Skip to content

Latest commit

 

History

History
1270 lines (836 loc) · 60.2 KB

ch4.md

File metadata and controls

1270 lines (836 loc) · 60.2 KB
Raw

You Don't Know JS Yet: Objects & Classes - 2nd Edition

Chapter 4: This Works

NOTE:
Work in progress

We've seen the this keyword used quite a bit so far, but haven't really dug in to understand exactly how it works in JS. It's time we do so.

But to properly understand this in JS, you need to set aside any preconceptions you may have, especially assumptions from how this works in other programming languages you may have experience in.

Here's the most important thing to understand about this: the determination of what value (usually, object) this points at is not made at author time, but rather determined at runtime. That means you cannot simply look at a this-aware function (even a method in a class definition) and know for sure what this will hold while that function runs.

Instead, you have to find each place the function is invoked, and look at how it's invoked (not even where matters). That's the only way to fully answer what this will point to.

In fact, a single this-aware function can be invoked at least four different ways, and any of those approaches will end up assigning a different this for that particular function invocation.

So the typical question we might ask when reading code -- "What does this point to the function?" -- is not actually a valid question. The question you really have to ask is, "When the function is invoked a certain way, what this will be assigned for that invocation?"

If your brain is already twisting around just reading this chapter intro... good! Settle in for a rewiring of how you think about this in JS.

This Aware

I used the phrase this-aware just a moment ago. But what exactly do I mean by that?

Any function that has a this keyword in it.

If a function does not have this in it anywhere, then the rules of how this behaves don't affect that function in any way. But if it does have even a single this in it, then you absolutely cannot determine how the function will behave without figuring out, for each invocation of the function, what this will point to.

It's sort of like the this keyword is a placeholder in a template. That placeholder's value-replacement doesn't get determined when we author the code; it gets determined while the code is running.

You might think I'm just playing word games here. Of course, when you write the program, you write out all the calls to each function, so you've already determined what the this is going to be when you authored the code, right? Right!?

Not so fast!

First of all, you don't always write all the code that invokes your function(s). Your this-aware function(s) might be passed as a callback(s) to some other code, either in your code base, or in a third-party framework/utility, or even inside a native built-in mechanism of the language or environment that's hosting your program.

But even aside from passing functions as callbacks, several mechanisms in JS allow for conditional runtime behaviors to determine which value (again, usually object) will be set for the this of a particular function invocation. So even though you may have written all that code, you at best will have to mentally execute the different conditions/paths that end up affecting the function invocation.

And why does all this matter?

Because it's not just you, the author of the code, that needs to figure this stuff out. It's every single reader of your code, forever. If anyone (even your future self) wants to read a piece of code that defines a this-aware function, that inevitably means that, to fully understand and predict its behavior, that person will have to find, read, and understand every single invocation of that function.

This Confuses Me

Now, in fairness, that's already partially true if we consider a function's parameters. To understand how a function is going to work, we need to know what is being passed into it. So any function with at least one parameter is, in a similar sense, argument-aware -- meaning, what argument(s) is/are passed in and assigned to the parameter(s) of the function.

But with parameters, we often have a bit more of a hint from the function itself what the parameters will do and hold.

We often see the names of the parameters declared right in the function header, which goes a long way to explaining their nature/purpose. And if there are defaults for the parameters, we often see them declared inline with = whatever clauses. Moreover, depending on the code style of the author, we may see in the first several lines of the function a set of logic that applies to these parameters; this could be assertions about the values (disallowed values, etc), or even modifications (type conversion, formatting, etc).

Actually, this is very much like a parameter to a function, but it's an implicit parameter rather than an explicit one. You don't see any signal that this is going to be used, in the function header anywhere. You have to read the entire function body to see if this appears anywhere.

The "parameter" name is always this, so we don't get much of a hint as to its nature/purpose from such a general name. In fact, there's historically a lot of confusion of what "this" even is supposed to mean. And we rarely see much if anything done to validate/convert/etc the this value applied to a function invocation. In fact, virtually all this-aware code I've seen just neatly assumes the this "parameter" is holding exactly what value is expected. Talk about a trap for unexpected bugs!

So What Is This?

If this is an implicit parameter, what's its purpose? What's being passed in?

Hopefully you have already read the "Scope & Closures" book of this series. If not, I strongly encourage you to circle back and read that one once you've finished this one. In that book, I explained at length how scopes (and closures!) work, an especially important characteristic of functions.

Lexical scope (including all the variables closed over) represents a static context for the function's lexical identifier references to be evaluated against. It's fixed/static because at author time, when you place functions and variable declarations in various (nested) scopes, those decisions are fixed, and unaffected by any runtime conditions.

By contrast, a different programming language might offer dynamic scope, where the context for a function's variable references is not determined by author-time decisions but by runtime conditions. Such a system would be undoubtedly more flexible than static context -- though with flexibility often comes complexity.

To be clear: JS scope is always and only lexical and static (if we ignore non-strict mode cheats like eval(..) and with). However, one of the truly powerful things about JS is that it offers another mechanism with similar flexibility and capabilities to dynamic scope.

The this mechanism is, effectively, dynamic context (not scope); it's how a this-aware function can be dynamically invoked against different contexts -- something that's impossible with closure and lexical scope identifiers!

Why Is This So Implicit?

You might wonder why something as important as a dynamic context is handled as an implicit input to a function, rather than being an explicit argument passed in.

That's a very important question, but it's not one we can quite answer, yet. Hold onto that question though.

Can We Get On With This?

So why have I belabored this subject for a couple of pages now? You get it, right!? You're ready to move on.

My point is, you the author of code, and all other readers of the code even years or decades in the future, need to be this-aware. That's the choice, the burden, you place on the reading of such code. And yes, that goes for the choice to use class (see Chapter 3), as most class methods will be this-aware out of necessity.

Be aware of this this choice in code you write. Do it intentionally, and do it in such a way as to produce more outcome benefit than burden. Make sure this usage in your code carries its own weight.

