When should I use arrow functions in ECMAScript 6?

Question

With () => {} and function () {} we are getting two very similar ways to write functions in ES6. In other languages lambda functions often distinguish themselves by being anonymous, but in ECMAScript any function can be anonymous. Each of the two types have unique usage domains (namely when this needs to either be bound explicitly or explicitly not be bound). Between those domains there are a vast number of cases where either notation will do.

Arrow functions in ES6 have at least two limitations:

• Don't work with new and cannot be used when creating prototype
• Fixed this bound to scope at initialisation

These two limitations aside, arrow functions could theoretically replace regular functions almost anywhere. What is the right approach using them in practice? Should arrow functions be used e.g.:

• "everywhere they work", i.e. everywhere a function does not have to be agnostic about the this variable and we are not creating an object.
• only "everywhere they are needed", i.e. event listeners, timeouts, that need to be bound to a certain scope
• with 'short' functions, but not with 'long' functions
• only with functions that do not contain another arrow function

I am looking for a guideline to selecting the appropriate function notation in the future version of ECMAScript. The guideline will need to be clear, so that it can be taught to developers in a team, and to be consistent so that it does not require constant refactoring back and forth from one function notation to another.

_{The question is directed at people who have thought about code style in the context of the upcoming ECMAScript 6 (Harmony) and who have already worked with the language.}

Answer 1

A while ago our team migrated all its code (a mid-sized AngularJS app) to JavaScript compiled using Traceur Babel. I'm now using the following rule of thumb for functions in ES6 and beyond:

• Use function in the global scope and for Object.prototype properties.
• Use class for object constructors.
• Use => everywhere else.

Why use arrow functions almost everywhere?

1. Scope safety: When arrow functions are used consistently, everything is guaranteed to use the same thisObject as the root. If even a single standard function callback is mixed in with a bunch of arrow functions there's a chance the scope will become messed up.
2. Compactness: Arrow functions are easier to read and write. (This may seem opinionated so I will give a few examples further on.)
3. Clarity: When almost everything is an arrow function, any regular function immediately sticks out for defining the scope. A developer can always look up the next-higher function statement to see what the thisObject is.

Why always use regular functions on the global scope or module scope?

1. To indicate a function that should not access the thisObject.
2. The window object (global scope) is best addressed explicitly.
3. Many Object.prototype definitions live in the global scope (think String.prototype.truncate, etc.) and those generally have to be of type function anyway. Consistently using function on the global scope helps avoid errors.
4. Many functions in the global scope are object constructors for old-style class definitions.
5. Functions can be named¹. This has two benefits: (1) It is less awkward to writefunction foo(){} than const foo = () => {} — in particular outside other function calls. (2) The function name shows in stack traces. While it would be tedious to name every internal callback, naming all the public functions is probably a good idea.
6. Function declarations are hoisted, (meaning they can be accessed before they are declared), which is a useful attribute in a static utility function.

Object constructors

Attempting to instantiate an arrow function throws an exception:

var x = () => {}; new x(); // TypeError: x is not a constructor 

One key advantage of functions over arrow functions is therefore that functions double as object constructors:

function Person(name) {     this.name = name; } 

However, the functionally identical² ECMAScript Harmony draft class definition is almost as compact:

class Person {     constructor(name) {         this.name = name;     } } 

I expect that use of the former notation will eventually be discouraged. The object constructor notation may still be used by some for simple anonymous object factories where objects are programmatically generated, but not for much else.

Where an object constructor is needed one should consider converting the function to a class as shown above. The syntax works with anonymous functions/classes as well.

Readability of arrow functions

The probably best argument for sticking to regular functions - scope safety be damned - would be that arrow functions are less readable than regular functions. If your code is not functional in the first place, then arrow functions may not seem necessary, and when arrow functions are not used consistently they look ugly.

ECMAScript has changed quite a bit since ECMAScript 5.1 gave us the functional Array.forEach, Array.map and all of these functional programming features that have us use functions where for loops would have been used before. Asynchronous JavaScript has taken off quite a bit. ES6 will also ship a Promise object, which means even more anonymous functions. There is no going back for functional programming. In functional JavaScript, arrow functions are preferable over regular functions.

