Skip to main content

Electron 中的消息端口

¥MessagePorts in Electron

MessagePort 是一项 Web 功能,允许在不同上下文之间传递消息。就像 window.postMessage,但是在不同的通道上。本文档的目标是描述 Electron 如何扩展 Channel Messaging 模型,并提供一些示例来说明如何在应用中使用 MessagePorts。

¥MessagePorts are a web feature that allow passing messages between different contexts. It's like window.postMessage, but on different channels. The goal of this document is to describe how Electron extends the Channel Messaging model, and to give some examples of how you might use MessagePorts in your app.

下面是一个关于 MessagePort 是什么及其工作原理的非常简短的示例:

¥Here is a very brief example of what a MessagePort is and how it works:

renderer.js (Renderer Process)
// MessagePorts are created in pairs. A connected pair of message ports is
// called a channel.
const channel = new MessageChannel()

// The only difference between port1 and port2 is in how you use them. Messages
// sent to port1 will be received by port2 and vice-versa.
const port1 = channel.port1
const port2 = channel.port2

// It's OK to send a message on the channel before the other end has registered
// a listener. Messages will be queued until a listener is registered.
port2.postMessage({ answer: 42 })

// Here we send the other end of the channel, port1, to the main process. It's
// also possible to send MessagePorts to other frames, or to Web Workers, etc.
ipcRenderer.postMessage('port', null, [port1])
main.js (Main Process)
// In the main process, we receive the port.
ipcMain.on('port', (event) => {
// When we receive a MessagePort in the main process, it becomes a
// MessagePortMain.
const port = event.ports[0]

// MessagePortMain uses the Node.js-style events API, rather than the
// web-style events API. So .on('message', ...) instead of .onmessage = ...
port.on('message', (event) => {
// data is { answer: 42 }
const data = event.data
})

// MessagePortMain queues messages until the .start() method has been called.
port.start()
})

渠道消息 API 文档是了解有关 MessagePorts 如何工作的更多信息的好方法。

¥The Channel Messaging API documentation is a great way to learn more about how MessagePorts work.

主进程中的消息端口

¥MessagePorts in the main process

在渲染器中,MessagePort 类的行为与在 Web 上的行为完全相同。不过,主进程不是网页,它没有 Blink 集成,因此它没有 MessagePortMessageChannel 类。为了在主进程中处理 MessagePorts 并与之交互,Electron 添加了两个新类:MessagePortMainMessageChannelMain。它们的行为与渲染器中的类似类类似。

¥In the renderer, the MessagePort class behaves exactly as it does on the web. The main process is not a web page, though—it has no Blink integration—and so it does not have the MessagePort or MessageChannel classes. In order to handle and interact with MessagePorts in the main process, Electron adds two new classes: MessagePortMain and MessageChannelMain. These behave similarly to the analogous classes in the renderer.

MessagePort 对象可以在渲染器或主进程中创建,并使用 ipcRenderer.postMessageWebContents.postMessage 方法来回传递。注意,常用的 IPC 方法如 sendinvoke 不能用于传输 MessagePort,只有 postMessage 方法可以传输 MessagePort

¥MessagePort objects can be created in either the renderer or the main process, and passed back and forth using the ipcRenderer.postMessage and WebContents.postMessage methods. Note that the usual IPC methods like send and invoke cannot be used to transfer MessagePorts, only the postMessage methods can transfer MessagePorts.

通过通过主进程传递 MessagePort,你可以连接两个可能无法通信的页面(例如由于同源限制)。

¥By passing MessagePorts via the main process, you can connect two pages that might not otherwise be able to communicate (e.g. due to same-origin restrictions).

扩展:close 事件

¥Extension: close event

Electron 向 MessagePort 添加了一项 Web 上不存在的功能,以使 MessagePorts 更加有用。这就是 close 事件,当通道的另一端关闭时会发出该事件。端口也可以通过垃圾收集来隐式关闭。

¥Electron adds one feature to MessagePort that isn't present on the web, in order to make MessagePorts more useful. That is the close event, which is emitted when the other end of the channel is closed. Ports can also be implicitly closed by being garbage-collected.

在渲染器中,你可以通过分配给 port.onclose 或调用 port.addEventListener('close', ...) 来监听 close 事件。在主进程中,可以通过调用 port.on('close', ...) 来监听 close 事件。

¥In the renderer, you can listen for the close event either by assigning to port.onclose or by calling port.addEventListener('close', ...). In the main process, you can listen for the close event by calling port.on('close', ...).

