path: root/doc/api/addons.md

# C++ Addons

<!--introduced_in=v0.10.0-->
<!-- type=misc -->

Node.js Addons are dynamically-linked shared objects, written in C++, that
can be loaded into Node.js using the [`require()`][require] function, and used
just as if they were an ordinary Node.js module. They are used primarily to
provide an interface between JavaScript running in Node.js and C/C++ libraries.

At the moment, the method for implementing Addons is rather complicated,
involving knowledge of several components and APIs:

 - V8: the C++ library Node.js currently uses to provide the
   JavaScript implementation. V8 provides the mechanisms for creating objects,
   calling functions, etc. V8's API is documented mostly in the
   `v8.h` header file (`deps/v8/include/v8.h` in the Node.js source
   tree), which is also available [online][v8-docs].

 - [libuv][]: The C library that implements the Node.js event loop, its worker
   threads and all of the asynchronous behaviors of the platform. It also
   serves as a cross-platform abstraction library, giving easy, POSIX-like
   access across all major operating systems to many common system tasks, such
   as interacting with the filesystem, sockets, timers, and system events. libuv
   also provides a pthreads-like threading abstraction that may be used to
   power more sophisticated asynchronous Addons that need to move beyond the
   standard event loop. Addon authors are encouraged to think about how to
   avoid blocking the event loop with I/O or other time-intensive tasks by
   off-loading work via libuv to non-blocking system operations, worker threads
   or a custom use of libuv's threads.

 - Internal Node.js libraries. Node.js itself exports a number of C++ APIs
   that Addons can use &mdash; the most important of which is the
   `node::ObjectWrap` class.

 - Node.js includes a number of other statically linked libraries including
   OpenSSL. These other libraries are located in the `deps/` directory in the
   Node.js source tree. Only the libuv, OpenSSL, V8 and zlib symbols are
   purposefully re-exported by Node.js and may be used to various extents by
   Addons.
   See [Linking to Node.js' own dependencies][] for additional information.

All of the following examples are available for [download][] and may
be used as the starting-point for an Addon.

## Hello world

This "Hello world" example is a simple Addon, written in C++, that is the
equivalent of the following JavaScript code:

```js
module.exports.hello = () => 'world';
```

First, create the file `hello.cc`:

```cpp
// hello.cc
#include <node.h>

namespace demo {

using v8::FunctionCallbackInfo;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Object;
using v8::String;
using v8::Value;

void Method(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  args.GetReturnValue().Set(String::NewFromUtf8(
      isolate, "world", NewStringType::kNormal).ToLocalChecked());
}

void Initialize(Local<Object> exports) {
  NODE_SET_METHOD(exports, "hello", Method);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, Initialize)

}  // namespace demo
```

Note that all Node.js Addons must export an initialization function following
the pattern:

```cpp
void Initialize(Local<Object> exports);
NODE_MODULE(NODE_GYP_MODULE_NAME, Initialize)
```

There is no semi-colon after `NODE_MODULE` as it's not a function (see
`node.h`).

The `module_name` must match the filename of the final binary (excluding
the `.node` suffix).

In the `hello.cc` example, then, the initialization function is `Initialize`
and the addon module name is `addon`.

When building addons with `node-gyp`, using the macro `NODE_GYP_MODULE_NAME` as
the first parameter of `NODE_MODULE()` will ensure that the name of the final
binary will be passed to `NODE_MODULE()`.

### Context-aware addons

There are environments in which Node.js addons may need to be loaded multiple
times in multiple contexts. For example, the [Electron][] runtime runs multiple
instances of Node.js in a single process. Each instance will have its own
`require()` cache, and thus each instance will need a native addon to behave
correctly when loaded via `require()`. From the addon's perspective, this means
that it must support multiple initializations.

A context-aware addon can be constructed by using the macro
`NODE_MODULE_INITIALIZER`, which expands to the name of a function which Node.js
will expect to find when it loads an addon. An addon can thus be initialized as
in the following example:

```cpp
using namespace v8;

extern "C" NODE_MODULE_EXPORT void
NODE_MODULE_INITIALIZER(Local<Object> exports,
                        Local<Value> module,
                        Local<Context> context) {
  /* Perform addon initialization steps here. */
}
```

Another option is to use the macro `NODE_MODULE_INIT()`, which will also
construct a context-aware addon. Unlike `NODE_MODULE()`, which is used to
construct an addon around a given addon initializer function,
`NODE_MODULE_INIT()` serves as the declaration of such an initializer to be
followed by a function body.

The following three variables may be used inside the function body following an
invocation of `NODE_MODULE_INIT()`:
* `Local<Object> exports`,
* `Local<Value> module`, and
* `Local<Context> context`

The choice to build a context-aware addon carries with it the responsibility of
carefully managing global static data. Since the addon may be loaded multiple
times, potentially even from different threads, any global static data stored
in the addon must be properly protected, and must not contain any persistent
references to JavaScript objects. The reason for this is that JavaScript
objects are only valid in one context, and will likely cause a crash when
accessed from the wrong context or from a different thread than the one on which
they were created.

