Extensions and apps can exchange messages with native applications using an API that is similar to the other message passing APIs. Native applications that support this feature must register a native messaging host that knows how to communicate with the extension. Chrome starts the host in a separate process and communicates with it using standard input and standard output streams.
Native messaging host
In order to register a native messaging host the application must install a manifest file that defines the native messaging host configuration. Below is an example of the manifest file:
{
"name": "com.my_company.my_application",
"description": "My Application",
"path": "C:\\Program Files\\My Application\\chrome_native_messaging_host.exe",
"type": "stdio",
"allowed_origins": [
"chrome-extension://knldjmfmopnpolahpmmgbagdohdnhkik/"
]
}
The native messaging host manifest file must be valid JSON and contains the following fields:
Name | Description |
---|---|
name | Name of the native messaging host. Clients pass this string toruntime.connectNative or runtime.sendNativeMessage. This name can only contain lowercase alphanumeric characters, underscores and dots. The name cannot start or end with a dot, and a dot cannot be followed by another dot. |
description | Short application description. |
path | Path to the native messaging host binary. On Linux and OSX the path must be absolute. On Windows it can be relative to the directory in which the manifest file is located. The host process is started with the current directory set to the directory that contains the host binary. For example if this parameter is set to C:\Application\nm_host.exe then it will be started with current directory C:\Application\ . |
type | Type of the interface used to communicate with the native messaging host. Currently there is only one possible value for this parameter: stdio . It indicates that Chrome should use stdin and stdout to communicate with the host. |
allowed_origins | List of extensions that should have access to the native messaging host. Wildcards such as chrome-extension://*/* are not allowed. |
Native messaging host location
The location of the manifest file depends on the platform.
On Windows, the manifest file can be located anywhere in the file system. The application installer must create registry keyHKEY_LOCAL_MACHINE\SOFTWARE\Google\Chrome\NativeMessagingHosts\com.my_company.my_application
orHKEY_CURRENT_USER\SOFTWARE\Google\Chrome\NativeMessagingHosts\com.my_company.my_application
, and set default value of that key to the full path to the manifest file. For example, using the following command:
REG ADD "HKCU\Software\Google\Chrome\NativeMessagingHosts\com.my_company.my_application" /ve /t REG_SZ /d "C:\path\to\nmh-manifest.json" /f
or using the following .reg
file:
Windows Registry Editor Version 5.00
[HKEY_CURRENT_USER\Software\Google\Chrome\NativeMessagingHosts\com.my_company.my_application]
@="C:\\path\\to\\nmh-manifest.json"
When Chrome looks for native messaging hosts, first the 32-bit registry is queried, then the 64-bit registry.
On OS X and Linux, the location of the native messaging host's manifest file varies by the browser (Google Chrome or Chromium). The system-wide native messaging hosts are looked up at a fixed location, while the user-level native messaging hosts are looked up in a subdirectory within the user profile directory called NativeMessagingHosts
.
OS X (system-wide)
Google Chrome: /Library/Google/Chrome/NativeMessagingHosts/com.my_company.my_application.json
Chromium: /Library/Application Support/Chromium/NativeMessagingHosts/com.my_company.my_application.json
OS X (user-specific, default path)
Google Chrome: ~/Library/Application Support/Google/Chrome/NativeMessagingHosts/com.my_company.my_application.json
Chromium: ~/Library/Application Support/Chromium/NativeMessagingHosts/com.my_company.my_application.json
Linux (system-wide)
Google Chrome: /etc/opt/chrome/native-messaging-hosts/com.my_company.my_application.json
Chromium: /etc/chromium/native-messaging-hosts/com.my_company.my_application.json
Linux (user-specific, default path)
Google Chrome: ~/.config/google-chrome/NativeMessagingHosts/com.my_company.my_application.json
Chromium: ~/.config/chromium/NativeMessagingHosts/com.my_company.my_application.json
Native messaging protocol
Chrome starts each native messaging host in a separate process and communicates with it using standard input (stdin
) and standard output (stdout
). The same format is used to send messages in both directions: each message is serialized using JSON, UTF-8 encoded and is preceded with 32-bit message length in native byte order. The maximum size of a single message from the native messaging host is 1 MB, mainly to protect Chrome from misbehaving native applications. The maximum size of the message sent to the native messaging host is 4 GB.
