
type with React-Redux
Using the extracted By default, the React-Redux useDispatch
hook does not contain any types that take middlewares into account. If you need a more specific type for the dispatch
function when dispatching, you may specify the type of the returned dispatch
function, or create a custom-typed version of useSelector
. See the React-Redux documentation for details.
createAction
For most use cases, there is no need to have a literal definition of action.type
, so the following can be used:
createAction<number>('test')
This will result in the created action being of type PayloadActionCreator<number, string>
.
In some setups, you will need a literal type for action.type
, though.
Unfortunately, TypeScript type definitions do not allow for a mix of manually-defined and inferred type parameters, so you'll have to specify the type
both in the Generic definition as well as in the actual JavaScript code:
createAction<number, 'test'>('test')
If you are looking for an alternate way of writing this without the duplication, you can use a prepare callback so that both type parameters can be inferred from arguments, removing the need to specify the action type.
function withPayloadType<T>() { return (t: T) => ({ payload: t })}createAction('test', withPayloadType<string>())
action.type
Alternative to using a literally-typed If you are using action.type
as a discriminator on a discriminated union, for example to correctly type your payload in case
statements, you might be interested in this alternative:
Created action creators have a match
method that acts as a type predicate:
const increment = createAction<number>('increment')function test(action: Action) { if (increment.match(action)) { // action.payload inferred correctly here action.payload }}
This match
method is also very useful in combination with redux-observable
and RxJS's filter
method.
createReducer
The default way of calling createReducer
would be with a "lookup table" / "map object", like this:
createReducer(0, { increment: (state, action: PayloadAction<number>) => state + action.payload})
Unfortunately, as the keys are only strings, using that API TypeScript can neither infer nor validate the action types for you:
{ const increment = createAction<number, 'increment'>('increment') const decrement = createAction<number, 'decrement'>('decrement') createReducer(0, { [increment.type]: (state, action) => { // action is any here }, [decrement.type]: (state, action: PayloadAction<string>) => { // even though action should actually be PayloadAction<number>, TypeScript can't detect that and won't give a warning here. } })}
As an alternative, RTK includes a type-safe reducer builder API.
Building Type-Safe Reducer Argument Objects
Instead of using a simple object as an argument to createReducer
, you can also use a callback that receives a ActionReducerMapBuilder
instance:
const increment = createAction<number, 'increment'>('increment')const decrement = createAction<number, 'decrement'>('decrement')createReducer(0, builder => builder .addCase(increment, (state, action) => { // action is inferred correctly here }) .addCase(decrement, (state, action: PayloadAction<string>) => { // this would error out }))
We recommend using this API if stricter type safety is necessary when defining reducer argument objects.
createSlice
As createSlice
creates your actions as well as your reducer for you, you don't have to worry about type safety here.
Action types can just be provided inline:
{ const slice = createSlice({ name: 'test', initialState: 0, reducers: { increment: (state, action: PayloadAction<number>) => state + action.payload } }) // now available: slice.actions.increment(2) // also available: slice.caseReducers.increment(0, { type: 'increment', payload: 5 })}
If you have too many reducers and defining them inline would be messy, you can also define them outside the createSlice
call and type them as CaseReducer
:
type State = numberconst increment: CaseReducer<State, PayloadAction<number>> = (state, action) => state + action.payload
createSlice({ name: 'test', initialState: 0, reducers: { increment }})
Defining the Initial State Type
You might have noticed that it is not a good idea to pass your SliceState
type as a generic to createSlice
. This is due to the fact that in almost all cases, follow-up generic parameters to createSlice
need to be inferred, and TypeScript cannot mix explicit declaration and inference of generic types within the same "generic block".
The standard approach is to declare an interface or type for your state, create an initial state value that uses that type, and pass the initial state value to createSlice. You can also use the construct
initialState: myInitialState as SliceState`.
type SliceState = { state: 'loading' } | { state: 'finished'; data: string }
// First approach: define the initial state using that typeconst initialState: SliceState = { state: 'loading' }
createSlice({ name: 'test1', initialState, // type SliceState is inferred for the state of the slice reducers: {}})
// Or, cast the initial state as necessarycreateSlice({ name: 'test2', initialState: { state: 'loading' } as SliceState, reducers: {}})
which will result in a Slice<SliceState, ...>
.
prepare
Callbacks
Defining Action Contents with If you want to add a meta
or error
property to your action, or customize the payload
of your action, you have to use the prepare
notation.