Take for instance this (particularly confusing) piece of code³:

function CommentController(articles) {     this.comments = [];      articles.getList()         .then(articles => Promise.all(articles.map(article => article.comments.getList())))         .then(commentLists => commentLists.reduce((a, b) => a.concat(b)));         .then(comments => {             this.comments = comments;         }) } 

The same piece of code with regular functions:

function CommentController(articles) {     this.comments = [];      articles.getList()         .then(function (articles) {             return Promise.all(articles.map(function (article) {                 return article.comments.getList();             }));         })         .then(function (commentLists) {             return commentLists.reduce(function (a, b) {                 return a.concat(b);             });         })         .then(function (comments) {             this.comments = comments;         }.bind(this)); } 

While any one of the arrow functions can be replaced by a standard function, there would be very little to gain from doing so. Which version is more readable? I would say the first one.

I think the question whether to use arrow functions or regular functions will become less relevant over time. Most functions will either become class methods, which make away with the function keyword, or they will become classes. Functions will remain in use for patching classes through the Object.prototype. In the mean time I suggest reserving the function keyword for anything that should really be a class method or a class.

---

Notes

1. Named arrow functions have been deferred in the ES6 specification. They might still be added a future version.
2. According to the draft specification, "Class declarations/expressions create a constructor function/prototype pair exactly as for function declarations" as long as a class does not use the extend keyword. A minor difference is that class declarations are constants, whereas function declarations are not.
3. Note on blocks in single statement arrow functions: I like to use a block wherever an arrow function is called for the side effect alone (e.g., assignment). That way it is clear that the return value can be discarded.

Answer 2

According to the proposal, arrows aimed "to address and resolve several common pain points of traditional function expressions". They intended to improve matters by binding this lexically and offering terse syntax.

However,

• One cannot consistently bind this lexically
• Arrow function syntax is delicate and ambiguous

Therefore, arrow functions create opportunities for confusion and errors, and should be excluded from a JavaScript programmer's vocabulary, replaced with function exclusively.

Regarding lexical this

this is problematic:

function Book(settings) {     this.settings = settings;     this.pages = this.createPages(); } Book.prototype.render = function () {     this.pages.forEach(function (page) {         page.draw(this.settings);     }, this); }; 

Arrow functions intend to fix the problem where we need to access a property of this inside a callback. There are already several ways to do that: One could assign this to a variable, use bind, or use the third argument available on the Array aggregate methods. Yet arrows seem to be the simplest workaround, so the method could be refactored like this:

this.pages.forEach(page => page.draw(this.settings)); 

However, consider if the code used a library like jQuery, whose methods bind this specially. Now, there are two this values to deal with:

Book.prototype.render = function () {     var book = this;     this.$pages.each(function (index) {         var $page = $(this);         book.draw(book.currentPage + index, $page);     }); }; 

We must use function in order for each to bind this dynamically. We can't use an arrow function here.

Dealing with multiple this values can also be confusing, because it's hard to know which this an author was talking about:

function Reader() {     this.book.on('change', function () {         this.reformat();     }); } 

Did the author actually intend to call Book.prototype.reformat? Or did he forget to bind this, and intend to call Reader.prototype.reformat? If we change the handler to an arrow function, we will similarly wonder if the author wanted the dynamic this, yet chose an arrow because it fit on one line:

function Reader() {     this.book.on('change', () => this.reformat()); } 

One may pose: "Is it exceptional that arrows could sometimes be the wrong function to use? Perhaps if we only rarely need dynamic this values, then it would still be okay to use arrows most of the time."

But ask yourself this: "Would it be 'worth it' to debug code and find that the result of an error was brought upon by an 'edge case?'" I'd prefer to avoid trouble not just most of the time, but 100% of the time.

There is a better way: Always use function (so this can always be dynamically bound), and always reference this via a variable. Variables are lexical and assume many names. Assigning this to a variable will make your intentions clear:

function Reader() {     var reader = this;     reader.book.on('change', function () {         var book = this;         book.reformat();         reader.reformat();     }); } 

Furthermore, always assigning this to a variable (even when there is a single this or no other functions) ensures one's intentions remain clear even after the code is changed.

