JavaScriptモジュール:ビギナーズガイド

JavaScriptを初めて使用する場合は、「モジュールバンドラーとモジュールローダー」、「WebpackとBrowserify」、「AMDとCommonJS」などの専門用語がすぐに圧倒される可能性があります。

JavaScriptモジュールシステムは恐ろしいかもしれませんが、それを理解することはWeb開発者にとって不可欠です。

この投稿では、これらの流行語を平易な英語(およびいくつかのコードサンプル)で解凍します。お役に立てば幸いです。

注:簡単にするために、これは2つのセクションに分かれています。パート1では、モジュールとは何か、およびモジュールを使用する理由について説明します。パート2(来週投稿)では、モジュールをバンドルすることの意味と、そのためのさまざまな方法について説明します。

パート1:誰かがモジュールが再び何であるかを説明できますか?

優れた著者は、本を章とセクションに分けています。優れたプログラマーは、プログラムをモジュールに分割します。

本の章のように、モジュールは単なる単語のクラスター(または場合によってはコード)です。

ただし、優れたモジュールは、独自の機能を備えた高度な自己完結型であるため、システム全体を中断することなく、必要に応じてシャッフル、削除、または追加できます。

なぜモジュールを使用するのですか?

広大で相互に依存するコードベースを優先してモジュールを使用することには、多くの利点があります。私の意見では、最も重要なものは次のとおりです。

1)保守性:定義上、モジュールは自己完結型です。適切に設計されたモジュールは、コードベースの一部への依存を可能な限り少なくして、独立して拡張および改善できるようにすることを目的としています。モジュールが他のコードから切り離されている場合、単一のモジュールの更新ははるかに簡単です。

本の例に戻ると、本の章を更新したい場合、1つの章を少し変更するだけで、他のすべての章も微調整する必要があるとしたら、それは悪夢です。代わりに、他の章に影響を与えることなく改善できるように、各章を作成する必要があります。

2)名前空間: JavaScriptでは、トップレベル関数のスコープ外の変数はグローバルです(つまり、誰でもそれらにアクセスできます)。このため、完全に無関係なコードがグローバル変数を共有する「名前空間汚染」が発生するのが一般的です。

無関係なコード間でグローバル変数を共有することは、開発においては大したことではありません。

この投稿の後半で説明するように、モジュールを使用すると、変数用のプライベートスペースを作成することで、名前空間の汚染を回避できます。

3)再利用性:ここで正直に言うと、以前に作成したコードはすべて、ある時点で新しいプロジェクトにコピーされています。たとえば、前のプロジェクトから現在のプロジェクトに作成したいくつかのユーティリティメソッドをコピーしたとします。

それはすべてうまくいっていますが、そのコードの一部を書くためのより良い方法を見つけた場合は、戻って、それを書いた他のすべての場所で更新することを忘れないでください。

これは明らかに時間の大きな無駄です。何度も再利用できるモジュールがあったら、それを待つのははるかに簡単ではないでしょうか。

モジュールをどのように組み込むことができますか?

モジュールをプログラムに組み込む方法はたくさんあります。それらのいくつかを見ていきましょう:

モジュールパターン

モジュールパターンは、クラスの概念を模倣するために使用されます(JavaScriptはクラスをネイティブにサポートしていないため)。これにより、Javaなどの他のプログラミング言語でクラスが使用されるのと同様に、パブリックメソッドとプライベートメソッドおよび変数の両方を単一のオブジェクト内に格納できます。またはPython。これにより、プライベート変数とメソッドをクロージャスコープにカプセル化しながら、世界に公開したいメソッドの公開APIを作成できます。

モジュールパターンを実現するには、いくつかの方法があります。この最初の例では、匿名クロージャを使用します。これは、すべてのコードを無名関数に配置することで、目標を達成するのに役立ちます。(注意:JavaScriptでは、関数が新しいスコープを作成する唯一の方法です。)

例1:匿名クロージャ

(function () { // We keep these variables private inside this closure scope var myGrades = [93, 95, 88, 0, 55, 91]; var average = function() { var total = myGrades.reduce(function(accumulator, item) { return accumulator + item}, 0); return 'Your average grade is ' + total / myGrades.length + '.'; } var failing = function(){ var failingGrades = myGrades.filter(function(item) { return item < 70;}); return 'You failed ' + failingGrades.length + ' times.'; } console.log(failing()); }()); // ‘You failed 2 times.’