Let me put it this way: don't use this-aware code unless you really can justify it, and you've carefully weighed the costs. Just because you've seen a lot of code examples slinging around this in others' code, doesn't mean that this belongs in this code you're writing.

The this mechanism in JS, paired with [[Prototype]] delegation, is an extremely powerful pillar of the language. But as the cliche goes: "with great power comes great responsibility". Anecdotally, even though I really like and appreciate this pillar of JS, probably less than 5% of the JS code I ever write uses it. And when I do, it's with restraint. It's not my default, go-to JS capability.

This Is It!

OK, enough of the wordy lecture. You're ready to dive into this code, right?

Let's revisit (and extend) Point2d from Chapter 3, but just as an object with data properties and functions on it, instead of using class:

var point = {
    x: null,
    y: null,

    init(x,y) {
        this.x = x;
        this.y = y;
    },
    rotate(angleRadians) {
        var rotatedX = this.x * Math.cos(angleRadians) -
            this.y * Math.sin(angleRadians);
        var rotatedY = this.x * Math.sin(angleRadians) +
            this.y * Math.cos(angleRadians);
        this.x = rotatedX;
        this.y = rotatedY;
    },
    toString() {
        return `(${this.x},${this.y})`;
    },
};

As you can see, the init(..), rotate(..), and toString() functions are this-aware. You might be in the habit of assuming that the this reference will obviously always hold the point object. But that's not guaranteed in any way.

Keep reminding yourself as you go through the rest of this chapter: the this value for a function is determined by how the function is invoked. That means you can't look at the function's definition, nor where the function is defined (not even the enclosing class!). In fact, it doesn't even matter where the function is called from.

We only need to look at how the functions are called; that's the only factor that matters.

Implicit Context Invocation

Consider this call:

point.init(3,4);

We're invoking the init(..) function, but notice the point. in front of it? This is an implicit context binding. It says to JS: invoke the init(..) function with this referencing point.

That is the normal way we'd expect a this to work, and that's also one of the most common ways we invoke functions. So the typical invocation gives us the intuitive outcome. That's a good thing!

Default Context Invocation

But what happens if we do this?

const init = point.init;
init(3,4);

You might assume that we'd get the same outcome as the previous snippet. But that's not how JS this assignment works.

The call-site for the function is init(3,4), which is different than point.init(3,4). When there's no implicit context (point.), nor any other kind of this assignment mechanism, the default context assignment occurs.

What will this reference when init(3,4) is invoked like that?

It depends.

Uh oh. Depends? That sounds confusing.

Don't worry, it's not as bad as it sounds. The default context assignment depends on whether the code is in strict-mode or not. But thankfully, virtually all JS code these days is running in strict-mode; for example, ESM (ES Modules) always run in strict-mode, as does code inside a class block. And virtually all transpiled JS code (via Babel, TypeScript, etc) is written to declare strict-mode.

So almost all of the time, modern JS code will be running in strict-mode, and thus the default assignment context won't "depend" on anything; it's pretty straightforward: undefined. That's it!

NOTE:
Keep in mind: undefined does not mean "not defined"; it means, "defined with the special empty undefined value". I know, I know... the name and meaning are mismatched. That's language legacy baggage, for you. (shrugging shoulders)

That means init(3,4), if run in strict-mode, would throw an exception. Why? Because the this.x reference in init(..) is a .x property access on undefined (i.e., undefined.x), which is not allowed:

"use strict";

var point = { /* .. */ };

const init = point.init;
init(3,4);
// TypeError: Cannot set properties of
// undefined (setting 'x')

Stop for a moment and consider: why would JS choose to default the context to undefined, so that any default context invocation of a this-aware function will fail with such an exception?

Because a this-aware function always needs a this. The invocation init(3,4) isn't providing a this, so that is a mistake, and should raise an exception so the mistake can be corrected. The lesson: never invoke a this-aware function without providing it a this!

Just for completeness sake: in the less common non-strict mode, the default context is the global object -- JS defines it as globalThis, which in browser JS is essentially an alias to window, and in Node it's global. So, when init(3,4) runs in non-strict mode, the this.x expression is globalThis.x -- also known as window.x in the browser, or global.x in Node. Thus, globalThis.x gets set as 3 and globalThis.y gets set as 4.

// no strict-mode here, beware!

var point = { /* .. */ };

const init = point.init;
init(3,4);

globalThis.x;   // 3
globalThis.y;   // 4
point.x;        // null
point.y;        // null

That's unfortunate, because it's almost certainly not the intended outcome. Not only is it bad if it's a global variable, but it's also not changing the property on our point object, so program bugs are guaranteed.

WARNING:
Ouch! Nobody wants accidental global variables implicitly created from all over the code. The lesson: always make sure your code is running in strict-mode!

Explicit Context Invocation

Functions can alternately be invoked with explicit context, using the built-in call(..) or apply(..) utilities:

var point = { /* .. */ };

const init = point.init;

init.call( point, 3, 4 );
// or: init.apply( point, [ 3, 4 ] )

point.x;        // 3
point.y;        // 4

init.call(point,3,4) is effectively the same as point.init(3,4), in that both of them assign point as the this context for the init(..) invocation.

NOTE:
Both call(..) and apply(..) utilities take as their first argument a this context value; that's almost always an object, but can technically can be any value (number, string, etc). The call(..) utility takes subsequent arguments and passes them through to the invoked function, whereas apply(..) expects its second argument to be an array of values to pass as arguments.

It might seem awkward to contemplate invoking a function with the explicit context assignment (call(..) / apply(..)) style in your program. But it's more useful than might be obvious at first glance.

Let's recall the original snippet:

var point = {
    x: null,
    y: null,

    init(x,y) {
        this.x = x;
        this.y = y;
    },
    rotate(angleRadians) { /* .. */ },
    toString() {
        return `(${this.x},${this.y})`;
    },
};

point.init(3,4);

var anotherPoint = {};
point.init.call( anotherPoint, 5, 6 );

point.x;                // 3
point.y;                // 4
anotherPoint.x;         // 5
anotherPoint.y;         // 6

Are you seeing what I did there?