The first argument to the native messaging host is the origin of the caller, usually chrome-extension://[ID of allowed extension]
. This allows native messaging hosts to identify the source of the message when multiple extensions are specified in the allowed_origins
key in the native messaging host manifest.
Warning: In Windows, in Chrome 54 and earlier, the origin was passed as the second parameter instead of the first parameter.
When a messaging port is created using runtime.connectNative Chrome starts native messaging host process and keeps it running until the port is destroyed. On the other hand, when a message is sent usingruntime.sendNativeMessage, without creating a messaging port, Chrome starts a new native messaging host process for each message. In that case the first message generated by the host process is handled as a response to the original request, i.e. Chrome will pass it to the response callback specified whenruntime.sendNativeMessage is called. All other messages generated by the native messaging host in that case are ignored.
On Windows, the native messaging host is also passed a command line argument with a handle to the calling chrome native window: --parent-window=<decimal handle value>
. This lets the native messaging host create native UI windows that are correctly focused.
Connecting to a native application
Sending and receiving messages to and from a native application is very similar to cross-extension messaging. The main difference is that runtime.connectNative is used instead of runtime.connect, and runtime.sendNativeMessage is used instead of runtime.sendMessage.
These methods can only be used if the "nativeMessaging" permission is declared in your extension's manifest file.
The Following example creates a runtime.Port object that's connected to native messaging host com.my_company.my_application
, starts listening for messages from that port and sends one outgoing message:
var port = chrome.runtime.connectNative('com.my_company.my_application');
port.onMessage.addListener(function(msg) {
console.log("Received" + msg);
});
port.onDisconnect.addListener(function() {
console.log("Disconnected");
});
port.postMessage({ text: "Hello, my_application" });
runtime.sendNativeMessage can be used to send a message to native application without creating a port, e.g.:
chrome.runtime.sendNativeMessage('com.my_company.my_application',
{ text: "Hello" },
function(response) {
console.log("Received " + response);
});
Debugging native messaging
When the native messaging host fails to start, writes to stderr
or when it violates the communication protocol, output is written to the error log of Chrome. On Linux and OS X, this log can easily be accessed by starting Chrome from the command line and watching its output in the terminal. On Windows, use --enable-logging
as explained at How to enable logging.
Here are some errors and tips for solving the issues:
Failed to start native messaging host.
Check whether you have sufficient permissions to execute the file.
Invalid native messaging host name specified.
Check whether the name contains any invalid characters. Only lowercase alphanumeric characters, underscores and dots are allowed. A name cannot start or end with a dot, and a dot cannot be followed by another dot.
Native host has exited.
The pipe to the native messaging host was broken before the message was read by Chrome. This is most likely initiated from your native messaging host.
Specified native messaging host not found.
- Is the name spelled correctly in the extension and in the manifest file?
- Is the manifest put in the right directory and with the correct name? See native messaging host location for the expected formats.
- Is the manifest file in the correct format? In particular, is the JSON syntax correct and do the values match the definition of a native messaging host manifest?
- Does the file specified in
path
exist? On Windows, paths may be relative, but on OS X and Linux, the paths must be absolute.
Native messaging host host name is not registered. (Windows-only)
The native messaging host was not found in the Windows registry. Double-check using regedit
whether the key was really created and matches the required format as documented at native messaging host location.
Access to the specified native messaging host is forbidden.
Is the extension's origin listed in allowed_origins
?
Error when communicating with the native messaging host.
This is a very common error and indicates an incorrect implementation of the communication protocol in the native messaging host.
- Make sure that all output in
stdout
adheres to the native messaging protocol. If you want to print some data for debugging purposes, write tostderr
. - Make sure that the 32-bit message length is in the platform's native integer format (little-endian / big-endian).
- The message length must not exceed 1024*1024.
- The message size must be equal to the number of bytes in the message. This may differ from the "length" of a string, because characters may be represented by multiple bytes.
- Windows-only: Make sure that the program's I/O mode is set to
O_BINARY
. By default, the I/O mode isO_TEXT
, which corrupts the message format as line breaks (\n
=0A
) are replaced with Windows-style line endings (\r\n
=0D 0A
). The I/O mode can be set using__setmode
.