この構成では、無名関数に独自の評価環境または「クロージャ」があり、すぐに評価します。これにより、親(グローバル)名前空間から変数を非表示にできます。

このアプローチの優れている点は、既存のグローバル変数を誤って上書きすることなく、この関数内でローカル変数を使用しながら、次のようにグローバル変数にアクセスできることです。

var global = 'Hello, I am a global variable :)'; (function () { // We keep these variables private inside this closure scope var myGrades = [93, 95, 88, 0, 55, 91]; var average = function() { var total = myGrades.reduce(function(accumulator, item) { return accumulator + item}, 0); return 'Your average grade is ' + total / myGrades.length + '.'; } var failing = function(){ var failingGrades = myGrades.filter(function(item) { return item < 70;}); return 'You failed ' + failingGrades.length + ' times.'; } console.log(failing()); console.log(global); }()); // 'You failed 2 times.' // 'Hello, I am a global variable :)'

キーワードfunctionで始まるステートメントは常に関数宣言と見なされるため、無名関数を括弧で囲む必要があることに注意してください(JavaScriptでは名前のない関数宣言を使用できないことに注意してください)。したがって、周囲の括弧は関数式を作成します。代わりに。興味がある場合は、こちらで詳細を読むことができます。

例2:グローバルインポート

Another popular approach used by libraries like jQuery is global import. It’s similar to the anonymous closure we just saw, except now we pass in globals as parameters:

(function (globalVariable) { // Keep this variables private inside this closure scope var privateFunction = function() { console.log('Shhhh, this is private!'); } // Expose the below methods via the globalVariable interface while // hiding the implementation of the method within the // function() block globalVariable.each = function(collection, iterator) { if (Array.isArray(collection)) { for (var i = 0; i < collection.length; i++) { iterator(collection[i], i, collection); } } else { for (var key in collection) { iterator(collection[key], key, collection); } } }; globalVariable.filter = function(collection, test) { var filtered = []; globalVariable.each(collection, function(item) { if (test(item)) { filtered.push(item); } }); return filtered; }; globalVariable.map = function(collection, iterator) { var mapped = []; globalUtils.each(collection, function(value, key, collection) { mapped.push(iterator(value)); }); return mapped; }; globalVariable.reduce = function(collection, iterator, accumulator) { var startingValueMissing = accumulator === undefined; globalVariable.each(collection, function(item) { if(startingValueMissing) { accumulator = item; startingValueMissing = false; } else { accumulator = iterator(accumulator, item); } }); return accumulator; }; }(globalVariable)); 

In this example, globalVariable is the only variable that’s global. The benefit of this approach over anonymous closures is that you declare the global variables upfront, making it crystal clear to people reading your code.

Example 3: Object interface

Yet another approach is to create modules using a self-contained object interface, like so:

var myGradesCalculate = (function () { // Keep this variable private inside this closure scope var myGrades = [93, 95, 88, 0, 55, 91]; // Expose these functions via an interface while hiding // the implementation of the module within the function() block return { average: function() { var total = myGrades.reduce(function(accumulator, item) { return accumulator + item; }, 0); return'Your average grade is ' + total / myGrades.length + '.'; }, failing: function() { var failingGrades = myGrades.filter(function(item) { return item < 70; }); return 'You failed ' + failingGrades.length + ' times.'; } } })(); myGradesCalculate.failing(); // 'You failed 2 times.' myGradesCalculate.average(); // 'Your average grade is 70.33333333333333.'

As you can see, this approach lets us decide what variables/methods we want to keep private (e.g. myGrades) and what variables/methods we want to expose by putting them in the return statement (e.g. average & failing).

Example 4: Revealing module pattern

This is very similar to the above approach, except that it ensures all methods and variables are kept private until explicitly exposed:

var myGradesCalculate = (function () { // Keep this variable private inside this closure scope var myGrades = [93, 95, 88, 0, 55, 91]; var average = function() { var total = myGrades.reduce(function(accumulator, item) { return accumulator + item; }, 0); return'Your average grade is ' + total / myGrades.length + '.'; }; var failing = function() { var failingGrades = myGrades.filter(function(item) { return item < 70; }); return 'You failed ' + failingGrades.length + ' times.'; }; // Explicitly reveal public pointers to the private functions // that we want to reveal publicly return { average: average, failing: failing } })(); myGradesCalculate.failing(); // 'You failed 2 times.' myGradesCalculate.average(); // 'Your average grade is 70.33333333333333.'