示例用例

¥Example use cases

在两个渲染器之间设置 MessageChannel

¥Setting up a MessageChannel between two renderers

在此示例中,主进程设置一个 MessageChannel,然后将每个端口发送到不同的渲染器。这允许渲染器相互发送消息,而不需要使用主进程作为中间人。

¥In this example, the main process sets up a MessageChannel, then sends each port to a different renderer. This allows renderers to send messages to each other without needing to use the main process as an in-between.

main.js (Main Process)
const { BrowserWindow, app, MessageChannelMain } = require('electron')

app.whenReady().then(async () => {
// create the windows.
const mainWindow = new BrowserWindow({
show: false,
webPreferences: {
contextIsolation: false,
preload: 'preloadMain.js'
}
})

const secondaryWindow = new BrowserWindow({
show: false,
webPreferences: {
contextIsolation: false,
preload: 'preloadSecondary.js'
}
})

// set up the channel.
const { port1, port2 } = new MessageChannelMain()

// once the webContents are ready, send a port to each webContents with postMessage.
mainWindow.once('ready-to-show', () => {
mainWindow.webContents.postMessage('port', null, [port1])
})

secondaryWindow.once('ready-to-show', () => {
secondaryWindow.webContents.postMessage('port', null, [port2])
})
})

然后,在预加载脚本中,你通过 IPC 接收端口并设置监听器。

¥Then, in your preload scripts you receive the port through IPC and set up the listeners.

preloadMain.js and preloadSecondary.js (Preload scripts)
const { ipcRenderer } = require('electron')

ipcRenderer.on('port', e => {
// port received, make it globally available.
window.electronMessagePort = e.ports[0]

window.electronMessagePort.onmessage = messageEvent => {
// handle message
}
})

在此示例中,messagePort 直接绑定到 window 对象。最好使用 contextIsolation 并为每个预期消息设置特定的 contextBridge 调用,但为了简单起见,我们没有这样做。你可以在本页的 在主进程和上下文隔离页面的主世界之间直接通信 处找到上下文隔离的示例

¥In this example messagePort is bound to the window object directly. It is better to use contextIsolation and set up specific contextBridge calls for each of your expected messages, but for the simplicity of this example we don't. You can find an example of context isolation further down this page at Communicating directly between the main process and the main world of a context-isolated page

这意味着 window.electronMessagePort 是全局可用的,你可以从应用中的任何位置调用 postMessage 以将消息发送到其他渲染器。

¥That means window.electronMessagePort is globally available and you can call postMessage on it from anywhere in your app to send a message to the other renderer.

renderer.js (Renderer Process)
// elsewhere in your code to send a message to the other renderers message handler
window.electronMessagePort.postMessage('ping')

工作进程

¥Worker process

在此示例中,你的应用有一个作为隐藏窗口实现的工作进程。你希望应用页面能够直接与工作进程通信,而无需通过主进程进行中继的性能开销。

¥In this example, your app has a worker process implemented as a hidden window. You want the app page to be able to communicate directly with the worker process, without the performance overhead of relaying via the main process.

main.js (Main Process)
const { BrowserWindow, app, ipcMain, MessageChannelMain } = require('electron')

app.whenReady().then(async () => {
// The worker process is a hidden BrowserWindow, so that it will have access
// to a full Blink context (including e.g. <canvas>, audio, fetch(), etc.)
const worker = new BrowserWindow({
show: false,
webPreferences: { nodeIntegration: true }
})
await worker.loadFile('worker.html')

// The main window will send work to the worker process and receive results
// over a MessagePort.
const mainWindow = new BrowserWindow({
webPreferences: { nodeIntegration: true }
})
mainWindow.loadFile('app.html')

// We can't use ipcMain.handle() here, because the reply needs to transfer a
// MessagePort.
// Listen for message sent from the top-level frame
mainWindow.webContents.mainFrame.ipc.on('request-worker-channel', (event) => {
// Create a new channel ...
const { port1, port2 } = new MessageChannelMain()
// ... send one end to the worker ...
worker.webContents.postMessage('new-client', null, [port1])
// ... and the other end to the main window.
event.senderFrame.postMessage('provide-worker-channel', null, [port2])
// Now the main window and the worker can communicate with each other
// without going through the main process!
})
})
worker.html
<script>
const { ipcRenderer } = require('electron')

const doWork = (input) => {
// Something cpu-intensive.
return input * 2
}

// We might get multiple clients, for instance if there are multiple windows,
// or if the main window reloads.
ipcRenderer.on('new-client', (event) => {
const [ port ] = event.ports
port.onmessage = (event) => {
// The event data can be any serializable object (and the event could even
// carry other MessagePorts with it!)
const result = doWork(event.data)
port.postMessage(result)
}
})
</script>
app.html
<script>
const { ipcRenderer } = require('electron')