I wanted to define anotherPoint, but I didn't want to repeat the definitions of those init(..) / rotate(..) / toString() functions from point. So I "borrowed" a function reference, point.init, and explicitly set the empty object anotherPoint as the this context, via call(..).

When init(..) is running at that moment, this inside it will reference anotherPoint, and that's why the x / y properties (values 5 / 6, respectively) get set there.

Any this-aware functions can be borrowed like this: point.rotate.call(anotherPoint, ..), point.toString.call(anotherPoint).

Revisiting Implicit Context Invocation

Another approach to share behavior between point and anotherPoint would have been:

var point = { /* .. */ };

var anotherPoint = {
    init: point.init,
    rotate: point.rotate,
    toString: point.toString,
};

anotherPoint.init(5,6);

anotherPoint.x;         // 5
anotherPoint.y;         // 6

This is another way of "borrowing" the functions, by adding shared references to the functions on any target object (e.g., anotherPoint). The call-site invocation anotherPoint.init(5,6) is the more natural/ergonomic style that relies on implicit context assignment.

It may seem this approach is a little cleaner, comparing anotherPoint.init(5,6) to point.init.call(anotherPoint,5,6).

But the main downside is having to modify any target object with such shared function references, which can be verbose, manual, and error-prone. Sometimes such an approach is acceptable, but many other times, explicit context assignment with call(..) / apply(..) is more preferable.

New Context Invocation

We've so far seen three different ways of context assignment at the function call-site: default, implicit, and explicit.

A fourth way to call a function, and assign the this for that invocation, is with the new keyword:

var point = {
    // ..

    init: function() { /* .. */ }

    // ..
};

var anotherPoint = new point.init(3,4);

anotherPoint.x;     // 3
anotherPoint.y;     // 4
TIP:
This example has a bit of nuance to be explained. The init: function() { .. } form shown here -- specifically, a function expression assigned to a property -- is required for the function to be validly called with the new keyword. From previous snippets, the concise method form of init() { .. } defines a function that cannot be called with new.

You've typically seen new used with class for creating instances. But as an underlying mechanism of the JS language, new is not inherently a class operation.

In a sense, the new keyword hijacks a function and forces its behavior into a different mode than a normal invocation. Here are the 4 special steps that JS performs when a function is invoked with new:

  1. create a brand new empty object, out of thin air.

  2. link the [[Prototype]] of that new empty object to the function's .prototype object (see Chapter 2).

  3. invoke the function with the this context set to that new empty object.

  4. if the function doesn't return its own object value explicitly (with a return .. statement), assume the function call should instead return the new object (from steps 1-3).

WARNING:
Step 4 implies that if you new invoke a function that does return its own object -- like return { .. }, etc -- then the new object from steps 1-3 is not returned. That's a tricky gotcha to be aware of, in that it effectively discards that new object before the program has a chance to receive and store a reference to it. Essentially, new should never be used to invoke a function that has explicit return .. statement(s) in it.

To understand these 4 new steps more concretely, I'm going to illustrate them in code, as an alternate to using the new keyword:

// alternative to:
//   var anotherPoint = new point.init(3,4)

var anotherPoint;
// this is a bare block to hide local
// `let` declarations
{
    // (Step 1)
    let tmpObj = {};

    // (Step 2)
    Object.setPrototypeOf(
        tmpObj, point.init.prototype
    );
    // or: tmpObj.__proto__ = point.init.prototype

    // (Step 3)
    let res = point.init.call(tmpObj,3,4);

    // (Step 4)
    anotherPoint = (
        typeof res !== "object" ? tmpObj : res
    );
}

Clearly, the new invocation streamlines that set of manual steps!

TIP:
The Object.setPrototypeOf(..) in step 2 could also have been done via the __proto__ property, such as tmpObj.__proto__ = point.init.prototype, or even as part of the object literal (step 1) with tmpObj = { __proto__: point.init.prototype }.

Skipping some of the formality of these steps, let's recall an earlier snippet and see how new approximates a similar outcome:

var point = { /* .. */ };

// this approach:
var anotherPoint = {};
point.init.call(anotherPoint,5,6);

// can instead be approximated as:
var yetAnotherPoint = new point.init(5,6);

That's a bit nicer! But there's a caveat here.

Using the other functions that point holds against anotherPoint / yetAnotherPoint, we won't want to do with new. Why? Because new is creating a new object, but that's not what we want if we intend to invoke a function against an existing object.

Instead, we'll likely use explicit context assignment:

point.rotate.call( anotherPoint, /*angleRadians=*/Math.PI );

point.toString.call( yetAnotherPoint );
// (5,6)

Review This

We've seen four rules for this context assignment in function calls. Let's put them in order of precedence:

  1. Is the function invoked with new, creating and setting a new this?

  2. Is the function invoked with call(..) or apply(..), explicitly setting this?

  3. Is the function invoked with an object reference at the call-site (e.g., point.init(..)), implicitly setting this?

  4. If none of the above... are we in non-strict mode? If so, default the this to globalThis. But if in strict-mode, default the this to undefined.

These rules, in this order, are how JS determines the this for a function invocation. If multiple rules match a call-site (e.g., new point.init.call(..)), the first rule from the list to match wins.

That's it, you're now master over the this keyword. Well, not quite. There's a bunch more nuance to cover. But you're well on your way!

An Arrow Points Somewhere

Everything I've asserted so far about this in functions, and how its determined based on the call-site, makes one giant assumption: that you're dealing with a regular function (or method).

So what's an irregular function?!? It looks like this:

const x = x => x <= x;
NOTE:
Yes, I'm being a tad sarcastic and unfair to call an arrow function "irregular" and to use such a contrived example. It's a joke, ok?

