Releases: reduxjs/redux-toolkit
v1.3.0 Beta 0
This release adds two new APIs: createEntityAdapter
to help manage normalized state, and createAsyncThunk
to abstract common data fetching behavior.
It also improves bundle size by fixing cases where APIs were accidentally being included in production and inlining some of our prior dependencies, as well as using a new version of Immer that tree-shakes better.
Note: this is a beta release. We feel that the APIs are stable and are hopefully ready for a final release soon, and they have been documented and tested. However, we'd still like some additional feedback from users before publishing a final release.
Please comment in issue #373: Roadmap: RTK v1.3.0 with any feedback on using these new APIs.
This version is available as @reduxjs/toolkit@beta
on NPM, or @reduxjs/toolkit@1.3.0-beta.0
.
New APIs
createAsyncThunk
The Redux docs have taught that async logic should typically dispatch "three-phase async actions" while doing data fetching: a "start" action before the request is made so that loading UI can be displayed, and then a "success" or "failure" action to handle loading the data or showing an error message. Writing these extra action types is tedious, as is writing thunks that dispatch these actions and differ only by what the async request is.
Given that this is a very common pattern, we've added a createAsyncThunk
API that abstracts this out. It accepts a base action type string and a callback function that returns a Promise, which is primarily intended to be a function that does a data fetch and returns a Promise containing the results. It then auto-generates the request lifecycle action types / creators, and generates a thunk that dispatches those lifecycle actions and runs the fetching callback.
From there, you can listen for those generated action types in your reducers, and handle loading state as desired.
createEntityAdapter
The Redux docs have also advised storing data in a "normalized" state shape, which typically means keeping each type of item in a structure that looks like {ids: [], entities: {} }
. However, the Redux core provides no APIs to help manage storing and updating your data using this approach. Many community libraries exist, with varying tradeoffs, but so far we haven't officially recommended any of them.
Caching data is a hard problem, and not one that we are interested in trying to solve ourselves. However, given that we do recommend this specific pattern, and that Redux Toolkit is intended to help simplify common use cases, we want to provide a minimal set of functionality to help users manage normalized state.
To help solve this, we've specifically ported the @ngrx/entity
library to work with Redux Toolkit, with some modifications.
The core API function is createEntityAdapter
. It generates a set of reducer functions and selectors that know how to work with data that has been stored in that normalized {ids: [], entities: {} }
format, and can be customized by passing in a function that returns the ID field for a given item. If you want to keep the item IDs in a sorted order, a comparison function can also be passed in.
The returned EntityAdapter
object contains generated CRUD functions for manipulating items within that state, and generated selector functions that know how to read from that state. You can then use the generated CRUD functions and selectors within your own code.
There is one very important difference between RTK's implementation and the original @ngrx/entity
implementation. With @ngrx/entity
, methods like addOne(item, state)
accept the data argument first and the state second. With RTK, the argument order has been flipped, so that the methods look like addOne(state, item)
, and the methods can also accept a standard Redux Toolkit PayloadAction
containing the data as the second argument. This allows them to be used as Redux case reducers directly, such as passing them in the reducers
argument for createSlice
. They can also be used as "mutating" helper functions inside createReducer
and createSlice
as well, thanks to use of Immer internally.
Documentation
See our beta API reference and usage guide docs for details:
- Usage:
- API reference:
Bundle Size Improvements and Dependency Updates
Immer 6.0
Immer has always been the largest chunk of code added to your bundle from using RTK. Until now, RTK specifically depended on Immer 4.x, since 5.x added support for handling Map
s and Set
s (which aren't useful in a Redux app) and that support added to its bundle size.
Immer's code was written in a way that kept it from tree-shaking properly. Fortunately, Immer author Michel Weststrate put in some amazing work refactoring the code to better support tree-shaking, and his efforts are now available as Immer 6.0.
Per the Immer documentation on customizing Immer's capabilities, Immer now uses a plugin architecture internally, and additional functionality has to be explicitly enabled as an opt-in. There are currently three Immer plugins that can be enabled: ES5 support (for environments without ES6 Proxies), Map/Set
support, and JSON Patch support.
