[Angular] ChangeDetection -- onPush

To understand how change detection can help us improve the proference, we need to understand when it works first.

There are some rules which can be applied when use change detection:

changeDetection: ChangeDetectionStrategy.OnPush

 

1. Change detection compares @Input value, so applied for dump components

Mostly change detection will be applied for dump component not smart component. Because if the data is getting from service, then change detection won't work.

 

<ul class="message-list" #list>
  <li class="message-list-item" *ngFor="let message of messages">
    <message [message]="message"></message>
  </li>
</ul>

For this code, <message> is a dump component:

复制代码
@Component({
  selector: 'message',
  templateUrl: './message.component.html',
  styleUrls: ['./message.component.css'],
  changeDetection: ChangeDetectionStrategy.OnPush
})
export class MessageComponent {
  @Input() message: MessageVM;
}
复制代码

 

 

2. Reducer: If the data is getting from the 'store' (ngrx/store), then you need to be careful about how to write your reducer. We should keep AppState immutable and reuseable as much as possible.

For example:

复制代码
function newMessagesReceivedAction(state: StoreData, action: NewMessagesReceivedAction) {
  const cloneState = cloneDeep(state);
  const newMessages = action.payload.unreadMessages,
    currentThreadId = action.payload.currentThreadId,
    currentUserId = action.payload.currentUserId;

  newMessages.forEach(message => {
    cloneState.messages[message.id] = message;
    cloneState.threads[message.threadId].messageIds.push(message.id);

    if(message.threadId !== currentThreadId) {
      cloneState.threads[message.threadId].participants[currentUserId] += 1;
    }
  });

  return cloneState;
}
复制代码
复制代码
export interface StoreData {
  participants: {
    [key: number]: Participant
  };
  threads: {
    [key: number]: Thread
  };
  messages: {
    [key: number]: Message
  };
}
复制代码

As we can see that, the 'state' is implements 'StoreData' interface.

We did a deep clone of current state:

const cloneState = cloneDeep(state);

new every props in StateData interface will get a new reference. But this is not necessary, because in the code, we only modify 'messages' & 'threads' props, but not 'participants'.

Therefore it means we don't need to do a deep clone for all the props, so we can do:

复制代码
function newMessagesReceivedAction(state: StoreData, action: NewMessagesReceivedAction) {
  const cloneState = {
    participants: state.participants, // no need to update this, since it won't change from here
    threads: Object.assign({}, state.threads),
    messages: Object.assign({}, state.messages)
  };
  const newMessages = action.payload.unreadMessages,
    currentThreadId = action.payload.currentThreadId,
    currentUserId = action.payload.currentUserId;

  newMessages.forEach(message => {
    cloneState.messages[message.id] = message;

    // First clone 'cloneState.threads[message.threadId]',
    // create a new reference
    cloneState.threads[message.threadId] =
      Object.assign({}, state.threads[message.threadId]);

    // Then assign new reference to new variable
    const messageThread = cloneState.threads[message.threadId];

    messageThread.messageIds = [
      ...messageThread.messageIds,
      message.id
    ];

    if (message.threadId !== currentThreadId) {
      messageThread.participants = Object.assign({}, messageThread.participants);
      messageThread.participants[currentUserId] += 1;
    }
  });

  return cloneState;
}
复制代码

So in the updated code, we didn't do a deep clone, instead, we using Object.assign() to do a shadow clone, but only for 'messages' & 'threads'.

  const cloneState = {
    participants: state.participants, // no need to update this, since it won't change from here
    threads: Object.assign({}, state.threads),
    messages: Object.assign({}, state.messages)
  };

And BE CAREFUL here, since we use Object.assign, what it dose is just a shadow copy, if we still do:

cloneState.messages[message.id] = message;

It actually modify the origial state, instead what we should do is do a shadow copy of 'state.messages', then modify the value based on new messages clone object:

    // First clone 'cloneState.threads[message.threadId]',
    // create a new reference
    cloneState.threads[message.threadId] =
      Object.assign({}, state.threads[message.threadId]);

...

 

3. Selector: Using memoization to remember previous selector's data.

But only 1 & 2 are still not enough for Change Detection. Because the application state is what we get from BE, it is good to keep it immutable and reuse the old object reference as much as possible, but what we pass into component are not Application state, it is View model state. This will cause the whole list be re-render, if we set time interval 3s, to fetch new messages. 

For example we smart component:

复制代码
@Component({
  selector: 'message-section',
  templateUrl: './message-section.component.html',
  styleUrls: ['./message-section.component.css']
})
export class MessageSectionComponent {

  participantNames$: Observable<string>;
  messages$: Observable<MessageVM[]>;
  uiState: UiState;

  constructor(private store: Store<AppState>) {
    this.participantNames$ = store.select(this.participantNamesSelector);
    this.messages$ = store.select(this.messageSelector.bind(this));
    store.subscribe(state => this.uiState = Object.assign({}, state.uiState));
  }

...