Examples
The examples/api/nativeMessaging directory contains an example application that uses native messaging to communicate with a Python script that serves as a native messaging host. The sample host's directory also contains scripts to install/remove the native messaging host.
To try out the example, first download and extract the sample app and sample host. Run install_host.bat
(Windows) or install_host.sh
(Linux / OS X) to install the native messaging host. Then load the app and interact with the app. Run uninstall_host.bat
or uninstall_host.sh
to unregister the native messaging host when you are done.
Messaging System
This section describes the messaging system used to communicate between the JavaScript code and the Native Client module’s C or C++ code in a Native Client application. It introduces the concept of asynchronous programming and the basic steps required to set up a Native Client module that sends messages to and receive messages from JavaScript. This section assumes you are familiar with the material presented in the Application Structure section.
The “Hello, World” example for getting started with NaCl is used here to illustrate basic programming techniques. You can find this code in the /getting_started/part2
directory in the Native Client SDK download.
Reference information
For reference information related to the Pepper messaging API, see the following documentation:
- pp::Instance class HandleMessage(), PostMessage())
- pp::Module class
- pp::Var class
Introduction to the messaging system
Native Client modules and JavaScript communicate by sending messages to each other. The most basic form of a message is a string. Messages support many JavaScript types, including ints, arrays, array buffers, and dictionaries (see pp::Var, pp:VarArrayBuffer, and the general messaging system documentation). It’s up to you to decide on the type of message and define how to process the messages on both the JavaScript and Native Client side. For the “Hello, World” example, we will work with string-typed messages only.
When JavaScript posts a message to the Native Client module, the Pepper HandleMessage()
function is invoked on the module side. Similarly, the Native Client module can post a message to JavaScript, and this message triggers a JavaScript event listener for message
events in the DOM. (See the W3C specification onDocument Object Model Events for more information.) In the “Hello, World” example, the JavaScript functions for posting and handling messages are named postMessage()
and handleMessage()
(but any names could be used). On the Native Client C++ side, the Pepper Library functions for posting and handling messages are:
void pp::Instance::PostMessage(const Var &message)
virtual void pp::Instance::HandleMessage(const Var &message)
If you want to receive messages from JavaScript, you need to implement thepp::Instance::HandleMessage()
function in your Native Client module.
Design of the messaging system
The Native Client messaging system is analogous to the system used by the browser to allow web workers to communicate (see the W3 web worker specification). The Native Client messaging system is designed to keep the web page responsive while the Native Client module is performing potentially heavy processing in the background. When JavaScript sends a message to the Native Client module, the postMessage()
call returns as soon as it sends its message to the Native Client module. The JavaScript does not wait for a reply from Native Client, thus avoiding bogging down the main JavaScript thread. On the JavaScript side, you set up an event listener to respond to the message sent by the Native Client module when it has finished the requested processing and returns a message.
This asynchronous processing model keeps the main thread free while avoiding the following problems:
- The JavaScript engine hangs while waiting for a synchronous call to return.
- The browser pops up a dialog when a JavaScript entry point takes longer than a few moments.
- The application hangs while waiting for an unresponsive Native Client module.
Communication tasks in the “Hello, World” example
The following sections describe how the “Hello, World” example posts and handles messages on both the JavaScript side and the Native Client side of the application.
JavaScript code
The JavaScript code and HTML in the “Hello, World” example can be found in the example.js
, common.js
, and index.html
files. The important steps are:
- Sets up an event listener to listen for
message
events from the Native Client module. - Implements an event handler that the event listener invokes to handle incoming
message
events. - Calls
postMessage()
to communicate with the NaCl module, after the page loads.
Step 1: From common.js
function attachDefaultListeners() {
// The NaCl module embed is created within the listenerDiv
var listenerDiv = document.getElementById('listener');
// ...
// register the handleMessage function as the message event handler.
listenerDiv.addEventListener('message', handleMessage, true);
// ...
}
Step 2: From example.js
// This function is called by common.js when a message is received from the
// NaCl module.
function handleMessage(message) {
// In the example, we simply log the data that's received in the message.
var logEl = document.getElementById('log');
logEl.textContent += message.data;
}
// In the index.html we have set up the appropriate divs:
<body {attrs}>
<!-- ... -->
<div id="listener"></div>
<div id="log"></div>
</body>
Step 3: From example.js
// From example.js, Step 3:
function moduleDidLoad() {
// After the NaCl module has loaded, common.naclModule is a reference to the
// NaCl module's <embed> element.