Redux Toolkit force-enables ES5 support. This is because we expect RTK to be used in multiple environments that do not support Proxies, such as Internet Explorer and React Native. It's also how Immer previously behaved, so we want to keep that behavior consistent and not break code given that this is a minor release of RTK. (In a hypothetical future major release, we may stop force-enabling the ES5 plugin and ask you to do it if necessary.)
Overall, this should drop a couple KB off your app's minified bundle size.
You may choose to enable the other plugins in your app code if that functionality is desired.
Store Configuration Dependencies
Since its creation, RTK has depended on leoasis/redux-immutable-state-invariant
to throw errors if accidental mutations are detected, and the zalmoxisus/redux-devtools-extension
NPM package to handle setup and configuration of the Redux DevTools Extension as the store is created.
Unfortunately, neither of these dependencies is currently published as ES Modules, and we recently found out that the immutable middleware was actually being included in production bundles despite our attempts to ensure it is excluded.
Given that the repo for the immutable middleware has had no activity in the last 3 years, we've opted to fork the package and include the code directly inside Redux Toolkit. We've also inlined the tiny-invariant
and json-stringify-safe
packages that the immutable middleware depended on.
The DevTools setup package, while tiny, suffers from the same issue, and so we've forked it as well.
Based on tests locally, these changes should reduce your production bundle sizes by roughly 2.5K minified.
Changes since Alpha
This beta release has a few additional changes since the 1.3.0-alpha.10 release.
createAsyncThunk
Error Handling Improvements
We've spent a lot of time going over the error handling semantics for createAsyncThunk
to ensure that errors can be handled outside the thunk, be correctly typed, and that validation errors from a server can be processed correctly. We also now detect if AbortController
is available in the environment, and if not, provide a tiny polyfill that suggests adding a real polyfill to your application.
See PR #393: createAsyncThunk
: reject with typed value for much of the discussion and work, as well as the API reference docs.
Middleware Updates
We found that the serializable invariant middleware was partly being included in production. We've decided that both the immutable and serializable middleware should always be no-ops in prod, both to ensure minimum bundle size, and to eliminate any unwanted slowdowns.
In addition, the serializable middleware now ignores meta.args
in every action by default. This is because createAsyncThunk
automatically takes any arguments to its payload creator function and inserts them into dispatched actions. Since a user may be reasonably passing non-serializable values as arguments, and they're not intentionally inserting those into actions themselves, it seems sensible to ignore any potential non-serializable values in that field.
Immer 6 final
We've updated Immer 6 from an alpha build to its final 6.0.1 release. This fixes the ability to use RTK with TS 3.5 and 3.6, as Immer has re-added typings support for those TS versions.
Other Changes
TypeScript Support...
v.1.3.0-alpha.10
This release reduces bundle sizes by updating to the latest Immer 6 alpha to help reduce bundle sizes and inlining the nanoid
dependency.
Note: Due to the Immer upgrade, this RTK alpha version requires at least TypeScript 3.7 if you are consuming it in a TypeScript app. The final version of RTK 1.3 will hopefully support TS 3.5+ once immerjs/immer#541 has been merged in.
Changes
Bundle Size Improvements
Immer has always been the largest chunk of code added to your bundle from using RTK. Until now, RTK specifically depended on Immer 4.x, since 5.x added support for handling Map
s and Set
s (which aren't useful in a Redux app) and that support added to its bundle size.
Immer's code was written in a way that kept it from tree-shaking properly. Fortunately, Immer author Michel Weststrate has been hard at work refactoring the code to better support tree-shaking, and his effort are now available as Immer 6.x alpha.
Per the updated Immer alpha installation docs, Immer now uses a plugin architecture internally, and additional functionality has to be explicitly enabled as an opt-in. There are currently three Immer plugins that can be enabled: ES5 support (for environments without ES6 Proxies), Map/Set
support, and JSON Patch support.
In this alpha, Redux Toolkit force-enables ES5 support. This is because we expect RTK to be used in multiple environments that do not support Proxies, such as Internet Explorer and React Native. It's also how Immer previously behaved, so we want to keep that behavior consistent and not break code given that this is a minor release of RTK. (In a hypothetical future major release, we may stop force-enabling the ES5 plugin and ask you to do it if necessary.)