That may seem like a lot to take in, but it’s just the tip of the iceberg when it comes to module patterns. Here are a few of the resources I found useful in my own explorations:

  • Learning JavaScript Design Patterns by Addy Osmani: a treasure trove of details in an impressively succinct read
  • Adequately Good by Ben Cherry: a useful overview with examples of advanced usage of the module pattern
  • Blog of Carl Danley: module pattern overview and resources for other JavaScript patterns.

CommonJS and AMD

The approaches above all have one thing in common: the use of a single global variable to wrap its code in a function, thereby creating a private namespace for itself using a closure scope.

While each approach is effective in its own way, they have their downsides.

For one, as a developer, you need to know the right dependency order to load your files in. For instance, let’s say you’re using Backbone in your project, so you include the script tag for Backbone’s source code in your file.

However, since Backbone has a hard dependency on Underscore.js, the script tag for the Backbone file can’t be placed before the Underscore.js file.

As a developer, managing dependencies and getting these things right can sometimes be a headache.

Another downside is that they can still lead to namespace collisions. For example, what if two of your modules have the same name? Or what if you have two versions of a module, and you need both?

So you’re probably wondering: can we design a way to ask for a module’s interface without going through the global scope?

Fortunately, the answer is yes.

There are two popular and well-implemented approaches: CommonJS and AMD.

CommonJS

CommonJS is a volunteer working group that designs and implements JavaScript APIs for declaring modules.

A CommonJS module is essentially a reusable piece of JavaScript which exports specific objects, making them available for other modules to require in their programs. If you’ve programmed in Node.js, you’ll be very familiar with this format.

With CommonJS, each JavaScript file stores modules in its own unique module context (just like wrapping it in a closure). In this scope, we use the module.exports object to expose modules, and require to import them.

When you’re defining a CommonJS module, it might look something like this:

function myModule() { this.hello = function() { return 'hello!'; } this.goodbye = function() { return 'goodbye!'; } } module.exports = myModule;

We use the special object module and place a reference of our function into module.exports. This lets the CommonJS module system know what we want to expose so that other files can consume it.

Then when someone wants to use myModule, they can require it in their file, like so:

var myModule = require('myModule'); var myModuleInstance = new myModule(); myModuleInstance.hello(); // 'hello!' myModuleInstance.goodbye(); // 'goodbye!'

There are two obvious benefits to this approach over the module patterns we discussed before:

1. Avoiding global namespace pollution

2. Making our dependencies explicit

Moreover, the syntax is very compact, which I personally love.

Another thing to note is that CommonJS takes a server-first approach and synchronously loads modules. This matters because if we have three other modules we need to require, it’ll load them one by one.

Now, that works great on the server but, unfortunately, makes it harder to use when writing JavaScript for the browser. Suffice it to say that reading a module from the web takes a lot longer than reading from disk. For as long as the script to load a module is running, it blocks the browser from running anything else until it finishes loading. It behaves this way because the JavaScript thread stops until the code has been loaded. (I’ll cover how we can work around this issue in Part 2 when we discuss module bundling. For now, that’s all we need to know).

AMD

CommonJS is all well and good, but what if we want to load modules asynchronously? The answer is called Asynchronous Module Definition, or AMD for short.

Loading modules using AMD looks something like this:

define(['myModule', 'myOtherModule'], function(myModule, myOtherModule) { console.log(myModule.hello()); });

What’s happening here is that the define function takes as its first argument an array of each of the module’s dependencies. These dependencies are loaded in the background (in a non-blocking manner), and once loaded define calls the callback function it was given.

Next, the callback function takes, as arguments, the dependencies that were loaded — in our case, myModule and myOtherModule — allowing the function to use these dependencies. Finally, the dependencies themselves must also be defined using the define keyword.