Here's a real example of an => arrow function:

const clickHandler = evt =>
    evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation();

For comparison sake, let me also show the non-arrow equivalent:

const clickHandler = function(evt) {
    evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation();
};

Or if we went a bit old-school about it -- this is my jam! -- we could try the standalone function declaration form:

function clickHandler(evt) {
    evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation();
}

Or if the function appeared as a method in a class definition, or as a concise method in an object literal, it would look like this:

// ..
clickHandler(evt) {
    evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation();
}

What I really want to focus on is how each of these forms of the function will behave with respect to their this reference, and whether the first => form differs from the others (hint: it does!). But let's start with a little quiz to see if you've been paying attention.

For each of those function forms just shown, how do we know what each this will reference?

Where's The Call-site?

Hopefully, you responded with something like: "first, we need to see how the functions are called."

Fair enough.

Let's say our program looks like this:

var infoForm = {
    theFormElem: null,
    theSubmitBtn: null,

    init() {
        this.theFormElem =
            document.getElementById("the-info-form");
        this.theSubmitBtn =
            theFormElem.querySelector("button[type=submit]");

        // is *this* the call-site?
        this.theSubmitBtn.addEventListener(
            "click",
            this.clickHandler,
            false
        );
    },

    // ..
}

Ah, interesting. Half of you readers have never seen actual DOM API code like getElementById(..), querySelector(..), and addEventListener(..) before. I heard the confusion bells whistle just now!

NOTE:
Sorry, I'm dating myself, here. I've been doing this stuff long enough that I remember when we did that kind of code long before we had utilities like jQuery cluttering up the code with $ everywhere. And after many years of front-end evolution, we seem to have landed somewhere quite a bit more "modern" -- at least, that's the prevailing presumption.

I'm guessing many of you these days are used to seeing component-framework code (React, etc) somewhat like this:

// ..

infoForm(props) {
    return (
        <form ref={this.theFormElem}>
            <button type=submit onClick=this.clickHandler>
                Click Me
            </button>
        </form>
    );
}

// ..

Of course, there's a bunch of other ways that code might be shaped, depending on if you're using one framework or another, etc.

Or maybe you're not even using class / this style components anymore, because you've moved everything to hooks and closures. In any case, for our discussion purposes, this chapter is all about this, so we need to stick to a coding style like the above, to have code related to the discussion.

And neither of those two previous code snippets show the clickHandler function being defined. But I've said repeatedly so far, that doesn't matter; all that matters is ... what? say it with me... all that matters is how the function is invoked.

So how is clickHandler being invoked? What's the call-site, and which context assignment rule does it match?

Hidden From Sight

If you're stuck, don't worry. I'm deliberately making this difficult, to point something very important out.

When the "click" or onClick= event handler bindings happen, in both cases, we specified this.clickHandler, which implies that there is a this context object with a property on it called clickHandler, which is holding our function definition.

So, is this.clickHandler the call-site? If it was, what assignment rule applies? The implicit context rule (#3)?

Unfortunately, no.

The problem is, we cannot actually see the call-site in this program. Uh oh.

If we can't see the call-site, how do we know how the function is going to actually get called?

That's the exact point I'm making.

It doesn't matter that we passed in this.clickHandler. That is merely a reference to a function object value. It's not a call-site.

Under the covers, somewhere inside a framework, library, or even the JS environment itself, when a user clicks the button, a reference to the clickHandler(..) function is going to be invoked. And as we've implied, that call-site is even going to pass in the DOM event object as the evt argument.

Since we can't see the call-site, we have to imagine it. Might it look like...?

// ..
eventCallback( domEventObj );
// ..

If it did, which this rule would apply? The default context rule (#4)?

Or, what if the call-site looked like this...?

// ..
eventCallback.call( domElement, domEventObj );

Now which this rule would apply? The explicit context rule (#2)?

Unless you open and view the source code for the framework/library, or read the documentation/specification, you won't know what to expect of that call-site. Which means that predicting, ultimately, what this points to in the clickHandler function you write, is... to put it mildly... a bit convoluted.

This Is Wrong

To spare you any more pain here, I'll cut to the chase.

Pretty much all implementations of a click-handler mechanism are going to do something like the .call(..), and they're going to set the DOM element (e.g., button) the event listener is bound to, as the explicit context for the invocation.

Hmmm... is that ok, or is that going to be a problem?

Recall that our clickHandler(..) function is this-aware, and that its this.theFormElem reference implies referencing an object with a theFormElem property, which in turn is pointing at the parent <form> element. DOM buttons do not, by default, have a theFormElem property on them.

In other words, the this reference that our event handler will have set for it is almost certainly wrong. Oops.

Unless we want to rewrite the clickHandler function, we're going to need to fix that.

Fixing this

Let's consider some options to address the mis-assignment. To keep things focused, I'll stick to this style of event binding for the discussion:

this.submitBtnaddEventListener(
    "click",
    this.clickHandler,
    false
);

Here's one way to address it:

// store a fixed reference to the current
// `this` context
var context = this;

this.submitBtn.addEventListener(
    "click",
    function handler(evt){
        return context.clickHandler(evt);
    },
    false
);
TIP:
Most older JS code that uses this approach will say something like var self = this instead of the context name I'm giving it here. "Self" is a shorter word, and sounds cooler. But it's also entirely the wrong semantic meaning. The this keyword is not a "self" reference to the function, but rather the context for that current function invocation. Those may seem like the same thing at a glance, but they're completely different concepts, as different as apples and a Beatles song. So... to paraphrase them, "Hey developer, don't make it bad. Take a sad self and make it better context."

What's going on here? I recognized that the enclosing code, where the addEventListener call is going to run, has a current this context that is correct, and we need to ensure that same this context is applied when clickHandler(..) gets invoked.

I defined a surrounding function (handler(..)) and then forced the call-site to look like:

context.clickHandler(evt);
TIP:
Which this context assignment rule is applied here? That's right, the implicit context rule (#3).

Now, it doesn't matter what the internal call-site of the library/framework/environment looks like. But, why?

Because we're now actually in control of the call-site. It doesn't matter how handler(..) gets invoked, or what its this is assigned. It only matters than when clickHandler(..) is invoked, the this context is set to what we wanted.