//
// postMessage sends a message to it.
common.naclModule.postMessage('hello');
}
Native Client module
The C++ code in the Native Client module of the “Hello, World” example:
- Implements
pp::Instance::HandleMessage()
to handle messages sent by the JavaScript. - Processes incoming messages. This example simply checks that JavaScript has sent a “hello” message and not some other message.
- Calls
PostMessage()
to send an acknowledgement back to the JavaScript code. The acknowledgement is a string in the form of aVar
that the JavaScript code can process. In general, app::Var
can be several JavaScript types, see the messaging system documentation.
class HelloTutorialInstance : public pp::Instance {
public:
// ...
// === Step 1: Implement the HandleMessage function. ===
virtual void HandleMessage(const pp::Var& var_message) {
// === Step 2: Process the incoming message. ===
// Ignore the message if it is not a string.
if (!var_message.is_string())
return;
// Get the string message and compare it to "hello".
std::string message = var_message.AsString();
if (message == kHelloString) {
// === Step 3: Send the reply. ===
// If it matches, send our response back to JavaScript.
pp::Var var_reply(kReplyString);
PostMessage(var_reply);
}
}
};
Messaging in JavaScript code: More details.
This section describes in more detail the messaging system code in the JavaScript portion of the “Hello, World” example.
Setting up an event listener and handler
The following JavaScript code sets up an event listener for messages posted by the Native Client module. It then defines a message handler that simply logs the content of messages received from the module.
Setting up the ‘message’ handler on load
// From common.js
// Listen for the DOM content to be loaded. This event is fired when
// parsing of the page's document has finished.
document.addEventListener('DOMContentLoaded', function() {
var body = document.body;
// ...
var loadFunction = common.domContentLoaded;
// ... set up parameters ...
loadFunction(...);
}
// This function is exported as common.domContentLoaded.
function domContentLoaded(...) {
// ...
if (common.naclModule == null) {
// ...
attachDefaultListeners();
// initialize common.naclModule ...
} else {
// ...
}
}
function attachDefaultListeners() {
var listenerDiv = document.getElementById('listener');
// ...
listenerDiv.addEventListener('message', handleMessage, true);
// ...
}
Implementing the handler
// From example.js
function handleMessage(message) {
var logEl = document.getElementById('log');
logEl.textContent += message.data;
}
Note that the handleMessage()
function is handed a message_event containing data
that you can display or manipulate in JavaScript. The “Hello, World” application simply logs this data to the log
div.
Messaging in the Native Client module: More details.
This section describes in more detail the messaging system code in the Native Client module portion of the “Hello, World” example.
Implementing HandleMessage()
If you want the Native Client module to receive and handle messages from JavaScript, you need to implement a HandleMessage()
function for your module’s pp::Instance
class. TheHelloWorldInstance::HandleMessage()
function examines the message posted from JavaScript. First it examines that the type of the pp::Var
is indeed a string (not a double, etc.). It then interprets the data as a string with var_message.AsString()
, and checks that the string matches kHelloString
. After examining the message received from JavaScript, the code calls PostMessage()
to send a reply message back to the JavaScript side.
namespace {
// The expected string sent by the JavaScript.
const char* const kHelloString = "hello";
// The string sent back to the JavaScript code upon receipt of a message
// containing "hello".
const char* const kReplyString = "hello from NaCl";
} // namespace
class HelloTutorialInstance : public pp::Instance {
public:
// ...
virtual void HandleMessage(const pp::Var& var_message) {
// Ignore the message if it is not a string.
if (!var_message.is_string())
return;
// Get the string message and compare it to "hello".
std::string message = var_message.AsString();
if (message == kHelloString) {
// If it matches, send our response back to JavaScript.
pp::Var var_reply(kReplyString);
PostMessage(var_reply);
}
}
};
Implementing application-specific functions
While the “Hello, World” example is very simple, your Native Client module will likely include application-specific functions to perform custom tasks in response to messages. For example the application could be a compression and decompression service (two functions exported). The application could set up an application-specific convention that messages coming from JavaScript are colon-separated pairs of the form <command>:<data>
. The Native Client module message handler can then split the incoming string along the :
character to determine which command to execute. If the command is “compress”, then data to process is an uncompressed string. If the command is “uncompress”, then data to process is an already-compressed string. After processing the data asynchronously, the application then returns the result to JavaScript.