Using this notation with TypeScript looks like this:
const blogSlice = createSlice({ name: 'blogData', initialState, reducers: { receivedAll: { reducer( state, action: PayloadAction<Page[], string, { currentPage: number }> ) { state.all = action.payload state.meta = action.meta }, prepare(payload: Page[], currentPage: number) { return { payload, meta: { currentPage } } } } }})
Generated Action Types for Slices
As TS cannot combine two string literals (slice.name
and the key of actionMap
) into a new literal, all actionCreators created by createSlice
are of type 'string'. This is usually not a problem, as these types are only rarely used as literals.
In most cases that type
would be required as a literal, the slice.action.myAction.match
type predicate should be a viable alternative:
const slice = createSlice({ name: 'test', initialState: 0, reducers: { increment: (state, action: PayloadAction<number>) => state + action.payload }})
function myCustomMiddleware(action: Action) { if (slice.actions.increment.match(action)) { // `action` is narrowed down to the type `PayloadAction<number>` here. }}
If you actually need that type, unfortunately there is no other way than manual casting.
extraReducers
Type safety with Reducer lookup tables that map an action type
string to a reducer function are not easy to fully type correctly. This affects both createReducer
and the extraReducers
argument for createSlice
. So, like with createReducer
, you may also use the "builder callback" approach for defining the reducer object argument.
This is particularly useful when a slice reducer needs to handle action types generated by other slices, or generated by specific calls to createAction
(such as the actions generated by createAsyncThunk
).
const fetchUserById = createAsyncThunk( 'users/fetchById', // if you type your function argument here async (userId: number) => { const response = await fetch(`https://reqres.in/api/users/${userId}`) return (await response.json()) as Returned })
interface UsersState { entities: [] loading: 'idle' | 'pending' | 'succeeded' | 'failed'}
const initialState: UsersState = { entities: [], loading: 'idle'}
const usersSlice = createSlice({ name: 'users', initialState, reducers: { // fill in primary logic here }, extraReducers: builder => { builder.addCase(fetchUserById.pending, (state, action) => { // both `state` and `action` are now correctly typed // based on the slice state and the `pending` action creator }) }})
createSlice
Wrapping If you need to reuse reducer logic, it is common to write "higher-order reducers" that wrap a reducer function with additional common behavior. This can be done with createSlice
as well, but due to the complexity of the types for createSlice
, you have to use the SliceCaseReducers
and ValidateSliceCaseReducers
types in a very specific way.
Here is an example of such a "generic" wrapped createSlice
call:
interface GenericState<T> { data?: T status: 'loading' | 'finished' | 'error'}
const createGenericSlice = < T, Reducers extends SliceCaseReducers<GenericState<T>>>({ name = '', initialState, reducers}: { name: string initialState: GenericState<T> reducers: ValidateSliceCaseReducers<GenericState<T>, Reducers>}) => { return createSlice({ name, initialState, reducers: { start(state) { state.status = 'loading' }, /** * If you want to write to values of the state that depend on the generic * (in this case: `state.data`, which is T), you might need to specify the * State type manually here, as it defaults to `Draft<GenericState<T>>`, * which can sometimes be problematic with yet-unresolved generics. * This is a general problem when working with immer's Draft type and generics. */ success(state: GenericState<T>, action: PayloadAction<T>) { state.data = action.payload state.status = 'finished' }, ...reducers } })}
const wrappedSlice = createGenericSlice({ name: 'test', initialState: { status: 'loading' } as GenericState<string>, reducers: { magic(state) { state.status = 'finished' state.data = 'hocus pocus' } }})
createAsyncThunk
In the most common use cases, you should not need to explicitly declare any types for the createAsyncThunk
call itself.
Just provide a type for the first argument to the payloadCreator
argument as you would for any function argument, and the resulting thunk will accept the same type as its input parameter.