The context-aware addon can be structured to avoid global static data by
performing the following steps:
* defining a class which will hold per-addon-instance data. Such
a class should include a `v8::Persistent<v8::Object>` which will hold a weak
reference to the addon's `exports` object. The callback associated with the weak
reference will then destroy the instance of the class.
* constructing an instance of this class in the addon initializer such that the
`v8::Persistent<v8::Object>` is set to the `exports` object.
* storing the instance of the class in a `v8::External`, and
* passing the `v8::External` to all methods exposed to JavaScript by passing it
to the `v8::FunctionTemplate` constructor which creates the native-backed
JavaScript functions. The `v8::FunctionTemplate` constructor's third parameter
accepts the `v8::External`.

This will ensure that the per-addon-instance data reaches each binding that can
be called from JavaScript. The per-addon-instance data must also be passed into
any asynchronous callbacks the addon may create.

The following example illustrates the implementation of a context-aware addon:

```cpp
#include <node.h>

using namespace v8;

class AddonData {
 public:
  AddonData(Isolate* isolate, Local<Object> exports):
      call_count(0) {
    // Link the existence of this object instance to the existence of exports.
    exports_.Reset(isolate, exports);
    exports_.SetWeak(this, DeleteMe, WeakCallbackType::kParameter);
  }

  ~AddonData() {
    if (!exports_.IsEmpty()) {
      // Reset the reference to avoid leaking data.
      exports_.ClearWeak();
      exports_.Reset();
    }
  }

  // Per-addon data.
  int call_count;

 private:
  // Method to call when "exports" is about to be garbage-collected.
  static void DeleteMe(const WeakCallbackInfo<AddonData>& info) {
    delete info.GetParameter();
  }

  // Weak handle to the "exports" object. An instance of this class will be
  // destroyed along with the exports object to which it is weakly bound.
  v8::Persistent<v8::Object> exports_;
};

static void Method(const v8::FunctionCallbackInfo<v8::Value>& info) {
  // Retrieve the per-addon-instance data.
  AddonData* data =
      reinterpret_cast<AddonData*>(info.Data().As<External>()->Value());
  data->call_count++;
  info.GetReturnValue().Set((double)data->call_count);
}

// Initialize this addon to be context-aware.
NODE_MODULE_INIT(/* exports, module, context */) {
  Isolate* isolate = context->GetIsolate();

  // Create a new instance of AddonData for this instance of the addon.
  AddonData* data = new AddonData(isolate, exports);
  // Wrap the data in a v8::External so we can pass it to the method we expose.
  Local<External> external = External::New(isolate, data);

  // Expose the method "Method" to JavaScript, and make sure it receives the
  // per-addon-instance data we created above by passing `external` as the
  // third parameter to the FunctionTemplate constructor.
  exports->Set(context,
               String::NewFromUtf8(isolate, "method", NewStringType::kNormal)
                  .ToLocalChecked(),
               FunctionTemplate::New(isolate, Method, external)
                  ->GetFunction(context).ToLocalChecked()).FromJust();
}
```

#### Worker support

In order to support [`Worker`][] threads, addons need to clean up any resources
they may have allocated when such a thread exists. This can be achieved through
the usage of the `AddEnvironmentCleanupHook()` function:

```c++
void AddEnvironmentCleanupHook(v8::Isolate* isolate,
                               void (*fun)(void* arg),
                               void* arg);
```

This function adds a hook that will run before a given Node.js instance shuts
down. If necessary, such hooks can be removed using
`RemoveEnvironmentCleanupHook()` before they are run, which has the same
signature.

### Building

Once the source code has been written, it must be compiled into the binary
`addon.node` file. To do so, create a file called `binding.gyp` in the
top-level of the project describing the build configuration of the module
using a JSON-like format. This file is used by [node-gyp][] — a tool written
specifically to compile Node.js Addons.

```json
{
  "targets": [
    {
      "target_name": "addon",
      "sources": [ "hello.cc" ]
    }
  ]
}
```

A version of the `node-gyp` utility is bundled and distributed with
Node.js as part of `npm`. This version is not made directly available for
developers to use and is intended only to support the ability to use the
`npm install` command to compile and install Addons. Developers who wish to
use `node-gyp` directly can install it using the command
`npm install -g node-gyp`. See the `node-gyp` [installation instructions][] for
more information, including platform-specific requirements.

Once the `binding.gyp` file has been created, use `node-gyp configure` to
generate the appropriate project build files for the current platform. This
will generate either a `Makefile` (on Unix platforms) or a `vcxproj` file
(on Windows) in the `build/` directory.