Overall, this should drop a couple KB off your app's minified bundle size.
You may choose to enable the other plugins in your app code if that functionality is desired.
Inlined nanoid
We received reports that the nanoid
library, which we used for generating unique request IDs in createAsyncThunk
, prints warnings on React Native due to a lack of crypto availability. Since we don't need anything cryptographically secure, just reasonably random, we've inlined the function from nanoid/no-secure
and dropped the nanoid
dependency.
Documentation
We've added TypeScript usage guides for createAsyncThunk
and createEntityAdapter
to the alpha docs:
Alpha docs: createAsyncThunk
and createEntityAdapter
TypeScript usage guide
Changelog
- Use Immer 6 alpha (@markerikson - #396)
v1.3.0-alpha.9
This release reduces bundle sizes by forking and inlining the redux-immutable-state-invariant
and redux-devtools-extension
dependencies, and modifies the types for createEntityAdapter
.
Changes
Dependency Updates and Bundle Size
Since its creation, RTK has depended on leoasis/redux-immutable-state-invariant
to throw errors if accidental mutations are detected, and the zalmoxisus/redux-devtools-extension
NPM package to handle setup and configuration of the Redux DevTools Extension as the store is created.
Unfortunately, neither of these dependencies is currently published as ES Modules, and we recently found out that the immutable middleware was actually being included in production bundles despite our attempts to ensure it is excluded.
Given that the repo for the immutable middleware has had no activity in the last 3 years, we've opted to fork the package and include the code directly inside Redux Toolkit. We've also inlined the tiny-invariant
and json-stringify-safe
packages that the immutable middleware depended on.
The DevTools setup package, while tiny, suffers from the same issue, and so we've forked it as well.
Based on tests locally, these changes should reduce your production bundle sizes by roughly 2.5K minified.
As a future note, Immer is currently being reworked to enable better shakeability, and once that is released, we plan on updating RTK to use the new Immer version as well. This is unlikely to make it into RTK 1.3.0, but we'll put out another update once we've been able to confirm that the changes work for us.
createEntityAdapter
Types
We made a few more tweaks to the type definitions for createEntityAdapter
. Shouldn't be anything breaking, but clarifying the possible overloads for the generated methods.
Changelog
- Immutable middleware cleanup (@markerikson - #385)
- Feature/immutable invariant (@msutkowski - #381)
- Fork redux-devtools-extension (@markerikson - #384)
- remove
any
where applicable (@phryneas - #377)
v1.3.0-alpha.8
This release fixes a couple edge case bugs with entity update operations, and documents expected update behavior. Also, since the new APIs look to be somewhat stabilized, we've merged the original PR into a v1.3.0 tracking branch where we can integrate additional planned changes.
Roadmap
We've put up a v1.3.0 roadmap issue to track other planned work before 1.3.0 is released.
Changes
createEntityAdapter
Update Logic
We identified a couple potential bugs and edge cases inside the updateMany
method implementation. We've fixed a potential issue that might have appeared when multiple updates were passed in attempting to rename the same entity to different IDs in a row, fixed a wrong type definition for comparer callbacks, and documented expected behavior when ID renames do occur.
Documentation
The alpha docs are available at https://deploy-preview-374--redux-starter-kit-docs.netlify.com/. In particular, see the API references for:
Changelog
- Port ngrx/entity and add createAsyncThunk (@markerikson, @phryneas - #352) 219be24
v1.3.0-alpha.7
This release reworks the cancellation functionality in createAsyncThunk
.
Changes
createAsyncThunk
cancellation
We previously added AbortController
support to createAsyncThunk
, to allow other code to trigger whatever cancellation might be applicable to the async logic inside.
The thunk now exits early when the abort()
method is called and doesn't keep waiting for the payloadCreator
to finish. This should be an improvement especially in cases where the payloadCreator
didn't care about the abort signal, but something outside was awaiting the promise from dispatch.
Also, before this, a non-signal-aware payloadCreator
could still finish which would have caused a dispatch of a "fulfilled" action after an "rejected" (abort) action, which was confusing.
