I'm new to React and still struggling in understanding in using redux with React.
For example, below is some code :
const mapStateToProps = (storeData) => ({
editing: storeData.isEditMode;
})
const mapDispatchToProps = {
saveCallback: xxx
}
const connectFunction = connect(mapStateToProps, mapDispatchToProps);
export const ProductDisplay = connectFunction(
class extends Component {
render() {
if (this.props.editing) {
...
} else {
...
}
}
}
)
and a component that uses ProductDisplay
export default class App extends Component {
render() {
return <ProductDisplay/>
}
}
Let's say storeData.isEditMode is changed by other component, so the wrapped component's props.editing is changed. As we know that the only way to trigger update process is to use setState() but how does react know that ProductDisplay component needs to be updated since there is no setState() method involved?
The connect function generates a wrapper component that subscribes to the store. When an action is dispatched, the wrapper component's callback is notified. It then runs your mapState function, and shallow-compares the result object from this time vs the result object from last time (so if you were to rewrite a redux store field with its same value, it would not trigger a re-render). If the results are different, then it passes the results to your "real" component" as props.
Dan Abramov wrote a great simplified version of connect at (connect.js) that illustrates the basic idea, although it doesn't show any of the optimization work.
update
React-Redux v6.0.0 made some major internal changes to how connected components receive their data from the store.
For more details: https://spin.atomicobject.com/2018/04/02/redux-rerendering/
I have an app where after login, user details are acquired and some user profiles will have a metric setting and others will have a non-metric (imperial).
I want to set a boolean (isMetric) in React Context as there are stateless components all over that need to reference it and I want to avoid passing an isMetric prop all through the component hierarchy.
I've been following the React 16 examples at https://reactjs.org/docs/context.html and so I have a unit-context.js of:
export const UnitContext = React.createContext({
isMetric: true,
toggleUnits: () => {},
});
and an App.js of
import { UnitContext } from './unit-context';
class App extends React.Component {
constructor(props) {
super(props);
this.toggleUnits = () => {
this.setState(state => ({
isMetric: !state.isMetric
}));
};
this.state = {
isMetric: true,
toggleUnits: this.toggleUnits,
};
}
render() {
return (
<UnitContext.Provider value={this.state}>
< ... />
</UnitContext.Provider>
);
}
}
The above is the dynamic pattern that provides toggleUnits in the context itself so that state (context) can be set.
With the above pattern, I've confirmed that in my components I can access the isMetric value fine.
But I use redux-saga to handle api calls and it is retrieving the user's metric / non-metric setting.
Once I've acquired the user details, how can I set Context? (redux-saga seems to have it's own context aspect but I'm referring to React Context).
One option is to import the UnitContext with
import { UnitContext } from '.unit-context';
and then access the Consumer to get the state and thus the function. But I'm unsure how to reference what would normally be a Component in a redux-saga saga which is imperative code.
Alternatively, could I move the App state to redux? That seems odd and unsound to have a function in a redux store. I know that redux-saga provides a select method to access redux store via selectors and I have used that in the past.
Update:
As the context is provided by <UnitContext.Provider value={this.state}> then conceivably I could replace that value property setting with a redux store value instead of using local state.
On my React + Redux client app, I need to get the active user info (fetch from myapp.com/users/me) and keep it somewhere so that I can access it from multiple components.
I guess window.activeUser = data would not be the best practice. But I could not find any resource about the best practice of doing that. What would be the best way to do what I want?
you can keep it in a separate reducer, and then import multiple parts of your state with connect() in your components.
Say if you have 2 reducers called users.js and tags.js which are combined with combineReducers when setting up your store. You would simply pull different parts by passing a function to your connect() call. So using es6 + decorators:
const mapStateToProps = state => {
return {users: state.users, tags: state.tags}
}
#connect(mapStateToProps)
export default class MyComponent extends React.Component {
and then down in your render function:
return (
<div>
<p>{this.props.users.activeUsernameOrWhatever}</p>
<p>{this.props.tags.activeTags.join('|')}</p>
</div>
);
So your different reducer states become objects on this.props.
You can use React's context, HOC or global variables to make your data available to multiple components, via context
Something like...
class ParentDataComponent extends React.Component {
// make data accessible for children
getChildContext() {
return {
data: "your-fetchet-data"
};
}
render() {
return < Child />
}
}
class Child extends React.Component {
render() {
// access data from Parent's context
return (<div> {this.context.data} </div>);
}
}
create an action creator and call it on componentWillMount of the appropriate component so it runs right before your component mounts, and in the action creator fetch the data you need and pass it to a reducer. in that reducer you can keep the data you want throughout your application. so whenever you needed the data you can retrieve it from redux state. this tutorial from official redux website covers everything you need to know. mention me if you had any questions.