Next, invoke the `node-gyp build` command to generate the compiled `addon.node`
file. This will be put into the `build/Release/` directory.

When using `npm install` to install a Node.js Addon, npm uses its own bundled
version of `node-gyp` to perform this same set of actions, generating a
compiled version of the Addon for the user's platform on demand.

Once built, the binary Addon can be used from within Node.js by pointing
[`require()`][require] to the built `addon.node` module:

```js
// hello.js
const addon = require('./build/Release/addon');

console.log(addon.hello());
// Prints: 'world'
```

Please see the examples below for further information or
<https://github.com/arturadib/node-qt> for an example in production.

Because the exact path to the compiled Addon binary can vary depending on how
it is compiled (i.e. sometimes it may be in `./build/Debug/`), Addons can use
the [bindings][] package to load the compiled module.

Note that while the `bindings` package implementation is more sophisticated
in how it locates Addon modules, it is essentially using a try-catch pattern
similar to:

```js
try {
  return require('./build/Release/addon.node');
} catch (err) {
  return require('./build/Debug/addon.node');
}
```

### Linking to Node.js' own dependencies

Node.js uses a number of statically linked libraries such as V8, libuv and
OpenSSL. All Addons are required to link to V8 and may link to any of the
other dependencies as well. Typically, this is as simple as including
the appropriate `#include <...>` statements (e.g. `#include <v8.h>`) and
`node-gyp` will locate the appropriate headers automatically. However, there
are a few caveats to be aware of:

* When `node-gyp` runs, it will detect the specific release version of Node.js
and download either the full source tarball or just the headers. If the full
source is downloaded, Addons will have complete access to the full set of
Node.js dependencies. However, if only the Node.js headers are downloaded, then
only the symbols exported by Node.js will be available.

* `node-gyp` can be run using the `--nodedir` flag pointing at a local Node.js
source image. Using this option, the Addon will have access to the full set of
dependencies.

### Loading Addons using require()

The filename extension of the compiled Addon binary is `.node` (as opposed
to `.dll` or `.so`). The [`require()`][require] function is written to look for
files with the `.node` file extension and initialize those as dynamically-linked
libraries.

When calling [`require()`][require], the `.node` extension can usually be
omitted and Node.js will still find and initialize the Addon. One caveat,
however, is that Node.js will first attempt to locate and load modules or
JavaScript files that happen to share the same base name. For instance, if
there is a file `addon.js` in the same directory as the binary `addon.node`,
then [`require('addon')`][require] will give precedence to the `addon.js` file
and load it instead.

## Native Abstractions for Node.js

Each of the examples illustrated in this document make direct use of the
Node.js and V8 APIs for implementing Addons. It is important to understand
that the V8 API can, and has, changed dramatically from one V8 release to the
next (and one major Node.js release to the next). With each change, Addons may
need to be updated and recompiled in order to continue functioning. The Node.js
release schedule is designed to minimize the frequency and impact of such
changes but there is little that Node.js can do currently to ensure stability
of the V8 APIs.

The [Native Abstractions for Node.js][] (or `nan`) provide a set of tools that
Addon developers are recommended to use to keep compatibility between past and
future releases of V8 and Node.js. See the `nan` [examples][] for an
illustration of how it can be used.

## N-API

> Stability: 2 - Stable

N-API is an API for building native Addons. It is independent from
the underlying JavaScript runtime (e.g. V8) and is maintained as part of
Node.js itself. This API will be Application Binary Interface (ABI) stable
across versions of Node.js. It is intended to insulate Addons from
changes in the underlying JavaScript engine and allow modules
compiled for one version to run on later versions of Node.js without
recompilation. Addons are built/packaged with the same approach/tools
outlined in this document (node-gyp, etc.). The only difference is the
set of APIs that are used by the native code. Instead of using the V8
or [Native Abstractions for Node.js][] APIs, the functions available
in the N-API are used.

Creating and maintaining an addon that benefits from the ABI stability
provided by N-API carries with it certain
[implementation considerations](n-api.html#n_api_implications_of_abi_stability).

To use N-API in the above "Hello world" example, replace the content of
`hello.cc` with the following. All other instructions remain the same.

```cpp
// hello.cc using N-API
#include <node_api.h>

namespace demo {

napi_value Method(napi_env env, napi_callback_info args) {
  napi_value greeting;
  napi_status status;

  status = napi_create_string_utf8(env, "hello", NAPI_AUTO_LENGTH, &greeting);
  if (status != napi_ok) return nullptr;
  return greeting;
}

napi_value init(napi_env env, napi_value exports) {
  napi_status status;
  napi_value fn;

  status = napi_create_function(env, nullptr, 0, Method, nullptr, &fn);
  if (status != napi_ok) return nullptr;

  status = napi_set_named_property(env, exports, "hello", fn);
  if (status != napi_ok) return nullptr;
  return exports;
}

NAPI_MODULE(NODE_GYP_MODULE_NAME, init)

}  // namespace demo
```

The functions available and how to use them are documented in the
section titled [C/C++ Addons - N-API](n-api.html).