We've also removed the meta.abortReason
property as it's no longer possible to override the error in the "rejected" action.
Changelog
v1.3.0-alpha.6
This release alters the internal behavior of the createEntityAdapter
CRUD methods to allow them to work as "mutating helper functions" when not directly used as case reducers, and adds initial API reference documentation for createEntityAdapter
.
Changes
createEntityAdapter
and Immer
The original version of createEntityAdapter
that we ported from the @ngrx/entity
library used hand-written immutable update logic internally. We replaced that with use of Immer, as it made it consistent with how createReducer
and createSlice
work, and simplified some of the logic.
Previously, the CRUD methods such as addOne()
always called Immer's createNextState()
internally to wrap the updating logic. This worked okay when the CRUD method was used as a case reducer directly.
However, when called manually as a helper function inside an existing case reducer, the behavior was confusing. The case reducer had already received state
as an Immer Draft
value, but that draft was being passed into createNextState
again. In theory, updates to the nested draft are supposed to propagate back to the parent draft, but we didn't see updates propagating upwards as expected.
We've updated the CRUD methods to first check whether the incoming state
value is an Immer Draft
, or just a plain JS value. If it's already a Draft
, we pass that draft to the mutating internal update logic, so the changes are applied correctly. If it's a plain object, we still call createNextState()
and return a plain value.
So, when using the CRUD methods as helper functions, treat them as mutating the existing state
value:
const booksSlice = createSlice({
name: "books",
initialState: booksAdapter.getInitialState({totalBooks: 0),
reducers: {
bookAdded(state, action) {
// Ignore return value here, and "mutate" state
booksAdapter.addOne(state, action.payload.book)
// can continue to mutate state here
state.totalBooks++
}
}
})
Documentation
A first draft of API documentation for createEntityAdapter
is now available:
createEntityAdapter
draft API reference
Changelog
v1.3.0-alpha.5
This release reworks the createAsyncThunk
types to enable more flexibility in declaring optional generic arguments.
Changes
createAsyncThunk
Generic Types
Per notes in alpha.4
, our original types made it actually impossible to declare the correct State
type for use by getState
in your promise payload creator callback.
We came up with a stopgap solution, but since TS doesn't allow you to specify individual generic arguments by name, it meant that specifying types for some of the thunk options might require specifying all generic types whether you wanted to or not.
Fortunately, @Ethan-Arrowood has found a novel technique for optionally overriding specific generics by name, in a syntax similar to object destructuring, and we've been able to apply that here.
Per the previous example, the most common use case for createAsyncThunk
still does not require specifying any types by hand, as the types for the Returned
value and the thunkArg
argument will be inferred from your payload creator callback:
// basic usage:
const thunk1 = createAsyncThunk("a", async (arg: string) => {
return 42
})
To specify the type of State
, you'll need to specifically declare the types of Returned
and thunkArg
. Then, pass in an object as the third generic argument, and declare the type of a field named state
inside that object:
// specify state type for getState usage
const thunk2 = createAsyncThunk<Promise<number>, string, {state: RootState}>(
"a",
async (arg: string, {getState}) => {
const state = getState();
return 42;
}
)
If you only want to declare the type of the extra
thunk argument, do the same thing, but override the extra
field instead of state
:
interface UserAPI {
fetchUserById: (id: number) => Promise<User>
}
// specify state type for getState usage
const thunk2 = createAsyncThunk<Promise<User>, string, {extra: UserAPI}>(
"a",
async (arg: string, {extra}) => {
const user = await extra.fetchUserById(123)
return user;
}
)
Previously, that would have required also declaring the type of the state
generic, since state
was listed before extra
in the generics definition.
Changelog
- use ThunkApiConfig for optional type arguments (@phryneas , @Ethan-Arrowood - #364)
v1.3.0-alpha.4
This alpha release rearranges the TS generic types of createAsyncThunk
to fix broken usage.
Changes
createAsyncThunk
Types
createAsyncThunk
gives you access to getState
in the thunkAPI
object argument. However, we hadn't tried to exercise that yet in our tests, and it turns out there was no valid way to specify the correct type of the state returned by getState
.