Sending messages back to the JavaScript code
The Native Client module sends messages back to the JavaScript code using PostMessage()
. The Native Client module always returns its values in the form of a pp::Var
that can be processed by the browser’s JavaScript. In this example, the message is posted at the end of the Native Client module’s HandleMessage()
function:
PostMessage(var_reply);
Sending and receiving other pp::Var
types
Besides strings, pp::Var
can represent other types of JavaScript objects. For example, messages can be JavaScript objects. These richer types can make it easier to implement an application’s messaging protocol.
To send a dictionary from the NaCl module to JavaScript simply create a pp::VarDictionary
and then call PostMessage
with the dictionary.
pp::VarDictionary dictionary;
dictionary.Set(pp::Var("command"), pp::Var(next_command));
dictionary.Set(pp::Var("param_int"), pp::Var(123));
pp::VarArray an_array;
an_array.Set(0, pp::Var("string0"));
an_array.Set(1, pp::Var("string1"))
dictionary.Set(pp::Var("param_array"), an_array);
PostMessage(dictionary);
Here is how to create a similar object in JavaScript and send it to the NaCl module:
var dictionary = {
command: next_command,
param_int: 123,
param_array: ['string0', 'string1']
}
nacl_module.postMessage(dictionary);
To receive a dictionary-typed message in the NaCl module, test that the message is truly a dictionary type, then convert the message with the pp::VarDictionary
class.
virtual void HandleMessage(const pp::Var& var) {
if (var.is_dictionary()) {
pp::VarDictionary dictionary(var);
// Use the dictionary
pp::VarArray keys = dictionary.GetKeys();
// ...
} else {
// ...
}
}
Welcome to Native Client
To get the SDK and
installation instructions
visit the SDK Download page.
Native Client is a sandbox for running compiled C and C++ code in the browser efficiently and securely, independent of the user’s operating system. Portable Native Client extends that technology with architecture independence, letting developers compile their code once to run in any website and on any architecture with ahead-of-time (AOT) translation.
In short, Native Client brings the performance and low-level control of native code to modern web browsers, without sacrificing the security and portability of the web. Watch the video below for an overview of Native Client, including its goals, how it works, and how Portable Native Client lets developers run native compiled code on the web.
This site uses several examples of Native Client. For the best experience, consider downloading the latest version of Chrome. When you come back, be sure to check out our demos.
Two Types of Modules
Native Client comes in two flavors.
- Portable Native Client (PNaCl): Pronounced ‘pinnacle’, PNaCl runs single, portable (pexe) executables and is available in most implementations of Chrome. A translator built into Chrome translates the pexe into native code for the client hardware. The entire module is translated before any code is executed rather than as the code is executed. PNaCl modules can be hosted from any web server.
- Native Client (NaCl): Also called traditional or non-portable Native Client, NaCl runs architecture-dependent (nexe) modules, which are packaged into an application. At runtime, the browser decides which nexe to load based on the architecture of the client machine. Apps and Extensions installed via the Chrome Web Store (CWS) can use NaCl modules without additional prompting. NaCl apps can also be installed from chrome://extensions or the command-line during development, however, this is not a recommended distribution mechanism.
These flavors are described in more depth in PNaCl and NaCl
Hello World
To jump right in take the tutorial that walks you through a basic web application for Portable Native Client (PNaCl). This is a client-side application that uses HTML, JavaScript, and a Native Client module written in C++.
A Little More Advanced
If you’ve already got the basics down, you’re probably trying to get a real application ready for production. You’re building, debugging or ready to distribute.
Nuts and Bolts
You’ve been working on a Native Client module for a while now and you’ve run into an arcane problem. You may need to refer to the PNaCl Bitcode Reference or the Sandbox internals.
I Want to Know Everything
So, you like to read now and try later. Start with our Technical Overview
Send us comments and feedback on the native-client-discuss mailing list, or ask questions using Stack Overflow’s google-nativeclient tag.