The return type of the payloadCreator
will also be reflected in all generated action types.
interface MyData { // ...}
const fetchUserById = createAsyncThunk( 'users/fetchById', // Declare the type your function argument here: async (userId: number) => { const response = await fetch(`https://reqres.in/api/users/${userId}`) // Inferred return type: Promise<MyData> return (await response.json()) as MyData })
// the parameter of `fetchUserById` is automatically inferred to `number` here// and dispatching the resulting thunkAction will return a Promise of a correctly// typed "fulfilled" or "rejected" action.const lastReturnedAction = await store.dispatch(fetchUserById(3))
The second argument to the payloadCreator
, known as thunkApi
, is an object containing references to the dispatch
, getState
, and extra
arguments from the thunk middleware as well as a utility function called rejectWithValue
. If you want to use these from within the payloadCreator
, you will need to define some generic arguments, as the types for these arguments cannot be inferred. Also, as TS cannot mix explicit and inferred generic parameters, from this point on you'll have to define the Returned
and ThunkArg
generic parameter as well.
To define the types for these arguments, pass an object as the third generic argument, with type declarations for some or all of these fields: {dispatch?, state?, extra?, rejectValue?}
.
const fetchUserById = createAsyncThunk< // Return type of the payload creator MyData, // First argument to the payload creator number, { dispatch: AppDispatch state: State extra: { jwt: string } }>('users/fetchById', async (userId, thunkApi) => { const response = await fetch(`https://reqres.in/api/users/${userId}`, { headers: { Authorization: `Bearer ${thunkApi.extra.jwt}` } }) return (await response.json()) as MyData})
If you are performing a request that you know will typically either be a success or have an expected error format, you can pass in a type to rejectValue
and return rejectWithValue(knownPayload)
in the action creator. This allows you to reference the error payload in the reducer as well as in a component after dispatching the createAsyncThunk
action.
interface MyKnownError { errorMessage: string // ...}interface UserAttributes { id: string first_name: string last_name: string email: string}
const updateUser = createAsyncThunk< // Return type of the payload creator MyData, // First argument to the payload creator UserAttributes, // Types for ThunkAPI { extra: { jwt: string } rejectValue: MyKnownError }>('users/update', async (user, thunkApi) => { const { id, ...userData } = user const response = await fetch(`https://reqres.in/api/users/${id}`, { method: 'PUT', headers: { Authorization: `Bearer ${thunkApi.extra.jwt}` }, body: JSON.stringify(userData) }) if (response.status === 400) { // Return the known error for future handling return thunkApi.rejectWithValue((await response.json()) as MyKnownError) } return (await response.json()) as MyData})
While this notation for state
, dispatch
, extra
and rejectValue
might seem uncommon at first, it allows you to provide only the types for these you actually need - so for example, if you are not accessing getState
within your payloadCreator
, there is no need to provide a type for state
. The same can be said about rejectValue
- if you don't need to access any potential error payload, you can ignore it.
In addition, you can leverage checks against action.payload
and match
as provided by createAction
as a type-guard for when you want to access known properties on defined types. Example:
- In a reducer
const usersSlice = createSlice({ name: 'users', initialState: { entities: {}, error: null }, reducers: {}, extraReducers: builder => { builder.addCase(updateUser.fulfilled, (state, { payload }) => { state.entities[payload.id] = payload }) builder.addCase(updateUser.rejected, (state, action) => { if (action.payload) { // Since we passed in `MyKnownError` to `rejectType` in `updateUser`, the type information will be available here. state.error = action.payload.errorMessage } else { state.error = action.error } }) }})
- In a component
const handleUpdateUser = async userData => { const resultAction = await dispatch(updateUser(userData)) if (updateUser.fulfilled.match(resultAction)) { const user = unwrapResult(resultAction) showToast('success', `Updated ${user.name}`) } else { if (resultAction.payload) { // Since we passed in `MyKnownError` to `rejectType` in `updateUser`, the type information will be available here. // Note: this would also be a good place to do any handling that relies on the `rejectedWithValue` payload, such as setting field errors showToast('error', `Update failed: ${resultAction.payload.errorMessage}`) } else { showToast('error', `Update failed: ${resultAction.error.message}`) } }}
createEntityAdapter
Typing createEntityAdapter
only requires you to specify the entity type as the single generic argument.
The example from the createEntityAdapter
documentation would look like this in TypeScript:
interface Book { bookId: number title: string // ...}
const booksAdapter = createEntityAdapter<Book>({ selectId: book => book.bookId, sortComparer: (a, b) => a.title.localeCompare(b.title)})
const booksSlice = createSlice({ name: 'books', initialState: booksAdapter.getInitialState(), reducers: { bookAdded: booksAdapter.addOne, booksReceived(state, action: PayloadAction<{ books: Book[] }>) { booksAdapter.setAll(state, action.payload.books) } }})