I pulled off that trick not only by defining a surrounding function (handler(..)) so I can control the call-site, but... and this is important, so don't miss it... I defined handler(..) as a NON-this-aware function! There's no this keyword inside of handler(..), so whatever this gets set (or not) by the library/framework/environment, is completely irrelevant.

The var context = this line is critical to the trick. It defines a lexical variable context, which is not some special keyword, holding a snapshot of the value in the outer this. Then inside clickHandler, we merely reference a lexical variable (context), no relative/magic this keyword.

Lexical This

The name for this pattern, by the way, is "lexical this", meaning a this that behaves like a lexical scope variable instead of like a dynamic context binding.

But it turns out JS has an easier way of performing the "lexical this" magic trick. Are you ready for the trick reveal!?

...

The => arrow function! Tada!

That's right, the => function is, unlike all other function forms, special, in that it's not special at all. Or, rather, that it doesn't define anything special for this behavior whatsoever.

In an => function, the this keyword... is not a keyword. It's absolutely no different from any other variable, like context or happyFace or foobarbaz.

Let me illustrate this point more directly:

function outer() {
    console.log(this.value);

    // define a return an "inner"
    // function
    var inner = () => {
        console.log(this.value);
    };

    return inner;
}

var one = {
    value: 42,
};
var two = {
    value: "sad face",
};

var innerFn = outer.call(one);
// 42

innerFn.call(two);
// 42   <-- not "sad face"

The innerFn.call(two) would, for any regular function definition, have resulted in "sad face" here. But since the inner function we defined and returned (and assigned to innerFn) was an irregular => arrow function, it has no special this behavior, but instead has "lexical this" behavior.

When the innerFn(..) (aka inner(..)) function is invoked, even with an explicit context assignment via .call(..), that assignment is ignored.

NOTE:
I'm not sure why => arrow functions even have a call(..) / apply(..) on them, since they are silent no-op functions. I guess it's for consistency with normal functions. But as we'll see later, there are other inconsistencies between regular functions and irregular => arrow functions.

When a this is encountered (this.value) inside an => arrow function, this is treated like a normal lexical variable, not a special keyword. And since there is no this variable in that function itself, JS does what it always does with lexical variables: it goes up one level of lexical scope -- in this case, to the surrounding outer(..) function, and it checks to see if there's any registered this in that scope.

Luckily, outer(..) is a regular function, which means it has a normal this keyword. And the outer.call(one) invocation assigned one to its this.

So, innerFn.call(two) is invoking inner(), but when inner() looks up a value for this, it gets... one, not two.

Back To The... Button

You thought I was going to make a pun joke and say "future" there, didn't you!?

A more direct and appropriate way of solving our earlier issue, where we had done var context = this to get a sort of faked "lexical this" behavior, is to use the => arrow function, since its primary design feature is... "lexical this".

this.submitBtn.addEventListener(
    "click",
    evt => this.clickHandler(evt),
    false
);

Boom! Problem solved! Mic drop!

Hear me on this: the => arrow function is not -- I repeat, not -- about typing fewer characters. The primary point of the => function being added to JS was to give us "lexical this" behavior without having to resort to var context = this (or worse, var self = this) style hacks.

TIP:
If you need "lexical this", always prefer an => arrow function. If you don't need "lexical this", well... the => arrow function might not be the best tool for the job.

Confession Time

I've said all along in this chapter, that how you write a function, and where you write the function, has nothing to do with how its this will be assigned.

For regular functions, that's true. But when we consider an irregular => arrow function, it's not entirely accurate anymore.

Recall the original => form of clickHandler from earlier in the chapter?

const clickHandler = evt =>
    evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation();

If we use that form, in the same context as our event binding, it could look like this:

const clickHandler = evt =>
    evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation();

this.submitBtn.addEventListener("click",clickHandler,false);

A lot of developers prefer to even further reduce it, to an inline => arrow function:

this.submitBtn.addEventListener(
    "click",
    evt => evt.target.matches("button") ?
        this.theFormElem.submit() :
        evt.stopPropagation(),
    false
);

When we write an => arrow function, we know for sure that its this binding will exactly be the current this binding of whatever surrounding function is running, regardless of what the call-site of the => arrow function looks like. So in other words, how we wrote the => arrow function, and where we wrote it, does matter.

That doesn't fully answer the this question, though. It just shifts the question to how the enclosing function was invoked. Actually, the focus on the call-site is still the only thing that matters.

But the nuance I'm confessing to having omitted until this moment is: it matters which call-site we consider, not just any call-site in the current call stack. The call-site that matters is, the nearest function-invocation in the current call stack that actually assigns a this context.

Since an => arrow function never has a this-assigning call-site (no matter what), that call-site isn't relevant to the question. We have to keep stepping up the call stack until we find a function invocation that is this-assigning -- even if such invoked function is not itself this-aware.

THAT is the only call-site that matters.

Find The Right Call-Site

Let me illustrate, with a convoluted mess of a bunch of nested functions/calls:

globalThis.value = { result: "Sad face" };

function one() {
    function two() {
        var three = {
            value: { result: "Hmmm" },

            fn: () => {
                const four = () => this.value;
                return four.call({
                    value: { result: "OK", },
                });
            },
        };
        return three.fn();
    };
    return two();
}

new one();          // ???

Can you run through that (nightmare) in your head and determine what will be returned from the new one() invocation?

It could be any of these:

// from `four.call(..)`:
{ result: "OK" }

// or, from `three` object:
{ result: "Hmmm" }

// or, from the `globalThis.value`:
{ result: "Sad face" }

// or, empty object from the `new` call:
{}

The call-stack for that new one() invocation is:

four         |
three.fn     |
two          | (this = globalThis)
one          | (this = {})
[ global ]   | (this = globalThis)

Since four() and fn() are both => arrow functions, the three.fn() and four.call(..) call-sites are not this-assigning; thus, they're irrelevant for our query. What's the next invocation to consider in the call-stack? two(). That's a regular function (it can accept this-assignment), and the call-site matches the default context assignment rule (#4). Since we're not in strict-mode, this is assigned globalThis.