I've come across a repeated pattern in my react-redux site:
A component displays data from a web api, and it should be populated on load, automatically, without any user interaction.
I want to initiate the async fetch from a container component, but as far as I can tell the only way to do it is from a lifecycle event in a display component. This seems to make it impossible to put all the logic in the container and only use dumb stateless functional components for display.
This means I can't use a stateless functional component for any component that needs async data. That doesn't seem right.
It seems like the "right" way to do this would be to somehow initiate async calls from the container. Then when the call returned, the state would be updated and the container would get the new state and would in turn pass those to its stateless component via mapStateToProps().
Making async calls in mapStateToProps and mapDispatchToProps (I mean actually calling the async function, as opposed to returning it as a property) doesn't make sense.
So what I've ended up doing is putting the async call(s) in a refreshData() function exposed by mapDispatchToProps(), then calling it from two or more of the React lifecycle methods: componentDidMount and componentWillReceiveProps.
Is there a clean way to update the redux store state without putting lifecycle method calls in every component that needs async data?
Should I be making these calls higher up the component heierarchy (thereby reducing the scope of this issue, since only "top-level" components would need to listen to lifecycle events)?
Edit:
Just so there's no confusion what I mean by a connect()ed container component, here's a very simple example:
import React from 'react';
import { connect } from 'react-redux';
import {action} from './actions.js';
import MyDumbComponent from './myDumbComponent.jsx';
function mapStateToProps(state)
{
return { something: state.xxxreducer.something };
}
function mapDispatchToProps(dispatch)
{
return {
doAction: ()=>{dispatch(action())}
};
}
const MyDumbComponentContainer = connect(
mapStateToProps,
mapDispatchToProps
)(MyDumbComponent);
// Uh... how can I hook into to componentDidMount()? This isn't
// a normal React class.
export default MyDumbComponentContainer;
Jamie Dixon has written a package to do this!
https://github.com/JamieDixon/react-lifecycle-component
Usage would look like this:
const mapDispatchToProps = {
componentDidMount: getAllTehDatas
}
...
export default connectWithLifecycle(mapStateToProps, mapDispatchToProps)(WrappedComponent)
edit
With hooks you are now able to implement lifecycle callbacks in a stateless functional component. While this may not directly address all of the points in the question, it may also get around some of the reason for wanting to do what was originally proposed.
edit to original answer
After the discussion in comments and thinking about it more, this answer is more exploratory and can serve as a piece of the conversation. But I don't think it's the right answer.
original answer
On the Redux site there's an example that shows you don't have to do both mapStateToProps and mapDispatchToProps. You can just leverage connect's awesomeness for the props, and use a class and implement the lifecycle methods on the dumb component.
In the example, the connect call is even in the same file and the dumb component isn't even exported, so to the user of the component it's looks the same.
I can understand not wanting to issue async calls from the display component. I think there's a distinction between issuing the async calls from there and dispatching an action which, with thunks, moves the issuing of the async calls up into the actions (even more decoupled from the React code).
As an example, here's a splash screen component where I'd like to do some async action (like asset preloading) when the display component mounts:
SplashContainer.js
import { connect } from 'react-redux'
import Splash from '../components/Splash'
import * as actions from '../actions'
const mapStateToProps = (state) => {
return {
// whatever you need here
}
}
const mapDispatchToProps = (dispatch) => {
return {
onMount: () => dispatch(actions.splashMount())
}
}
const SceneSplash = connect(
mapStateToProps,
mapDispatchToProps
)(Splash)
export default SceneSplash
Splash.js
import React from 'react'
class Splash extends React.Component {
render() {
return (
<div className="scene splash">
<span className="fa fa-gear fa-spin"></span>
</div>
)
}
componentDidMount() {
const { onMount } = this.props
onMount()
}
}
export default Splash
You can see the the dispatch happens in the connected container, and you can imagine in the actions.splashMount() call we issue an async http request or do other async things via thunks or promises.
edit to clarify
Allow me to try to defend the approach. I re-read the question and am not 100% sure I'm addressing the main thing it's after, but bear with me. If I am still not quite on track, I have a modified approach below that may be closer to the mark.
"it should be populated on load" - the example above accomplishes this
"I want to initiate the async fetch from a container" - in the example it's not initiated from the display component or the container, but from an async action
"This seems to make it impossible to put all the logic in the container" - I think you can still put any additional logic needed in the container. As noted, the data loading code isn't in the display component (or the container) but in the async action creator.