## Addon examples

Following are some example Addons intended to help developers get started. The
examples make use of the V8 APIs. Refer to the online [V8 reference][v8-docs]
for help with the various V8 calls, and V8's [Embedder's Guide][] for an
explanation of several concepts used such as handles, scopes, function
templates, etc.

Each of these examples using the following `binding.gyp` file:

```json
{
  "targets": [
    {
      "target_name": "addon",
      "sources": [ "addon.cc" ]
    }
  ]
}
```

In cases where there is more than one `.cc` file, simply add the additional
filename to the `sources` array:

```json
"sources": ["addon.cc", "myexample.cc"]
```

Once the `binding.gyp` file is ready, the example Addons can be configured and
built using `node-gyp`:

```console
$ node-gyp configure build
```

### Function arguments

Addons will typically expose objects and functions that can be accessed from
JavaScript running within Node.js. When functions are invoked from JavaScript,
the input arguments and return value must be mapped to and from the C/C++
code.

The following example illustrates how to read function arguments passed from
JavaScript and how to return a result:

```cpp
// addon.cc
#include <node.h>

namespace demo {

using v8::Exception;
using v8::FunctionCallbackInfo;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Number;
using v8::Object;
using v8::String;
using v8::Value;

// This is the implementation of the "add" method
// Input arguments are passed using the
// const FunctionCallbackInfo<Value>& args struct
void Add(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();

  // Check the number of arguments passed.
  if (args.Length() < 2) {
    // Throw an Error that is passed back to JavaScript
    isolate->ThrowException(Exception::TypeError(
        String::NewFromUtf8(isolate,
                            "Wrong number of arguments",
                            NewStringType::kNormal).ToLocalChecked()));
    return;
  }

  // Check the argument types
  if (!args[0]->IsNumber() || !args[1]->IsNumber()) {
    isolate->ThrowException(Exception::TypeError(
        String::NewFromUtf8(isolate,
                            "Wrong arguments",
                            NewStringType::kNormal).ToLocalChecked()));
    return;
  }

  // Perform the operation
  double value =
      args[0].As<Number>()->Value() + args[1].As<Number>()->Value();
  Local<Number> num = Number::New(isolate, value);

  // Set the return value (using the passed in
  // FunctionCallbackInfo<Value>&)
  args.GetReturnValue().Set(num);
}

void Init(Local<Object> exports) {
  NODE_SET_METHOD(exports, "add", Add);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, Init)

}  // namespace demo
```

Once compiled, the example Addon can be required and used from within Node.js:

```js
// test.js
const addon = require('./build/Release/addon');

console.log('This should be eight:', addon.add(3, 5));
```

### Callbacks

It is common practice within Addons to pass JavaScript functions to a C++
function and execute them from there. The following example illustrates how
to invoke such callbacks:

```cpp
// addon.cc
#include <node.h>

namespace demo {

using v8::Context;
using v8::Function;
using v8::FunctionCallbackInfo;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Null;
using v8::Object;
using v8::String;
using v8::Value;

void RunCallback(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  Local<Context> context = isolate->GetCurrentContext();
  Local<Function> cb = Local<Function>::Cast(args[0]);
  const unsigned argc = 1;
  Local<Value> argv[argc] = {
      String::NewFromUtf8(isolate,
                          "hello world",
                          NewStringType::kNormal).ToLocalChecked() };
  cb->Call(context, Null(isolate), argc, argv).ToLocalChecked();
}

void Init(Local<Object> exports, Local<Object> module) {
  NODE_SET_METHOD(module, "exports", RunCallback);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, Init)

}  // namespace demo
```

Note that this example uses a two-argument form of `Init()` that receives
the full `module` object as the second argument. This allows the Addon
to completely overwrite `exports` with a single function instead of
adding the function as a property of `exports`.

To test it, run the following JavaScript:

```js
// test.js
const addon = require('./build/Release/addon');

addon((msg) => {
  console.log(msg);
// Prints: 'hello world'
});
```

Note that, in this example, the callback function is invoked synchronously.