For example, myModule might look like this:

define([], function() { return { hello: function() { console.log('hello'); }, goodbye: function() { console.log('goodbye'); } }; });

So again, unlike CommonJS, AMD takes a browser-first approach alongside asynchronous behavior to get the job done. (Note, there are a lot of people who strongly believe that dynamically loading files piecemeal as you start to run code isn’t favorable, which we’ll explore more when in the next section on module-building).

Aside from asynchronicity, another benefit of AMD is that your modules can be objects, functions, constructors, strings, JSON and many other types, while CommonJS only supports objects as modules.

That being said, AMD isn’t compatible with io, filesystem, and other server-oriented features available via CommonJS, and the function wrapping syntax is a bit more verbose compared to a simple require statement.

UMD

For projects that require you to support both AMD and CommonJS features, there’s yet another format: Universal Module Definition (UMD).

UMD essentially creates a way to use either of the two, while also supporting the global variable definition. As a result, UMD modules are capable of working on both client and server.

Here’s a quick taste of how UMD goes about its business:

(function (root, factory) { if (typeof define === 'function' && define.amd) { // AMD define(['myModule', 'myOtherModule'], factory); } else if (typeof exports === 'object') { // CommonJS module.exports = factory(require('myModule'), require('myOtherModule')); } else { // Browser globals (Note: root is window) root.returnExports = factory(root.myModule, root.myOtherModule); } }(this, function (myModule, myOtherModule) { // Methods function notHelloOrGoodbye(){}; // A private method function hello(){}; // A public method because it's returned (see below) function goodbye(){}; // A public method because it's returned (see below) // Exposed public methods return { hello: hello, goodbye: goodbye } }));

For more examples of UMD formats, check out this enlightening repo on GitHub.

Native JS

Phew! Are you still around? I haven’t lost you in the woods here? Good! Because we have *one more* type of module to define before we’re done.

As you probably noticed, none of the modules above were native to JavaScript. Instead, we’ve created ways to emulate a modules system by using either the module pattern, CommonJS or AMD.

Fortunately, the smart folks at TC39 (the standards body that defines the syntax and semantics of ECMAScript) have introduced built-in modules with ECMAScript 6 (ES6).

ES6 offers up a variety of possibilities for importing and exporting modules which others have done a great job explaining — here are a few of those resources:

  • jsmodules.io
  • exploringjs.com

What’s great about ES6 modules relative to CommonJS or AMD is how it manages to offer the best of both worlds: compact and declarative syntax and asynchronous loading, plus added benefits like better support for cyclic dependencies.

Probably my favorite feature of ES6 modules is that imports are live read-only views of the exports. (Compare this to CommonJS, where imports are copies of exports and consequently not alive).

Here’s an example of how that works:

// lib/counter.js var counter = 1; function increment() { counter++; } function decrement() { counter--; } module.exports = { counter: counter, increment: increment, decrement: decrement }; // src/main.js var counter = require('../../lib/counter'); counter.increment(); console.log(counter.counter); // 1

In this example, we basically make two copies of the module: one when we export it, and one when we require it.

Moreover, the copy in main.js is now disconnected from the original module. That’s why even when we increment our counter it still returns 1 — because the counter variable that we imported is a disconnected copy of the counter variable from the module.

So, incrementing the counter will increment it in the module, but won’t increment your copied version. The only way to modify the copied version of the counter variable is to do so manually:

counter.counter++; console.log(counter.counter); // 2

On the other hand, ES6 creates a live read-only view of the modules we import:

// lib/counter.js export let counter = 1; export function increment() { counter++; } export function decrement() { counter--; } // src/main.js import * as counter from '../../counter'; console.log(counter.counter); // 1 counter.increment(); console.log(counter.counter); // 2

Cool stuff, huh? What I find really compelling about live read-only views is how they allow you to split your modules into smaller pieces without losing functionality.

Then you can turn around and merge them again, no problem. It just “works.”

Looking forward: bundling modules

Wow! Where does the time go? That was a wild ride, but I sincerely hope it gave you a better understanding of modules in JavaScript.

In the next section I’ll walk through module bundling, covering core topics including:

  • Why we bundle modules
  • Different approaches to bundling
  • ECMAScript’s module loader API
  • …and more. :)

NOTE: To keep things simple, I skipped over some of the nitty-gritty details (think: cyclic dependencies) in this post. If I left out anything important and/or fascinating, please let me know in the comments!