"This means I can't use a stateless functional component for any component that needs async data." - in the above example the display component is stateless and functional. The only link is the lifecycle method invoking a callback. It need not know or care what that callback does. It is not a case of the display component trying to be the owner of async data fetching - it's merely letting the code that does handle that know when a particular thing has happened.
So far I'm attempting to justify how the example given meets the question's requirements. That said, if what you're after is having a display component that contains absolutely no code related to the async data load even by indirect callbacks - that is, the only link it has is to consume that data via the props it's handed when that remote data comes down, then I would suggest something like this:
SplashContainer.js
import { connect } from 'react-redux'
import Splash from '../components/Splash'
import * as actions from '../actions'
const mapStateToProps = (state) => {
return {
// whatever you need here
}
}
const mapDispatchToProps = (dispatch) => {
dispatch(actions.splashMount())
return {
// whatever else here may be needed
}
}
const SceneSplash = connect(
mapStateToProps,
mapDispatchToProps
)(Splash)
export default SceneSplash
Splash.js
import React from 'react'
class Splash extends React.Component {
// incorporate any this.props references here as desired
render() {
return (
<div className="scene splash">
<span className="fa fa-gear fa-spin"></span>
</div>
)
}
}
export default Splash
By dispatching the action in mapDispatchToProps you are letting the code for that action reside entirely in the container. In fact, you are starting the async call as soon as the container is instantiated, rather than waiting for the connected display component to spin up and be mounted. However, if you cannot begin the async call until the componentDidMount() for the display component fires, I think you're inherently bound to have code like in my first example.
I haven't actually tested this second approach to see if react or redux will complain about it, but it should work. You have access to the dispatch method and should be able to call it without problems.
To be honest, this second example, while removing all code related to the async action from the display component does kind of strike me as a bit funny since we're doing non-mapping-of-dispatch-to-props things in the eponymous function. And containers don't actually have a componentDidMount to run it in otherwise. So I'm a bit squirmy with it and would lean toward the first approach. It's not clean in the "feels right" sense, but it is in the "simple 1-liner" sense.
Check out redux-saga https://github.com/yelouafi/redux-saga. It's a redux middleware component that creates long-lived watchers that look for specific store actions and can trigger functions or generator functions in response. The generator syntax is particular nice for handling async, and redux-saga has some nice helpers that allow you to treat async code in a synchronous fashion. See some of their examples. https://github.com/yelouafi/redux-saga/blob/master/examples/async/src/sagas/index.js . The generator syntax can be hard to grok at first, but based on our experience this syntax supports extremely complex async logic, including debounce, cancellation and joining/racing multiple requests.
You can do it from a container. Just make a component that extends React.Component but name it with "Container" somewhere in the name. Then use that container's componentDidMount instead of using componentDidMount in the presentational (dumb) component that the container component renders. Reducer will see that you've dispatched an action still, and still update state so your dumb component will be able to get at that data..
I TDD but even if I didn't TDD I separate out my dumb vs container components via file. I hate having too much in one file, especially if mixing dumb vs. container stuff in the same file, that's a mess. I know people do it, but I think that's awful.
I do this:
src/components/someDomainFolder/someComponent.js (dumb component)
src/components/someDomainFolder/someComponentContainer.js (for example you might use React-Redux..have connected container not a connected presentational component..and so in someComponentContainer.js you DO have a react class in this file, as stated, just call it someComponentContainer extends React.Component for example.
Your mapStateToProps() and mapDispatchToProps() would be global functions of that connected container component outside that container class. And connect() would render the container, which would render the presentational component but this allows you to keep all your behavior in your container file, away from dumb presentational component code.
that way you have tests around someComponent that are structural/state based and you have behavior tests around the Container component. Much better route to be going with maintaining and writing tests, and maintaining and making things easy for yourself or other devs to see what's going on and to manage dumb vs. behavioral components.
Doing things this way, your presentational stuff is separated both physically by file AND by code convention. AND your tests are grouped around the right areas of code...not an intermingled mess. AND if you do this, and use a reducer that listens to update state, your presentational component can remain totally stupid....and just look for that updates state via props...since you're using mapStateToProps().
Following up on #PositiveGuy suggestion, here is example code of how to implement a container component that can utilize lifecycle methods. I think this is a pretty clean approach that maintains separation of concerns keeping the presentation component "dumb":
import React from 'react';
import { connect } from 'react-redux'
import { myAction } from './actions/my_action_creator'
import MyPresentationComponent from './my_presentation_component'
const mapStateToProps = state => {
return {
myStateSlice: state.myStateSlice
}
}
const mapDispatchToProps = dispatch => {
return {
myAction: () => {
dispatch(myAction())
}
}
}
class Container extends React.Component {
componentDidMount() {
//You have lifecycle access now!!