We've rearranged the generic types and tweaked the defaults. You should now be able to use it without specifying any generics for the most common case (returning a value, with a potential argument for the payload callback), and specify three types if you need to declare what the state type is:
// basic usage:
const thunk1 = createAsyncThunk("a", async (arg: string) => {
return 42
})
// infers: return = Promise<number>, arg = string
// specify state type for getState usage
const thunk2 = createAsyncThunk<Promise<number>, string, RootState>(
"a",
async (arg: string, {getState}) => {
const state = getState();
return 42;
}
)
// declared: return = Promise<number>, arg = string, state: RootState
We have some ideas for additional potential improvements to these types that may make usage simpler, so please keep an eye out for further alpha releases.
Documentation
A first draft of API documentation for createAsyncThunk
is now available:
createAsyncThunk
draft API reference
Changes
v1.3.0-alpha.3...v1.3.0-alpha.4
See PR #352: Port ngrx/entity and add createAsyncThunk for the complete alpha changes.
v1.3.0-alpha.3
This alpha release alters the error-handling behavior of createAsyncThunk
(again! 😁 )
Changes
createAsyncThunk
Error Handling
In alpha.2
, we tried having the thunk always re-throw caught errors. That didn't work out, because they always show up in the browser's console as unhandled exceptions.
Instead, the thunk now always returns a resolved promise containing the last action dispatched, which will be either the fulfilled
action or the rejected
action. We also export an unwrapAction
utility that will either return action.payload
if fulfilled or throw action.error
if rejected, allowing the dispatching code to chain off the promise if desired:
import { unwrapResult } from '@reduxjs/toolkit'
// in the component
const onClick = () => {
dispatch(fetchUserById(userId))
.then(unwrapResult)
.then(originalPromiseResult => {})
.catch(serializedError => {})
}
Aborting Requests
Since createAsyncThunk
accepts a promise callback, we have no way of knowing if or how the async logic can be canceled. However, we now provide a way for your logic to signal that cancellation needs to occur. An AbortController
instance will be created for each request, and abort
method will be attached to the returned promise that calls abortController.abort()
and rejects the promise. The corresponding abortController.signal
object will be passed in to your payload creator callback in the thunkAPI
object as thunkAPI.signal
, allowing your async logic to check if a cancellation has been requested.
Meta Field Names
The action.meta
field in each action object previously looked like: {args, requestId}
The args
field was renamed to arg
to indicate it's only one value.
For the pending
and fulfilled
actions, action.meta
is now: {arg, requestId}
The rejected
action now includes details on whether the request was aborted, and
action.meta
is now: {arg, requestId, aborted, abortReason}
Documentation
A first draft of API documentation for createAsyncThunk
is now available:
createAsyncThunk
draft API reference
Changes
v1.3.0-alpha.2...v1.3.0-alpha.3
See PR #352: Port ngrx/entity and add createAsyncThunk for the complete alpha changes.
v1.3.0-alpha.2
This alpha release alters the error-handling behavior of createAsyncThunk
.
Changes
createAsyncThunk
Error Handling
createAsyncThunk
catches rejected promises from the promise payload callback, and dispatches a "rejected"
action in response with the error value. That means that the thunk itself always returns a resolved promise.
We had a request to re-throw errors from the thunk, in case the calling code wants to chain off the promise or handle it further, so we've implemented that.
We've also removed the "finished"
action that was previously being dispatched in a finally {}
clause at the end, as it shouldn't truly be necessary - app logic should just need to respond to either the "fulfilled"
or "rejected"
actions.
When an error is caught in the thunk, we try to put it into the "rejected"
action. But, since JS Error
objects aren't actually serializable, we now check for Error
objects and serialize them into plain JS objects with any of the relevant fields ( {name?, message?, stack?, code?}
), or just the value itself if it's not an Error
. Unfortunately, since the err
value in a catch(err) {}
clause doesn't have a type, the action.error
field will come through as a type of any
either way.
Finally, we've reordered the logic inside to avoid cases where an error while dispatching "success" gets swallowed and triggers the "AJAX failed" handling.