When four() is running, this is just a normal variable. It looks then to its containing function (three.fn()), but it again finds a function with no this. So it goes up another level, and finds a two() regular function that has a this defined. And that this is globalThis. So the this.value expression resolves to globalThis.value, which returns us... { result: "Sad face" }.

...

Take a deep breath. I know that's a lot to mentally process. And in fairness, that's a super contrived example. You'll almost never see all that complexity mixed in one call-stack.

But you absolutely will find mixed call-stacks in real programs. You need to get comfortable with the analysis I just illustrated, to be able to unwind the call-stack until you find the most recent this-assigning call-site.

Remember the addage I quoted earlier: "with great power comes great responsibility". Choosing this-oriented code (even classes) means choosing both the flexibility it affords us, as well as needing to be comfortable navigating the call-stack to understand how it will behave.

That's the only way to effectively write (and later read!) this-aware code.

This Is Bound To Come Up

Backing up a bit, there's another option if you don't want to use an => arrow function's "lexical this" behavior to address the button event handler functionality.

In addition to call(..) / apply(..) -- these invoke functions, remember! -- JS functions also have a third utility built in, called bind(..) -- which does not invoke the function, just to be clear.

The bind(..) utility defines a new wrapped/bound version of a function, where its this is preset and fixed, and cannot be overridden with a call(..) or apply(..), or even an implicit context object at the call-site:

this.submitBtn.addEventListener(
    "click",
    this.clickHandler.bind(this),
    false
);

Since I'm passing in a this-bound function as the event handler, it similarly doesn't matter how that utility tries to set a this, because I've already forced the this to be what I wanted: the value of this from the surrounding function invocation context.

Hardly New

This pattern is often referred to as "hard binding", since we're creating a function reference that is strongly bound to a particular this. A lot of JS writings have claimed that the => arrow function is essentially just syntax for the bind(this) hard-binding. It's not. Let's dig in.

If you were going to create a bind(..) utility, it might look kinda like this:

function bind(fn,context) {
    return function bound(...args){
        return fn.apply(context,args);
    };
}
NOTE:
This is not actually how bind(..) is implemented. The behavior is more sophisticated and nuanced. I'm only illustrating one portion of its behavior in this snippet.

Does that look familiar? It's using the good ol' fake "lexical this" hack. And under the covers, it's an explicit context assignment, in this case via apply(..).

So wait... doesn't that mean we could just do it with an => arrow function?

function bind(fn,context) {
    return (...args) => fn.apply(context,args);
}

Eh... not quite. As with most things in JS, there's a bit of nuance. Let me illustrate:

// candidate implementation, for comparison
function fakeBind(fn,context) {
    return (...args) => fn.apply(context,args);
}

// test subject
function thisAwareFn() {
    console.log(`Value: ${this.value}`);
}

// control data
var obj = {
    value: 42,
};

// experiment
var f = thisAwareFn.bind(obj);
var g = fakeBind(thisAwareFn,obj);

f();            // Value: 42
g();            // Value: 42

new f();        // Value: undefined
new g();        // <--- ???

First, look at the new f() call. That's admittedly a strange usage, to call new on a hard-bound function. It's probably quite rare that you'd ever do so. But it shows something kind of interesting. Even though f() was hard-bound to a this context of obj, the new operator was able to hijack the hard-bound function's this and re-bind it to the newly created and empty object. That object has no value property, which is why we see "Value: undefined" printed out.

If that feels strange, I agree. It's a weird corner nuance. It's not something you'd likely ever exploit. But I point it out not just for trivia. Refer back to the four rules presented earlier in this chapter. Remember how I asserted their order-of-precedence, and new was at the top (#1), ahead of explicit call(..) / apply(..) assignment rule (#2)?

Since we can sort of think of bind(..) as a variation of that rule, we now see that order-of-precedence proven. new is more precedent than, and can override, even a hard-bound function. Sort of makes you think the hard-bound function is maybe not so "hard"-bound, huh?!

But... what's going to happen with the new g() call, which is invoking new on the returned => arrow function? Do you predict the same outcome as new f()?

Sorry to disappoint.

That line will actually throw an exception, because an => function cannot be used with the new keyword.

But why? My best answer, not being authoritative on TC39 myself, is that conceptually and actually, an => arrow function is not a function with a hard-bound this, it's a function that has no this at all. As such, new makes no sense against such a function, so JS just disallows it.

NOTE:
Recall earlier, when I pointed out that => arrow functions have call(..), apply(..), and indeed even a bind(..). But we've see that such functions basically ignore these utilities as no-ops. It's a bit strange, in my opinion, that => arrow functions have all those utilities as pass-through no-ops, but for the new keyword, that's not just, again, a no-op pass-through, but rather disallowed with an exception.

But the main point is: an => arrow function is not a syntactic form of bind(this).

Losing This Battle

Returning once again to our button event handler example:

this.submitBtnaddEventListener(
    "click",
    this.clickHandler,
    false
);

There's a deeper concern we haven't yet addressed.

We've seen several different approaches to construct a different callback function reference to pass in there, in place of this.clickHandler.

But whichever of those ways we choose, they are producing literally a different function, not just an in-place modification to our existing clickHandler function.

Why does that matter?

Well, first of all, the more functions we create (and re-create), the more processing time (very slight) and the more memory (pretty small, usually) we're chewing up. And when we're re-creating a function reference, and throwing an old one away, that's also leaving un-reclaimed memory sitting around, which puts pressure on the garbage collector (GC) to more often, pause the universe of our program momentarily while it cleans up and reclaims that memory.

If hooking up this event listening is a one-time operation, no big deal. But if it's happening over and over again, the system-level performance effects can start to add up. Ever had an otherwise smooth animation jitter? That was probably the GC kicking in, cleaning up a bunch of reclaimable memory.

But another concern is, for things like event handlers, if we're going to remove an event listener at some later time, we need to keep a reference to the exact same function we attached originally. If we're using a library/framework, often (but not always!) they take care of that little dirty-work detail for you. But otherwise, it's on us to make sure that whatever function we plan to attach, we hold onto a reference just in case we need it later.

