Raising the Dead with JavaScript

After my last post, Gavin sent me this: Scope Cheatsheet. Besides its misleading wording, an interesting fact stood out and gave me another JavaScript challenge question. I’ll also show you how it allows JavaScript to raise the dead!

Background

Like its close cousin, Scheme, JavaScript is a Lisp-1: functions and variables share the same namespace. In Scheme, the define form defines new variables.

(define foo "Hello, world!")

Combine with the lambda form and it can be used to name functions,

(define square (lambda (x) (* x x)))

(square -4)  ;; => 16

The variable square is assigned to an anonymous function, and afterward it can be called as a function. Since this is so common, there’s a syntactic shorthand (sugar) for this,

(define (square x) (* x x))

Notice that the first argument to define is now a list rather than a symbol. This is a signal to define that a function is being defined, and that this should be expanded into the lambda example above. (Also note that the declaration mimics a function call, which is pretty neat.)

JavaScript also has syntactic sugar for the same purpose. The var statement establishes a binding in the current scope. This can be used to define both variables and functions, since they share a namespace. For convenience, in addition to defining an anonymous function, the function statement can be used to declare a variable and assign it a function. These definitions below are equivalent … most of the time.

var square = function(x) {
    return x * x;
}

function square(x) {
    return x * x;
}

The second definition is actually more magical than a syntactic shorthand, which leads into my quiz.

Quiz

function bar() {
    var foo = 0;
    function foo() {}
    return typeof foo;
}

bar(); // What does this return? Why?

function baz() {
    var foo;
    function foo() {}
    return typeof foo;
}

baz(); // How about now?

function quux() {
    var foo = 0;
    var foo = function () {}
    return typeof foo;
}

quux(); // How about now?

We have three functions, bar(), baz(), and quux(), each slightly different. Try to figure out the return value of each without running them in a JavaScript interpreter. Reading the cheatsheet should give you a good idea of the answer.

Answer

Figured it out? The first function, bar(), is the surprising one. If the special function form was merely syntactic sugar then all this means is that foo is redundantly declared (and re-assigned before accessing it, which the compiler could optimize). The final assignment is a function, so it should return 'function'.

However, this is not the case! This function returns 'number'. The first assignment listed in the code actually happens after the second assignment, the function definition. This is because functions defined using the special syntax are hoisted to the top of the function. The function assignments are evaluated before any other part of the function body. This is the extra magic behind the special function syntax.

The effect is more apparent when looking at the return value of quux(), which is 'function'. The special function syntax isn’t used so the assignments are performed in the order that they’re listed. This isn’t surprising, except for the fact that variables can be declared multiple times in a scope without any sort of warning.

The second function, baz(), returns 'function'. The function definition is still hoisted but the variable declaration performs no assignment. The function assignment is not overridden. Because of the lack of assignment, nothing actually happens at all for the variable declaration.

Now, this seems to be a cloudy concept for even skilled programmers: a variable declaration like var foo = 0 accomplishes two separate things. The merge of these two tasks into a single statement is merely one of convenience.

  1. Declaration: declares a variable, modifying the semantics of the function’s body. It changes what place in memory an identifier in the current scope will refer to. This is a compile-time activity. Nothing happens at run time — there is no when. When function definitions are hoisted, it’s the assignment (part 2) that gets hoisted. In C, variables are initially assigned to stack garbage (globals are zeroed). In JavaScript, variables are initially assigned to undefined.

  2. Assignment: binds a variable to a new value. This is evaluated at run time. It matters when this happens in relation to other evaluations.

Consider this,

var foo = foo;

The expression on the right-hand side is evaluated in the same scope as the variable declaration. foo is initially assigned to undefined, then it is re-assigned to undefined. This permits recursive functions to be defined with var — otherwise the identifier used to make the recursive call wouldn’t refer to the function itself.

var factorial = function(n) {
    if (n === 0)
        return 1;
    else
        return factorial(n - 1) * n;
};

In contrast, Lisp’s let does not evaluate the right-hand side within the scope of the let, so recursive definitions are not possible with a regular let. This is the purpose of letrec (Scheme) and labels (Common Lisp).

;; Compile error, x is unbound
(let ((x x))
  x)

Why function hoisting?

JavaScript’s original goal was to be easy for novices to program. I think that they wanted users to be able to define functions anywhere in a function (at the top level) without thinking about it. Novices generally don’t think of functions as values, so this is probably more intuitive for them. To accomplish this, the assignment needs to happen before the real body of the function. Unfortunately, this leads to surprising behavior, and, ultimately, it was probably a bad design choice.

Below, in any other language the function definition would be dead code, unreachable by any valid control flow, and the compiler would be free to toss it.

function foo() {
    return baz();
    function baz() { return 'Hello'; }
}

foo(); // => 'Hello'

But in JavaScript you can raise the dead!

Have a comment on this article? Start a discussion in my public inbox by sending an email to ~skeeto/public-inbox@lists.sr.ht [mailing list etiquette] , or see existing discussions.

null program

Chris Wellons

wellons@nullprogram.com (PGP)
~skeeto/public-inbox@lists.sr.ht (view)