}
render() {
return(
<MyPresentationComponent
myStateSlice={this.props.myStateSlice}
myAction={this.props.myAction}
/>
)
}
}
const ContainerComponent = connect(
mapStateToProps,
mapDispatchToProps
)(Container)
export default ContainerComponent
You CAN initiate an async fetch from the parent container (the smart container). You write the function in the smart container and you pass the function as a prop for the dumb container. For example:
var Parent = React.createClass({
onClick: function(){
dispatch(myAsyncAction());
},
render: function() {
return <childComp onClick={this.onClick} />;
}
});
var childComp = React.createClass({
propTypes:{
onClick: React.PropTypes.func
},
render: function() {
return <Button onClick={this.props.onClick}>Click me</Button>;
}
});
childComp is stateless since the onClick definition is determined by the parent.
EDIT: Added connected container example below, excluded other stuff for brevity. Doesn't really show much, and it's a bit cumbersome to setup on fiddle and stuff, point is, I do use lifecycle methods in connected containers and it works out fine for me.
class cntWorkloadChart extends Component {
...
componentWillReceiveProps(nextProps){
if(nextProps.myStuff.isData){
if (nextProps.myStuff.isResized) {
this.onResizeEnd();
}
let temp = this.updatePrintingData(nextProps)
this.selectedFilterData = temp.selectedFilterData;
this.selectedProjects = temp.selectedProjects;
let data = nextProps.workloadData.toArray();
let spread = [];
if(nextProps.myStuff.isSpread) {
spread = this.updateSelectedProjectSpread(nextProps);
for (var i = 0; i < data.length; i++) {
data[i].sumBillableHrsSelectedProjects = spread[data[i].weekCode] ? Number(spread[data[i].weekCode].sumBillableHrsSelectedProjects.toFixed(1)) : 0;
data[i].sumCurrentBudgetHrsSelectedProjects = spread[data[i].weekCode] ? Number(spread[data[i].weekCode].sumCurrentBudgetHrsSelectedProjects.toFixed(1)) : 0;
data[i].sumHistoricBudgetHrsSelectedProjects = spread[data[i].weekCode] ? Number(spread[data[i].weekCode].sumHistoricBudgetHrsSelectedProjects.toFixed(1)) : 0;
}
}
if (nextProps.potentialProjectSpread.length || this.props.potentialProjectSpread.length) { //nextProps.myStuff.isPpSpread) { ???? - that was undefined
let potential = nextProps.potentialProjectSpread;
let ppdd = _.indexBy(potential, 'weekCode');
for (var i = 0; i < data.length; i++) {
data[i].sumSelectedPotentialProjects = ppdd[data[i].weekCode] ? ppdd[data[i].weekCode].sumSelectedPotentialProjects.toFixed(1) : 0;
}
}
for (var i = 0; i < data.length; i++) {
let currObj = data[i];
currObj.sumCurrentBudgetHrs = currObj.currentBudgeted.sumWeekHours;
currObj.sumHistoricBudgetHrs = currObj.historicBudgeted.sumWeekHours;
currObj.fillAlpha = .6; //Default to .6 before any selections are made
//RMW-TODO: Perhaps we should update ALL line colors this way? This would clean up zero total bars in all places
this.updateLineColor(currObj, "sumSelectedPotentialProjects", "potentialLineColor", potentialLineColor);
this.updateLineColor(currObj, "sumHistoricBudgetHrs", "histLineColor", histBudgetLineColor);
this.updateLineColor(currObj, "sumHistoricBudgetHrsSelectedProjects", "histSelectedLineColor", selectedHistBudgetFillColor);
}
if(nextProps.myStuff.isSelectedWeek){
let currWeekIndex = nextProps.weekIndex.index;
let selectedWeek = data[currWeekIndex].fillAlpha = 1.0;
}
if(data.length > 0){
if(data[0].targetLinePercentages && data.length > 9) { //there are target lines and more than 10 items in the dataset
let tlHigh = data[0].targetLinePercentages.targetLineHigh;
let tlLow = data[0].targetLinePercentages.targetLineLow;
if (tlHigh > 0 && tlLow > 0) {
this.addTargetLineGraph = true;
this.upperTarget = tlHigh;
this.lowerTarget = tlLow;
}
}
else {
this.addTargetLineGraph = false;
this.upperTarget = null;
this.lowerTarget = null;
}
}
this.data = this.transformStoreData(data);
this.containsHistorical = nextProps.workloadData.some(currObj=> currObj.historicBudgeted.projectDetails.length);
}
}
...