So the point I'm making is: presetting a this assignment, no matter how you do it, so that it's predictable, comes with a cost. A system level cost and a program maintenance/complexity cost. It is never free.

One way of reacting to that fact is to decide, OK, we're just going to manufacture all those this-assigned function references once, ahead of time, up-front. That way, we're sure to reduce both the system pressure, and the code pressure, to a minimum.

Sounds reasonable, right? Not so fast.

Pre-Binding Function Contexts

If you have a one-off function reference that needs to be this-bound, and you use an => arrow or a bind(this) call, I don't see any problems with that.

But if most or all of the this-aware functions in a segment of your code invoked in ways where the this isn't the predictable context you expect, and so you decide you need to hard-bind them all... I think that's a big warning signal that you're going about things the wrong way.

Please recall the discussion in the "Avoid This" section from Chapter 3, which started with this snippet of code:

class Point2d {
    x = null
    getDoubleX = () => this.x * 2

    constructor(x,y) {
        this.x = x;
        this.y = y;
    }
    toString() { /* .. */ }
}

var point = new Point2d(3,4);

Now imagine we did this with that code:

const getX = point.getDoubleX;

// later, elsewhere

getX();         // 6

As you can see, the problem we were trying to solve is the same as we've been dealing with here in this chapter. It's that we wanted to be able to invoke a function reference like getX(), and have that mean and behave like point.getDoubleX(). But this rules on regular functions don't work that way.

So we used an => arrow function. No big deal, right!?

Wrong.

The real root problem is that we want two conflicting things out of our code, and we're trying to use the same hammer for both nails.

We want to have a this-aware method stored on the class prototype, so that there's only one definition for the function, and all our subclasses and instances nicely share that same function. And the way they all share is through the power of the dynamic this binding.

But at the same time, we also want those function references to magically stay this-assinged to our instance when we pass those function references around and other code is in charge of the call-site.

In other words, sometimes we want something like point.getDoubleX to mean, "give me a reference that's this-assigned to point", and other times we want the same expression point.getDoubleX to mean, give me a dynamic this-assignable function reference so it can properly get the context I need it to at this moment.

Perhaps JS could offer a different operator besides ., like :: or -> or something like that, which would let you distinguish what kind of function reference you're after. In fact, there's a long-standing proposal for a this-binding operator (::), that picks up attention from time to time, and then seems to stall out. Who knows, maybe someday such an operator will finally land, and we'll have better options.

But I strongly suspect that even if it does land someday, it's going to vend a whole new function reference, exactly as the => or bind(this) approaches we've already talked about. It won't come as a free and perfect solution. There will always be a tension between wanting the same function to sometimes be this-flexible and sometimes be this-predictable.

What JS authors of class-oriented code often run up against, sooner or later, is this exact tension. And you know what they do?

They don't consider the costs of simply pre-binding all the class's this-aware methods as instead => arrow functions in member properties. They don't realize that it's completely defeated the entire purpose of the [[Prototype]] chain. And they don't realize that if fixed-context is what they really need, there's an entirely different mechanism in JS that is better suited for that purpose.

Take A More Critical Look

So when you do this sort of thing:

class Point2d {
    x = null
    y = null
    getDoubleX = () => this.x * 2
    toString = () => `(${this.x},${this.y})`

    constructor(x,y) {
        this.x = x;
        this.y = y;
    }
}

var point = new Point2d(3,4);
var anotherPoint = new Point2d(5,6);

var f = point.getDoubleX;
var g = anotherPoint.toString;

f();            // 6
g();            // (5,6)

I say, "ick!", to the hard-bound this-aware methods getDoubleX() and toString() there. To me, that's a code smell. But here's an even worse approach that has been favored by many developers in the past:

class Point2d {
    x = null
    y = null

    constructor(x,y) {
        this.x = x;
        this.y = y;
        this.getDoubleX = this.getDoubleX.bind(this);
        this.toString = this.toString.bind(this);
    }
    getDoubleX() { return this.x * 2; }
    toString() { return `(${this.x},${this.y})`; }
}

var point = new Point2d(3,4);
var anotherPoint = new Point2d(5,6);

var f = point.getDoubleX;
var g = anotherPoint.toString;

f();            // 6
g();            // (5,6)

Double ick.

In both cases, you're using a this mechanism but completely betraying/neutering it, by taking away all the powerful dynamicism of this.

You really should at least be contemplating this alternate approach, which skips the whole this mechanism altogether:

function Point2d(px,py) {
    var x = px;
    var y = py;

    return {
        getDoubleX() { return x * 2; },
        toString() { return `(${x},${y})`; }
    };
}

var point = Point2d(3,4);
var anotherPoint = Point2d(5,6);

var f = point.getDoubleX;
var g = anotherPoint.toString;

f();            // 6
g();            // (5,6)

You see? No ugly or complex this to clutter up that code or worry about corner cases for. Lexical scope is super straightforward and intuitive.

When all we want is for most/all of our function behaviors to have a fixed and predictable context, the most appropriate solution, the most straightforward and even performant solution, is lexical variables and scope closure.

When you go to all to the trouble of sprinkling this references all over a piece of code, and then you cut off the whole mechanism at the knees with => "lexical this" or bind(this), you chose to make the code more verbose, more complex, more overwrought. And you got nothing out of it that was more beneficial, except to follow the this (and class) bandwagon.

...

Deep breath. Collect yourself.

I'm talking to myself, not you. But if what I just said bothers you, I'm talking to you, too!

OK, listen. That's just my opinion. If you don't agree, that's fine. But apply the same level of rigor to thinking about how these mechanisms work, as I have, when you decide what conclusions you want to arrive at.

Variations

Before we close out our lengthy discussion of this, there's a few irregular variations on function calls that we should discuss.

