Application Structure

This section of the Developer’s Guide describes the general structure of a Native Client application. The section assumes you are familiar with the material presented in the Technical Overview.

Application components

A Native Client application typically contains the following components:

  • an HTML file;
  • JavaScript code, which can be included in the HTML file or contained in one or more separate .js files;
  • CSS styles, which can be included in the HTML file or contained in one or more separate .css files;
  • a Native Client manifest file (with a .nmf extension) that specifies how to load a Native Client module for different processors; and
  • a Native Client module, written in C or C++, and compiled into a portable executable file (with a .pexe extension) or (if using the Chrome Web Store), architecture-specific executable files (with .nexe extensions).

Applications that are published in the Chrome Web Store also include a Chrome Web Store manifest file (manifest.json) and one or more icon files.

HTML file and the <embed> element

The <embed> element in an HTML file triggers the loading of a Native Client module and specifies the rectangle on the web page that is managed by the module. Here is the <embed> element from the “Hello, World” application:

<embed id="hello_tutorial"
  width=0 height=0
  type="application/x-pnacl" />

In the <embed> element:

is the DOM name attribute for the Native Client module (“nacl_module” is often used as a convention)
specifies the DOM ID for the Native Client module
width, height
specify the size in pixels of the rectangle on the web page that is managed by the Native Client module (if the module does not have a visible area, these values can be 0)
refers to the Native Client manifest file that is used to determine which version of a module to load based on the architecture of the user’s computer (see the following section for more information)
specifies the MIME type of the embedded content; for Portable Native Client modules the type must be “application/x-pnacl”. For architecture-specific Native Client modules the type must be “application/x-nacl”

Manifest Files

Native Client applications have two types of manifest files: a Chrome Web Store manifest file and a Native Client manifest file.

A Chrome Web Store manifest file is a file with information about a web application that is published in the Chrome Web Store. This file, named manifest.json, is required for applications that are published in the Chrome Web Store. For more information about this file see Distributing Your Application. and the Chrome Web Store manifest file format.

A Native Client manifest file is a file that specifies which Native Client module (executable) to load. For PNaCl it specifies a single portable executable; otherwise it specifies one for each of the supported end-user computer architectures (for example x86-32, x86-64, or ARM). This file is required for all Native Client applications. The extension for this file is .nmf.

Manifest files for applications that use PNaCl are simple. Here is the manifest for the hello world example:

  "program": {
    "portable": {
      "pnacl-translate": {
        "url": "hello_tutorial.pexe"

For Chrome Web Store applications that do not use PNaCl, a typical manifest file contains a JSON dictionary with a single top-level key/value pair: the “program” key and a value consisting of a nested dictionary. The nested dictionary contains keys corresponding to the names of the supported computer architectures, and values referencing the file to load for a given architecture—specifically, the URL of the .nexe file, given by the "url" key. URLs are specified relative to the location of the manifest file. Here is an example:

  "program": {
    "x86-32": {
      "url": "hello_tutorial_x86_32.nexe"
    "x86-64": {
      "url": "hello_tutorial_x86_64.nexe"
    "arm": {
      "url": "hello_tutorial_arm.nexe"

For applications that use the glibc library, the manifest file must also contain a “files” key that specifies the shared libraries that the applications use. This is discussed in detail in Dynamic Linking and Loading with glibc. To see some example manifest files, build some of the example applications in the SDK (run make in the example subdirectories) and look at the generated manifest files.

In most cases, you can simply use the Python script provided with the SDK,, to create a manifest file for your application as part of the compilation step (see the Makefile in any of the SDK examples for an illustration of how to do so). The manifest file format is also documented.

Modules and instances

A Native Client module is C or C++ code compiled into a PNaCl .pexe file or a NaCl .nexe file.

An instance is a rectangle on a web page that is managed by a module. An instance may have a dimension of width=0 and height=0, meaning that the instance does not have any visible component on the web page. An instance is created by including an <embed> element in a web page. The <embed> element references a Native Client manifest file that loads the appropriate version of the module (either portable, or specific to the end-user’s architecture). A module may be included in a web page multiple times by using multiple <embed> elements that refer to the module; in this case the Native Client runtime system loads the module once and creates multiple instances that are managed by the module.

Native Client modules: A closer look

A Native Client module must include three components:

  • a factory function called CreateModule()
  • a Module class (derived from the pp::Module class)
  • an Instance class (derived from the pp:Instance class)

In the “Hello tutorial” example (in the getting_started/part1 directory of the NaCl SDK), these three components are specified in the file Here is the factory function:

Module* CreateModule() {
  return new HelloTutorialModule();

Native Client modules do not have a main() function. The CreateModule() factory function is the main binding point between a module and the browser, and serves as the entry point into the module. The browser calls CreateModule() when a module is first loaded; this function returns a Module object derived from the pp::Module class. The browser keeps a singleton of the Module object.

Below is the Module class from the “Hello tutorial” example:

class HelloTutorialModule : public pp::Module {
  HelloTutorialModule() : pp::Module() {}
  virtual ~HelloTutorialModule() {}

  virtual pp::Instance* CreateInstance(PP_Instance instance) {
    return new HelloTutorialInstance(instance);

The Module class must include a CreateInstance() method. The browser calls the CreateInstance() method every time it encounters an <embed> element on a web page that references the same module. The CreateInstance() function creates and returns an Instance object derived from the pp::Instance class.

Below is the Instance class from the “Hello tutorial” example:

class HelloTutorialInstance : public pp::Instance {
  explicit HelloTutorialInstance(PP_Instance instance) : pp::Instance(instance) {}
  virtual ~HelloTutorialInstance() {}

  virtual void HandleMessage(const pp::Var& var_message) {}

As in the example above, the Instance class for your module will likely include an implementation of the HandleMessage() function. The browser calls an instance’s HandleMessage() function every time the JavaScript code in an application calls postMessage() to send a message to the instance. See the Native Client messaging system for more information about how to send messages between JavaScript code and Native Client modules.

The NaCl code is only invoked to handle various browser-issued events and callbacks. There is no need to shut down the NaCl instance by calling the exit() function. NaCl modules will be shut down when the user leaves the web page, or the NaCl module’s <embed> is otherwise destroyed. If the NaCl module does call the exit() function, the instance will issue a crash event which can be handled in Javascript.

While the CreateModule() factory function, the Module class, and the Instance class are required for a Native Client application, the code samples shown above don’t actually do anything. Subsequent sections in the Developer’s Guide build on these code samples and add more interesting functionality.