}
复制代码

Event the reducers data is immutable, but everytime we actually receive a new 'message$' which is Message view model, not the state model.

export interface MessageVM {
  id: number;
  text: string;
  participantName: string;
  timestamp: number;
}

And for view model:

复制代码
  messageSelector(state: AppState): MessageVM[] {
    const {currentSelectedID} = state.uiState;
    if (!currentSelectedID) {
      return [];
    }
    const messageIds = state.storeData.threads[currentSelectedID].messageIds;
    const messages = messageIds.map(id => state.storeData.messages[id]);
    return messages.map((message) => this.mapMessageToMessageVM(message, state));
  }

  mapMessageToMessageVM(message, state): MessageVM {
    return {
      id: message.id,
      text: message.text,
      participantName: (state.storeData.participants[message.participantId].name || ''),
      timestamp: message.timestamp
    }
  }
复制代码

As we can see, everytime it map to a new message view model, but this is not what we want, in the mssages list component:

First, we don't want the whole message list been re-render every 3s. Because there is no new data come in. But becaseu we everytime create a new view model, the list is actually re-rendered. To prevent that, we need to update our selector code and using memoization to do it.

Install:

npm i --save reselect 
复制代码
import {createSelector} from 'reselect';
/*
export const messageSelector = (state: AppState): MessageVM[] => {
  const messages = _getMessagesFromCurrentThread(state);
  const participants = _getParticipants(state);
  return _mapMessagesToMessageVM(messages, participants);
};*/

export const messageSelector = createSelector(
  _getMessagesFromCurrentThread,
  _getParticipants,
  _mapMessagesToMessageVM
);
复制代码
复制代码
function _getMessagesFromCurrentThread(state: AppState): Message[] {
  const {currentSelectedID} = state.uiState;
  if(!currentSelectedID) {
    return [];
  }
  const currentThread = state.storeData.threads[currentSelectedID];
  return currentThread.messageIds.map(msgId => state.storeData.messages[msgId])
}

function _getParticipants(state: AppState): {[key: number]: Participant} {
  return state.storeData.participants;
}

function _mapMessagesToMessageVM(messages: Message[] = [], participants) {
  return messages.map((message) => _mapMessageToMessageVM(message, participants));
}

function _mapMessageToMessageVM(message: Message, participants: {[key: number]: Participant}): MessageVM {
  return {
    id: message.id,
    text: message.text,
    participantName: (participants[message.participantId].name || ''),
    timestamp: message.timestamp
  }
}
复制代码

'createSelector' function takes getters methods and one mapping function. The advantage to using 'createSelector' is that it can help to memoizate the data, if the input are the same, then output will be the same (take out from memory, not need to calculate again.) It means:

  _getMessagesFromCurrentThread,
  _getParticipants,

only when '_getMessagesFromCurrentThread' and '_getParticipants' outputs different result, then the function '_mapMessagesToMessageVM' will be run.

This can help to prevent the message list be rerendered each three seconds if there is no new message come in.

But this still not help if new message come in, only render the new message, not the whole list re-render. We still need to apply rule No.4 .

 

4. lodash--> memoize: Prevent the whole list of messages been re-rendered when new message come in.

复制代码
function _mapMessagesToMessageVM(messages: Message[] = [], participants: {[key: number]: Participant}) {
  return messages.map((message) => {
    const participantNames = participants[message.participantId].name || '';
    return _mapMessageToMessageVM(message, participantNames);
  });
}

const _mapMessageToMessageVM = memoize((message: Message, participantName: string): MessageVM => {
  return {
    id: message.id,
    text: message.text,
    participantName: participantName,
    timestamp: message.timestamp
  }
}, (message, participantName) => message.id + participantName);
复制代码

Now if new message come in, only new message will be rendered to the list, the existing message won't be re-rendered.

 

Github

posted @   Zhentiw  阅读(1371)  评论(0编辑  收藏  举报
编辑推荐:
· SQL Server 2025 AI相关能力初探
· Linux系列:如何用 C#调用 C方法造成内存泄露
· AI与.NET技术实操系列(二):开始使用ML.NET
· 记一次.NET内存居高不下排查解决与启示
· 探究高空视频全景AR技术的实现原理
阅读排行:
· 阿里最新开源QwQ-32B,效果媲美deepseek-r1满血版,部署成本又又又降低了!
· Manus重磅发布:全球首款通用AI代理技术深度解析与实战指南
· 开源Multi-agent AI智能体框架aevatar.ai,欢迎大家贡献代码
· 被坑几百块钱后,我竟然真的恢复了删除的微信聊天记录!
· AI技术革命,工作效率10个最佳AI工具
历史上的今天:
2016-01-31 [Javascript] Redirect the browser using JavaScript
点击右上角即可分享
微信分享提示