### Object factory

Addons can create and return new objects from within a C++ function as
illustrated in the following example. An object is created and returned with a
property `msg` that echoes the string passed to `createObject()`:

```cpp
// addon.cc
#include <node.h>

namespace demo {

using v8::Context;
using v8::FunctionCallbackInfo;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Object;
using v8::String;
using v8::Value;

void CreateObject(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  Local<Context> context = isolate->GetCurrentContext();

  Local<Object> obj = Object::New(isolate);
  obj->Set(context,
           String::NewFromUtf8(isolate,
                               "msg",
                               NewStringType::kNormal).ToLocalChecked(),
                               args[0]->ToString(context).ToLocalChecked())
           .FromJust();

  args.GetReturnValue().Set(obj);
}

void Init(Local<Object> exports, Local<Object> module) {
  NODE_SET_METHOD(module, "exports", CreateObject);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, Init)

}  // namespace demo
```

To test it in JavaScript:

```js
// test.js
const addon = require('./build/Release/addon');

const obj1 = addon('hello');
const obj2 = addon('world');
console.log(obj1.msg, obj2.msg);
// Prints: 'hello world'
```

### Function factory

Another common scenario is creating JavaScript functions that wrap C++
functions and returning those back to JavaScript:

```cpp
// addon.cc
#include <node.h>

namespace demo {

using v8::Context;
using v8::Function;
using v8::FunctionCallbackInfo;
using v8::FunctionTemplate;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Object;
using v8::String;
using v8::Value;

void MyFunction(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  args.GetReturnValue().Set(String::NewFromUtf8(
      isolate, "hello world", NewStringType::kNormal).ToLocalChecked());
}

void CreateFunction(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();

  Local<Context> context = isolate->GetCurrentContext();
  Local<FunctionTemplate> tpl = FunctionTemplate::New(isolate, MyFunction);
  Local<Function> fn = tpl->GetFunction(context).ToLocalChecked();

  // omit this to make it anonymous
  fn->SetName(String::NewFromUtf8(
      isolate, "theFunction", NewStringType::kNormal).ToLocalChecked());

  args.GetReturnValue().Set(fn);
}

void Init(Local<Object> exports, Local<Object> module) {
  NODE_SET_METHOD(module, "exports", CreateFunction);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, Init)

}  // namespace demo
```

To test:

```js
// test.js
const addon = require('./build/Release/addon');

const fn = addon();
console.log(fn());
// Prints: 'hello world'
```

### Wrapping C++ objects

It is also possible to wrap C++ objects/classes in a way that allows new
instances to be created using the JavaScript `new` operator:

```cpp
// addon.cc
#include <node.h>
#include "myobject.h"

namespace demo {

using v8::Local;
using v8::Object;

void InitAll(Local<Object> exports) {
  MyObject::Init(exports);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, InitAll)

}  // namespace demo
```

Then, in `myobject.h`, the wrapper class inherits from `node::ObjectWrap`:

```cpp
// myobject.h
#ifndef MYOBJECT_H
#define MYOBJECT_H

#include <node.h>
#include <node_object_wrap.h>

namespace demo {

class MyObject : public node::ObjectWrap {
 public:
  static void Init(v8::Local<v8::Object> exports);

 private:
  explicit MyObject(double value = 0);
  ~MyObject();

  static void New(const v8::FunctionCallbackInfo<v8::Value>& args);
  static void PlusOne(const v8::FunctionCallbackInfo<v8::Value>& args);
  static v8::Persistent<v8::Function> constructor;
  double value_;
};

}  // namespace demo

#endif
```

In `myobject.cc`, implement the various methods that are to be exposed.
Below, the method `plusOne()` is exposed by adding it to the constructor's
prototype:

```cpp
// myobject.cc
#include "myobject.h"

namespace demo {

using v8::Context;
using v8::Function;
using v8::FunctionCallbackInfo;
using v8::FunctionTemplate;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Number;
using v8::Object;
using v8::Persistent;
using v8::String;
using v8::Value;

Persistent<Function> MyObject::constructor;

MyObject::MyObject(double value) : value_(value) {
}

MyObject::~MyObject() {
}

void MyObject::Init(Local<Object> exports) {
  Isolate* isolate = exports->GetIsolate();

  // Prepare constructor template
  Local<FunctionTemplate> tpl = FunctionTemplate::New(isolate, New);
  tpl->SetClassName(String::NewFromUtf8(
      isolate, "MyObject", NewStringType::kNormal).ToLocalChecked());
  tpl->InstanceTemplate()->SetInternalFieldCount(1);

  // Prototype
  NODE_SET_PROTOTYPE_METHOD(tpl, "plusOne", PlusOne);

  Local<Context> context = isolate->GetCurrentContext();
  constructor.Reset(isolate, tpl->GetFunction(context).ToLocalChecked());
  exports->Set(context, String::NewFromUtf8(
      isolate, "MyObject", NewStringType::kNormal).ToLocalChecked(),
               tpl->GetFunction(context).ToLocalChecked()).FromJust();
}

void MyObject::New(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  Local<Context> context = isolate->GetCurrentContext();

  if (args.IsConstructCall()) {
    // Invoked as constructor: `new MyObject(...)`
    double value = args[0]->IsUndefined() ?
        0 : args[0]->NumberValue(context).FromMaybe(0);
    MyObject* obj = new MyObject(value);
    obj->Wrap(args.This());
    args.GetReturnValue().Set(args.This());
  } else {
    // Invoked as plain function `MyObject(...)`, turn into construct call.
    const int argc = 1;
    Local<Value> argv[argc] = { args[0] };
    Local<Function> cons = Local<Function>::New(isolate, constructor);
    Local<Object> result =
        cons->NewInstance(context, argc, argv).ToLocalChecked();
    args.GetReturnValue().Set(result);
  }
}

void MyObject::PlusOne(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();

  MyObject* obj = ObjectWrap::Unwrap<MyObject>(args.Holder());
  obj->value_ += 1;

  args.GetReturnValue().Set(Number::New(isolate, obj->value_));
}

}  // namespace demo
```