render() {
return (
<div id="chartContainer" className="container">
<WorkloadChart workloadData={this.props.workloadData}
onClick={this.onClick}
onResizeEnd={this.onResizeEnd}
weekIndex={this.props.weekIndex}
getChartReference={this.getChartReference}
//projectSpread={this.props.projectSpread}
selectedRows={this.props.selectedRows}
potentialProjectSpread={this.props.potentialProjectSpread}
selectedCompany={this.props.selectedCompany}
cascadeFilters={this.props.cascadeFilters}
selectedRows={this.props.selectedRows}
resized={this.props.resized}
selectedFilterData={this.selectedFilterData}
selectedProjects={this.selectedProjects}
data={this.data}
upperTarget={this.upperTarget}
lowerTarget={this.lowerTarget}
containsHistorical={this.containsHistorical}
addTargetLineGraph={this.addTargetLineGraph}
/>
</div>
);
}
};
function mapStateToProps(state){
let myValues = getChartValues(state);
return {
myStuff: myValues,
workloadData: state.chartData || new Immutable.List(),
weekIndex: state.weekIndex || null,
//projectSpread: state.projectSpread || {},
selectedRows: state.selectedRows || [],
potentialProjectSpread: state.potentialProjectSpread || [],
selectedCompany: state.companyFilter.selectedItems || null,
brokenOutByCompany: state.workloadGrid.brokenOutByCompany || false,
gridSortName: state.projectGridSort.name,
gridSortOrder: state.projectGridSort.order,
cascadeFilters: state.cascadeFilters || null,
selectedRows: state.selectedRows || [],
resized: state.chartResized || false,
selectedPotentialProjects: state.selectedPotentialProjects || []
};
}
module.exports = connect(mapStateToProps)(cntWorkloadChart);
Imagine you have a relatively simple component you create as part of a component library (simplified for brevity):
class ExampleComponent extends React.Component {
componentDidMount() {
getAsyncData().then((response) => {
const {a} = response.data;
this.setState({a});
this.props.notify({a});
});
}
render() {
return (
<h1>{this.state.a}</h1>
);
}
}
The component is required to allow dropping it into an application (think Google Maps for relatively similar approach) and have it just work. It can, however, share its data from response with the rest of the application, via some sort of callback (see this.props.notify above) it may receive via its props. This is an actual requirement and not a architectural choice.
Since this is a part of a library - you don't know what kind of application it is going to get used in at all times, but you do know that in many many cases it is going to get used in a Redux application.
For Redux application the above self-contained approach is not necessarily the best - as the retrieved data in response is better kept in application state in Redux store, where it can be consumed by other parts of application.
Even more so - the ExampleComponent itself is better off being "passive" and not having state at all, rather using mapStateToProps to have Redux infrastructure inject the state update into it via props.
The idea is that when ExampleComponent is in Redux application - its setState call and reference to this.state in its render method are somehow abstracted and "re-routed" to props via Redux?
One way would be to make ExampleComponent to use dispatch that by default calls setState and can be overridden by injected Redux dispatch - basically take this to Redux:
class ExampleComponent extends React.Component {
constructor() {
super();
this.dispatch = this.props.dispatch || this.dispatch;
}
componentDidMount() {
getAsyncData().then((response) => {
this.dispatch({type: 'SOME_ACTION', data: response.data});
});
}
dispatch(action) {
swtich (action.type) {
case 'SOME_ACTION':
const {a} = action.data;
this.setState({a});
case 'ANOTHER_ACTION': ...
}
}
render() {
return (
<h1>{this.state.a}</h1>
);
}
}
The above example works very well, save for:
this.state.a and its kin being sprinkled around the code whereas in Redux it should be this.props.state
having to do this.dispatch = this.props.dispatch || this.dispatch; in every component
I would like to avoid the obvious BaseComponent solutions that would abstract setState into some kind of hybrid... as this would take the code, with time, further away from "canonical" React.
Do you see an elegant way where the two approaches can be combined, with Redux superseding the inherent one?
You're making a fundamental mistake in thinking that a React component with Redux is different from a React component without Redux.
In fact, a React component is just a React component.
This is all your component needs to look like:
function ExampleComponent({ a }) {
return (
<h1>{a}</h1>
);
}
Simple, clean, readable, testable.
There's no obvious reason why your asynchronous data fetch should be buried inside the component's componentDidMount() method. It can be triggered anywhere else in the application. And it should be.