Indirect Function Calls

Recall this example from earlier in the chapter?

var point = {
    x: null,
    y: null,

    init(x,y) {
        this.x = x;
        this.y = y;
    },
    rotate(angleRadians) { /* .. */ },
    toString() { /* .. */ },
};

var init = point.init;
init(3,4);                  // broken!

This is broken because the init(3,4) call-site doesn't provide the necessary this-assignment signal. But there's other ways to observe a similar breakage. For example:

(1,point.init)(3,4);        // broken!

This strange looking syntax is first evaluating an expression (1,point.init), which is a comma series expression. The result of such an expression is the final evaluated value, which in this case is the function reference (held by point.init).

So the outcome puts that function reference onto the expression stack, and then invokes that value with (3,4). That's an indirect invocation of the function. And what's the result? It actually matches the default context assignment rule (#4) we looked at earlier in the chapter.

Thus, in non-strict mode, the this for the point.init(..) call will be globalThis. Had we been in strict-mode, it would have been undefined, and the this.x = x operation would then have thrown an exception for invalidly accessing the x property on the undefined value.

There's several different ways to get an indirect function invocation. For example:

(()=>point.init)()(3,4);    // broken!

And another example of indirect function invocation is the Immediately Invoked Function Expression (IIFE) pattern:

(function(){
    // `this` assigned via "default" rule
})();

As you can see, the function expression value is put onto the expression stack, and then it's invoked with the () on the end.

But what about this code:

(point.init)(3,4);

What will be the outcome of that code?

By the same reasoning we've seen in the previous examples, it stands to reason that the point.init expression puts the function value onto the expression stack, and then invoked indirectly with (3,4).

Not quite, though! JS grammar has a special rule to handle the invocation form (someIdentifier)(..) as if it had been someIdentifier(..) (without the (..) around the identifier name).

Wondering why you might want to ever force the default context for this assignment via an indirect function invocation?

Accessing globalThis

Before we answer that, let's introduce another way of performing indirect function this assignment. Thus far, the indirect function invocation patterns shown are sensitive to strict-mode. But what if we wanted an indirect function this assignment that doesn't respect strict-mode.

The Function(..) constructor takes a string of code and dynamically defines the equivalent function. However, it always does so as if that function had been declared in the global scope. And furthermore, it ensures such function does not run in strict-mode, no matter the strict-mode status of the program. That's the same outcome as running an indirect

One niche usage of such strict-mode agnostic indirect function this assignment is for getting a reliable reference to the true global object prior to when the JS specification actually defined the globalThis identifier (for example, in a polyfill for it):

"use strict";

var gt = new Function("return this")();
gt === globalThis;                      // true

In fact, a similar outcome, using the comma operator trick (see previous section) and eval(..):

"use strict";

function getGlobalThis() {
    return (1,eval)("this");
}

getGlobalThis() === globalThis;      // true
NOTE:
eval("this") would be sensitive to strict-mode, but (1,eval)("this") is not, and therefor reliably gives us the globalThis in any program.

Unfortunately, the new Function(..) and (1,eval)(..) approaches both have an important limitation: that code will be blocked in browser-based JS code if the app is served with certain Content-Security-Policy (CSP) restrictions, disallowing dynamic code evaluation (for security reasons).

Can we get around this? Yes, mostly. 1

The JS specification says that a getter function defined on the global object, or on any object that inherits from it (like Object.prototype), runs the getter function with this context assigned to globalThis, regardless of the program's strict-mode.

// Adapted from: https://mathiasbynens.be/notes/globalthis#robust-polyfill
function getGlobalThis() {
    Object.defineProperty(Object.prototype,"__get_globalthis__",{
        get() { return this; },
        configurable: true
    });
    var gt = __get_globalthis__;
    delete Object.prototype.__get_globalthis__;
    return gt;
}

getGlobalThis() === globalThis;      // true

Yeah, that's super gnarly. But that's JS this for you!

Template Tag Functions

There's one more unusual variation of function invocation we should cover: tagged template functions.

Template strings -- what I prefer to call interpolated literals -- can be "tagged" with a prefix function, which is invoked with the parsed contents of the template literal:

function tagFn(/* .. */) {
    // ..
}

tagFn`actually a function invocation!`;

As you can see, there's no (..) invocation syntax, just the tag function (tagFn) appearing before the `template literal`; whitespace is allowed between them, but is very uncommon.

Despite the strange appearance, the function tagFn(..) will be invoked. It's passed the list of one or more string literals that were parsed from the template literal, along with any interpolated expression values that were encountered.

We're not going to cover all the ins and outs of tagged template functions -- they're seriously one of the most powerful and interesting features ever added to JS -- but since we're talking about this assignment in function invocations, for completeness sake we need to talk about how this will be assigned.

The other form for tag functions you may encounter is:

var someObj = {
    tagFn() { /* .. */ }
};

someObj.tagFn`also a function invocation!`;

Here's the easy explanation: tagFn`..` and someObj.tagFn`..` will each have this-assignment behavior corresponding to call-sites as tagFn(..) and someObj.tagFn(..), respectively. In other words, tagFn`..` behaves by the default context assignment rule (#4), and someObj.tagFn`..` behaves by the implicit context assignment rule (#3).

Luckily for us, we don't need to worry about the new or call(..) / apply(..) assignment rules, as those forms aren't possible with tag functions.

It should be pointed out that it's pretty rare for a tagged template literal function to be defined as this-aware, so it's fairly unlikely you'll need to apply these rules. But just in case, now you're in the know.

Stay Aware

So, that's this. I'm willing to bet for many of you, it was a bit more... shall we say, involved... than you might have been expecting.

The good news, perhaps, is that in practice you don't often trip over all these different complexities. But the more you use this, the more it requires you, and the readers of your code, to understand how it actually works.

The lesson here is that you should be intentional and aware of all aspects of this before you go sprinkling it about your code. Make sure you're using it most effectively and taking full advantage of this important pillar of JS.

Footnotes

  1. "A horrifying globalThis polyfill in universal JavaScript"; Mathias Bynens; April 18 2019; https://mathiasbynens.be/notes/globalthis#robust-polyfill ; Accessed July 2022 ↩