Also, dynamic this is hardly exceptional. jQuery is used on over 50 million websites (as of this writing in February 2016). Here are other APIs binding this dynamically:

• Mocha (~120k downloads yesterday) exposes methods for its tests via this.
• Grunt (~63k downloads yesterday) exposes methods for build tasks via this.
• Backbone (~22k downloads yesterday) defines methods accessing this.
• Event APIs (like the DOM's) refer to an EventTarget with this.
• Prototypal APIs that are patched or extended refer to instances with this.

(Statistics via http://trends.builtwith.com/javascript/jQuery and https://www.npmjs.com.)

You are likely to require dynamic this bindings already.

A lexical this is sometimes expected, but sometimes not; just as a dynamic this is sometimes expected, but sometimes not. Thankfully, there is a better way, which always produces and communicates the expected binding.

Regarding terse syntax

Arrow functions succeeded in providing a "shorter syntactical form" for functions. But will these shorter functions make you more successful?

Is x => x * x "easier to read" than function (x) { return x * x; }? Maybe it is, because it's more likely to produce a single, short line of code. According to Dyson's The influence of reading speed and line length on the effectiveness of reading from screen,

A medium line length (55 characters per line) appears to support effective reading at normal and fast speeds. This produced the highest level of comprehension . . .

Similar justifications are made for the conditional (ternary) operator, and for single-line if statements.

However, are you really writing the simple mathematical functions advertised in the proposal? My domains are not mathematical, so my subroutines are rarely so elegant. Rather, I commonly see arrow functions break a column limit, and wrap to another line due to the editor or style guide, which nullifies "readability" by Dyson's definition.

One might pose, "How about just using the short version for short functions, when possible?". But now a stylistic rule contradicts a language constraint: "Try to use the shortest function notation possible, keeping in mind that sometimes only the longest notation will bind this as expected." Such conflation makes arrows particularly prone to misuse.

There are numerous issues with arrow function syntax:

const a = x =>     doSomething(x);  const b = x =>     doSomething(x);     doSomethingElse(x); 

Both of these functions are syntactically valid. But doSomethingElse(x); is not in the body of b. It is just a poorly-indented, top-level statement.

When expanding to the block form, there is no longer an implicit return, which one could forget to restore. But the expression may only have been intended to produce a side-effect, so who knows if an explicit return will be necessary going forward?

const create = () => User.create();  const create = () => {     let user;     User.create().then(result => {         user = result;         return sendEmail();     }).then(() => user); };  const create = () => {     let user;     return User.create().then(result => {         user = result;         return sendEmail();     }).then(() => user); }; 

What may be intended as a rest parameter can be parsed as the spread operator:

processData(data, ...results => {}) // Spread processData(data, (...results) => {}) // Rest 

Assignment can be confused with default arguments:

const a = 1; let x; const b = x => {}; // No default const b = x = a => {}; // "Adding a default" instead creates a double assignment const b = (x = a) => {}; // Remember to add parentheses 

Blocks look like objects:

(id) => id // Returns `id` (id) => {name: id} // Returns `undefined` (it's a labeled statement) (id) => ({name: id}) // Returns an object 

What does this mean?

() => {} 

Did the author intend to create a no-op, or a function that returns an empty object? (With this in mind, should we ever place { after =>? Should we restrict ourselves to the expression syntax only? That would further reduce arrows' frequency.)

=> looks like <= and >=:

x => 1 ? 2 : 3 x <= 1 ? 2 : 3  if (x => 1) {} if (x >= 1) {} 

To invoke an arrow function expression immediately, one must place () on the outside, yet placing () on the inside is valid and could be intentional.

(() => doSomething()()) // Creates function calling value of `doSomething()` (() => doSomething())() // Calls the arrow function 

Although, if one writes (() => doSomething()()); with the intention of writing an immediately-invoked function expression, simply nothing will happen.

It's hard to argue that arrow functions are "more understandable" with all the above cases in mind. One could learn all the special rules required to utilize this syntax. Is it really worth it?

The syntax of function is unexceptionally generalized. To use function exclusively means the language itself prevents one from writing confusing code. To write procedures that should be syntactically understood in all cases, I choose function.

Regarding a guideline

You request a guideline that needs to be "clear" and "consistent." Using arrow functions will eventually result in syntactically-valid, logically-invalid code, with both function forms intertwined, meaningfully and arbitrarily. Therefore, I offer the following:

Guideline for Function Notation in ES6:

• Always create procedures with function.
• Always assign this to a variable. Do not use () => {}.