To build this example, the `myobject.cc` file must be added to the
`binding.gyp`:

```json
{
  "targets": [
    {
      "target_name": "addon",
      "sources": [
        "addon.cc",
        "myobject.cc"
      ]
    }
  ]
}
```

Test it with:

```js
// test.js
const addon = require('./build/Release/addon');

const obj = new addon.MyObject(10);
console.log(obj.plusOne());
// Prints: 11
console.log(obj.plusOne());
// Prints: 12
console.log(obj.plusOne());
// Prints: 13
```

The destructor for a wrapper object will run when the object is
garbage-collected. For destructor testing, there are command-line flags that
can be used to make it possible to force garbage collection. These flags are
provided by the underlying V8 JavaScript engine. They are subject to change
or removal at any time. They are not documented by Node.js or V8, and they
should never be used outside of testing.

### Factory of wrapped objects

Alternatively, it is possible to use a factory pattern to avoid explicitly
creating object instances using the JavaScript `new` operator:

```js
const obj = addon.createObject();
// instead of:
// const obj = new addon.Object();
```

First, the `createObject()` method is implemented in `addon.cc`:

```cpp
// addon.cc
#include <node.h>
#include "myobject.h"

namespace demo {

using v8::FunctionCallbackInfo;
using v8::Isolate;
using v8::Local;
using v8::Object;
using v8::String;
using v8::Value;

void CreateObject(const FunctionCallbackInfo<Value>& args) {
  MyObject::NewInstance(args);
}

void InitAll(Local<Object> exports, Local<Object> module) {
  MyObject::Init(exports->GetIsolate());

  NODE_SET_METHOD(module, "exports", CreateObject);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, InitAll)

}  // namespace demo
```

In `myobject.h`, the static method `NewInstance()` is added to handle
instantiating the object. This method takes the place of using `new` in
JavaScript:

```cpp
// myobject.h
#ifndef MYOBJECT_H
#define MYOBJECT_H

#include <node.h>
#include <node_object_wrap.h>

namespace demo {

class MyObject : public node::ObjectWrap {
 public:
  static void Init(v8::Isolate* isolate);
  static void NewInstance(const v8::FunctionCallbackInfo<v8::Value>& args);

 private:
  explicit MyObject(double value = 0);
  ~MyObject();

  static void New(const v8::FunctionCallbackInfo<v8::Value>& args);
  static void PlusOne(const v8::FunctionCallbackInfo<v8::Value>& args);
  static v8::Persistent<v8::Function> constructor;
  double value_;
};

}  // namespace demo

#endif
```

The implementation in `myobject.cc` is similar to the previous example:

```cpp
// myobject.cc
#include <node.h>
#include "myobject.h"

namespace demo {

using v8::Context;
using v8::Function;
using v8::FunctionCallbackInfo;
using v8::FunctionTemplate;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Number;
using v8::Object;
using v8::Persistent;
using v8::String;
using v8::Value;

Persistent<Function> MyObject::constructor;

MyObject::MyObject(double value) : value_(value) {
}

MyObject::~MyObject() {
}

void MyObject::Init(Isolate* isolate) {
  // Prepare constructor template
  Local<FunctionTemplate> tpl = FunctionTemplate::New(isolate, New);
  tpl->SetClassName(String::NewFromUtf8(
      isolate, "MyObject", NewStringType::kNormal).ToLocalChecked());
  tpl->InstanceTemplate()->SetInternalFieldCount(1);

  // Prototype
  NODE_SET_PROTOTYPE_METHOD(tpl, "plusOne", PlusOne);

  Local<Context> context = isolate->GetCurrentContext();
  constructor.Reset(isolate, tpl->GetFunction(context).ToLocalChecked());
}

void MyObject::New(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  Local<Context> context = isolate->GetCurrentContext();

  if (args.IsConstructCall()) {
    // Invoked as constructor: `new MyObject(...)`
    double value = args[0]->IsUndefined() ?
        0 : args[0]->NumberValue(context).FromMaybe(0);
    MyObject* obj = new MyObject(value);
    obj->Wrap(args.This());
    args.GetReturnValue().Set(args.This());
  } else {
    // Invoked as plain function `MyObject(...)`, turn into construct call.
    const int argc = 1;
    Local<Value> argv[argc] = { args[0] };
    Local<Function> cons = Local<Function>::New(isolate, constructor);
    Local<Object> instance =
        cons->NewInstance(context, argc, argv).ToLocalChecked();
    args.GetReturnValue().Set(instance);
  }
}

void MyObject::NewInstance(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();

  const unsigned argc = 1;
  Local<Value> argv[argc] = { args[0] };
  Local<Function> cons = Local<Function>::New(isolate, constructor);
  Local<Context> context = isolate->GetCurrentContext();
  Local<Object> instance =
      cons->NewInstance(context, argc, argv).ToLocalChecked();

  args.GetReturnValue().Set(instance);
}

void MyObject::PlusOne(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();

  MyObject* obj = ObjectWrap::Unwrap<MyObject>(args.Holder());
  obj->value_ += 1;

  args.GetReturnValue().Set(Number::New(isolate, obj->value_));
}

}  // namespace demo
```