// We request that the main process sends us a channel we can use to
// communicate with the worker.
ipcRenderer.send('request-worker-channel')

ipcRenderer.once('provide-worker-channel', (event) => {
// Once we receive the reply, we can take the port...
const [ port ] = event.ports
// ... register a handler to receive results ...
port.onmessage = (event) => {
console.log('received result:', event.data)
}
// ... and start sending it work!
port.postMessage(21)
})
</script>

响应信息流

¥Reply streams

Electron 内置的 IPC 方法只支持两种模式:即发即弃(例如 send),或请求-响应(例如 invoke)。使用 MessageChannel,你可以实现 "响应流",其中单个请求以数据流进行响应。

¥Electron's built-in IPC methods only support two modes: fire-and-forget (e.g. send), or request-response (e.g. invoke). Using MessageChannels, you can implement a "response stream", where a single request responds with a stream of data.

renderer.js (Renderer Process)
const makeStreamingRequest = (element, callback) => {
// MessageChannels are lightweight--it's cheap to create a new one for each
// request.
const { port1, port2 } = new MessageChannel()

// We send one end of the port to the main process ...
ipcRenderer.postMessage(
'give-me-a-stream',
{ element, count: 10 },
[port2]
)

// ... and we hang on to the other end. The main process will send messages
// to its end of the port, and close it when it's finished.
port1.onmessage = (event) => {
callback(event.data)
}
port1.onclose = () => {
console.log('stream ended')
}
}

makeStreamingRequest(42, (data) => {
console.log('got response data:', data)
})
// We will see "got response data: 42" 10 times.
main.js (Main Process)
ipcMain.on('give-me-a-stream', (event, msg) => {
// The renderer has sent us a MessagePort that it wants us to send our
// response over.
const [replyPort] = event.ports

// Here we send the messages synchronously, but we could just as easily store
// the port somewhere and send messages asynchronously.
for (let i = 0; i < msg.count; i++) {
replyPort.postMessage(msg.element)
}

// We close the port when we're done to indicate to the other end that we
// won't be sending any more messages. This isn't strictly necessary--if we
// didn't explicitly close the port, it would eventually be garbage
// collected, which would also trigger the 'close' event in the renderer.
replyPort.close()
})

在主进程和上下文隔离页面的主世界之间直接通信

¥Communicating directly between the main process and the main world of a context-isolated page

当启用 上下文隔离 时,从主进程到渲染器的 IPC 消息将传递到隔离世界,而不是主世界。有时你想直接向主世界传递消息,而不必穿越孤立的世界。

¥When context isolation is enabled, IPC messages from the main process to the renderer are delivered to the isolated world, rather than to the main world. Sometimes you want to deliver messages to the main world directly, without having to step through the isolated world.

main.js (Main Process)
const { BrowserWindow, app, MessageChannelMain } = require('electron')
const path = require('node:path')

app.whenReady().then(async () => {
// Create a BrowserWindow with contextIsolation enabled.
const bw = new BrowserWindow({
webPreferences: {
contextIsolation: true,
preload: path.join(__dirname, 'preload.js')
}
})
bw.loadURL('index.html')

// We'll be sending one end of this channel to the main world of the
// context-isolated page.
const { port1, port2 } = new MessageChannelMain()

// It's OK to send a message on the channel before the other end has
// registered a listener. Messages will be queued until a listener is
// registered.
port2.postMessage({ test: 21 })

// We can also receive messages from the main world of the renderer.
port2.on('message', (event) => {
console.log('from renderer main world:', event.data)
})
port2.start()

// The preload script will receive this IPC message and transfer the port
// over to the main world.
bw.webContents.postMessage('main-world-port', null, [port1])
})
preload.js (Preload Script)
const { ipcRenderer } = require('electron')

// We need to wait until the main world is ready to receive the message before
// sending the port. We create this promise in the preload so it's guaranteed
// to register the onload listener before the load event is fired.
const windowLoaded = new Promise(resolve => {
window.onload = resolve
})

ipcRenderer.on('main-world-port', async (event) => {
await windowLoaded
// We use regular window.postMessage to transfer the port from the isolated
// world to the main world.
window.postMessage('main-world-port', '*', event.ports)
})
index.html
<script>
window.onmessage = (event) => {
// event.source === window means the message is coming from the preload
// script, as opposed to from an <iframe> or other source.
if (event.source === window && event.data === 'main-world-port') {
const [ port ] = event.ports
// Once we have the port, we can communicate directly with the main
// process.
port.onmessage = (event) => {
console.log('from main process:', event.data)
port.postMessage(event.data.test * 2)
}
}
}
</script>