Once again, to build this example, the `myobject.cc` file must be added to the
`binding.gyp`:

```json
{
  "targets": [
    {
      "target_name": "addon",
      "sources": [
        "addon.cc",
        "myobject.cc"
      ]
    }
  ]
}
```

Test it with:

```js
// test.js
const createObject = require('./build/Release/addon');

const obj = createObject(10);
console.log(obj.plusOne());
// Prints: 11
console.log(obj.plusOne());
// Prints: 12
console.log(obj.plusOne());
// Prints: 13

const obj2 = createObject(20);
console.log(obj2.plusOne());
// Prints: 21
console.log(obj2.plusOne());
// Prints: 22
console.log(obj2.plusOne());
// Prints: 23
```

### Passing wrapped objects around

In addition to wrapping and returning C++ objects, it is possible to pass
wrapped objects around by unwrapping them with the Node.js helper function
`node::ObjectWrap::Unwrap`. The following examples shows a function `add()`
that can take two `MyObject` objects as input arguments:

```cpp
// addon.cc
#include <node.h>
#include <node_object_wrap.h>
#include "myobject.h"

namespace demo {

using v8::Context;
using v8::FunctionCallbackInfo;
using v8::Isolate;
using v8::Local;
using v8::Number;
using v8::Object;
using v8::String;
using v8::Value;

void CreateObject(const FunctionCallbackInfo<Value>& args) {
  MyObject::NewInstance(args);
}

void Add(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  Local<Context> context = isolate->GetCurrentContext();

  MyObject* obj1 = node::ObjectWrap::Unwrap<MyObject>(
      args[0]->ToObject(context).ToLocalChecked());
  MyObject* obj2 = node::ObjectWrap::Unwrap<MyObject>(
      args[1]->ToObject(context).ToLocalChecked());

  double sum = obj1->value() + obj2->value();
  args.GetReturnValue().Set(Number::New(isolate, sum));
}

void InitAll(Local<Object> exports) {
  MyObject::Init(exports->GetIsolate());

  NODE_SET_METHOD(exports, "createObject", CreateObject);
  NODE_SET_METHOD(exports, "add", Add);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, InitAll)

}  // namespace demo
```

In `myobject.h`, a new public method is added to allow access to private values
after unwrapping the object.

```cpp
// myobject.h
#ifndef MYOBJECT_H
#define MYOBJECT_H

#include <node.h>
#include <node_object_wrap.h>

namespace demo {

class MyObject : public node::ObjectWrap {
 public:
  static void Init(v8::Isolate* isolate);
  static void NewInstance(const v8::FunctionCallbackInfo<v8::Value>& args);
  inline double value() const { return value_; }

 private:
  explicit MyObject(double value = 0);
  ~MyObject();

  static void New(const v8::FunctionCallbackInfo<v8::Value>& args);
  static v8::Persistent<v8::Function> constructor;
  double value_;
};

}  // namespace demo

#endif
```

The implementation of `myobject.cc` is similar to before:

```cpp
// myobject.cc
#include <node.h>
#include "myobject.h"

namespace demo {

using v8::Context;
using v8::Function;
using v8::FunctionCallbackInfo;
using v8::FunctionTemplate;
using v8::Isolate;
using v8::Local;
using v8::NewStringType;
using v8::Object;
using v8::Persistent;
using v8::String;
using v8::Value;

Persistent<Function> MyObject::constructor;

MyObject::MyObject(double value) : value_(value) {
}

MyObject::~MyObject() {
}

void MyObject::Init(Isolate* isolate) {
  // Prepare constructor template
  Local<FunctionTemplate> tpl = FunctionTemplate::New(isolate, New);
  tpl->SetClassName(String::NewFromUtf8(
      isolate, "MyObject", NewStringType::kNormal).ToLocalChecked());
  tpl->InstanceTemplate()->SetInternalFieldCount(1);

  Local<Context> context = isolate->GetCurrentContext();
  constructor.Reset(isolate, tpl->GetFunction(context).ToLocalChecked());
}

void MyObject::New(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();
  Local<Context> context = isolate->GetCurrentContext();

  if (args.IsConstructCall()) {
    // Invoked as constructor: `new MyObject(...)`
    double value = args[0]->IsUndefined() ?
        0 : args[0]->NumberValue(context).FromMaybe(0);
    MyObject* obj = new MyObject(value);
    obj->Wrap(args.This());
    args.GetReturnValue().Set(args.This());
  } else {
    // Invoked as plain function `MyObject(...)`, turn into construct call.
    const int argc = 1;
    Local<Value> argv[argc] = { args[0] };
    Local<Function> cons = Local<Function>::New(isolate, constructor);
    Local<Object> instance =
        cons->NewInstance(context, argc, argv).ToLocalChecked();
    args.GetReturnValue().Set(instance);
  }
}

void MyObject::NewInstance(const FunctionCallbackInfo<Value>& args) {
  Isolate* isolate = args.GetIsolate();

  const unsigned argc = 1;
  Local<Value> argv[argc] = { args[0] };
  Local<Function> cons = Local<Function>::New(isolate, constructor);
  Local<Context> context = isolate->GetCurrentContext();
  Local<Object> instance =
      cons->NewInstance(context, argc, argv).ToLocalChecked();

  args.GetReturnValue().Set(instance);
}

}  // namespace demo
```

Test it with:

```js
// test.js
const addon = require('./build/Release/addon');

const obj1 = addon.createObject(10);
const obj2 = addon.createObject(20);
const result = addon.add(obj1, obj2);

console.log(result);
// Prints: 30
```

### AtExit hooks

An `AtExit` hook is a function that is invoked after the Node.js event loop
has ended but before the JavaScript VM is terminated and Node.js shuts down.
`AtExit` hooks are registered using the `node::AtExit` API.

#### void AtExit(callback, args)

* `callback` <span class="type">&lt;void (\*)(void\*)&gt;</span>
  A pointer to the function to call at exit.
* `args` <span class="type">&lt;void\*&gt;</span>
  A pointer to pass to the callback at exit.

Registers exit hooks that run after the event loop has ended but before the VM
is killed.

`AtExit` takes two parameters: a pointer to a callback function to run at exit,
and a pointer to untyped context data to be passed to that callback.

Callbacks are run in last-in first-out order.

The following `addon.cc` implements `AtExit`:

```cpp
// addon.cc
#include <assert.h>
#include <stdlib.h>
#include <node.h>

namespace demo {

using node::AtExit;
using v8::HandleScope;
using v8::Isolate;
using v8::Local;
using v8::Object;

static char cookie[] = "yum yum";
static int at_exit_cb1_called = 0;
static int at_exit_cb2_called = 0;

static void at_exit_cb1(void* arg) {
  Isolate* isolate = static_cast<Isolate*>(arg);
  HandleScope scope(isolate);
  Local<Object> obj = Object::New(isolate);
  assert(!obj.IsEmpty());  // assert VM is still alive
  assert(obj->IsObject());
  at_exit_cb1_called++;
}

static void at_exit_cb2(void* arg) {
  assert(arg == static_cast<void*>(cookie));
  at_exit_cb2_called++;
}

static void sanity_check(void*) {
  assert(at_exit_cb1_called == 1);
  assert(at_exit_cb2_called == 2);
}

void init(Local<Object> exports) {
  AtExit(at_exit_cb2, cookie);
  AtExit(at_exit_cb2, cookie);
  AtExit(at_exit_cb1, exports->GetIsolate());
  AtExit(sanity_check);
}

NODE_MODULE(NODE_GYP_MODULE_NAME, init)

}  // namespace demo
```

Test in JavaScript by running:

```js
// test.js
require('./build/Release/addon');
```

[`Worker`]: worker_threads.html#worker_threads_class_worker
[Electron]: https://electronjs.org/
[Embedder's Guide]: https://github.com/v8/v8/wiki/Embedder's%20Guide
[Linking to Node.js' own dependencies]: #addons_linking_to_node_js_own_dependencies
[Native Abstractions for Node.js]: https://github.com/nodejs/nan
[bindings]: https://github.com/TooTallNate/node-bindings
[download]: https://github.com/nodejs/node-addon-examples
[examples]: https://github.com/nodejs/nan/tree/master/examples/
[installation instructions]: https://github.com/nodejs/node-gyp#installation
[libuv]: https://github.com/libuv/libuv
[node-gyp]: https://github.com/nodejs/node-gyp
[require]: modules.html#modules_require_id
[v8-docs]: https://v8docs.nodesource.com/