javascript and memoized getters

javascript and memoized getters - javascript

This article describe getters. It has a section "
Smart / self-overwriting / lazy getters"
And it's unclear for me, are getters 'memoized' by default or should I implement this feature by myself
e.g.
class Foo() {
get boo() {
this._boo = this._boo || new Boo();
return this._boo;
}
}
or can I just write:
class Foo() {
get boo() {
return new Boo();
}
}
to have the same result?

The most interesting bit of that article was Smart / self-overwriting / lazy getters, which offers this technique:
class Foo {
get boo() {
delete this.boo;
return this.boo = new Boo();
}
}
With this your Foo objects don't go through the hassle of creating their boo properties until you ask for it. Then it's created once and further requests for it simply return the same object. This makes sense if new Boo() is in someway resource-intensive to create and reasonably often is not needed.
Theoretically, you could extend this to allow you to delete the current version and recreate it on next access. But that's a lot more code, and is probably a fairly rare need.
Update
A comment from vrugtehagel correctly pointed out that the above technique, while fine for plain objects, does not work for classes.
Here's a variant which does work:
class Boo {
static counter = 0
constructor () {
this.x = ++Boo.counter
console .log (`creating Boo(${this.x})`)
}
}
class Foo {
get boo () {
Object .defineProperty (
this,
"boo",
{ value: new Boo(), writable: false}
)
return this .boo;
}
}
const f = new Foo()
console .log (f.boo)
console .log (f.boo) // no 'creating Boo' log, Boo constructor only created once

No, there is no language-level support for memoized getters in JavaScript. In your second example, a new object would be created every time boo was accessed.

You're free to add memoization, e.g.
Non memoized,
class NonMemoized {
constructor(prefix) {
this.prefix = prefix;
}
get myFunc() {
return this.prefix + Math.random().toString();
}
}
let nonMemoized = new NonMemoized('new number each time ');
console.log(nonMemoized.myFunc);
console.log(nonMemoized.myFunc);
Memoized, nice for when u want to create an object once and always return the same object (but don't want to create in the constructor because maybe it's not necessary all the time or some other reason)
class MemoizedManually {
constructor(prefix) {
this.prefix = prefix;
}
get myFunc() {
return this._myFunc_ = this._myFunc_ || this.prefix + Math.random().toString();
}
}
let memoizedManually = new MemoizedManually('same number ');
console.log(memoizedManually.myFunc);
console.log(memoizedManually.myFunc);
Lastly, if you have a bunch of functions you want to memoize but don't want to repeat that this.x = this.x || something computation in each function (which you really shoudln't repeat, as it's not really the job of myFunc to memoize itself:
class Memoized {
constructor(prefix) {
this.prefix = prefix;
}
get myFunc() {
return this.prefix + Math.random().toString();
}
}
const memoizeGetter = (clazz, functionName) => {
let func = Object.getOwnPropertyDescriptor(clazz.prototype, functionName);
let cacheKey = `_${functionName}-cache_`;
Object.defineProperty(clazz.prototype, functionName, {
get: function () {
return this[cacheKey] = this[cacheKey] || func.get.call(this);
}
});
};
memoizeGetter(Memoized, 'myFunc');
let memoized = new Memoized('also same number ');
console.log(memoized.myFunc);
console.log(memoized.myFunc);
Nice thing about getters is they don't take arguments, so you don't have to worry about ...args, but do need to worry about binding this

consider this code:
class Person {
static get SHORT() { return 0; }//rvalue
}
versus
class Person {}
Person.SHORT = 0;//lvalue
Although both return the same result, the latter is actually faster (because it avoids the function call overhead); though the js engine can make optimizations that nullify one over the other.

You can do it all on one line:
class Foo() {
get boo() {
return this._boo = this._boo || new Boo();
}
}
This is easy to remember and isn't convoluted. Got to consider the maintenance factor, and try to keep things simple.

Related

Is it possible to instantiate a class without a `new` operator? [duplicate]

Given a simple class
class Foo {
constructor(x) {
if (!(this instanceof Foo)) return new Foo(x);
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
Is it possible to call the class constructor without the new keyword?
Usage should allow
(new Foo("world")).hello(); // "hello world"
Or
Foo("world").hello(); // "hello world"
But the latter fails with
Cannot call a class as a function

Classes have a "class body" that is a constructor.
If you use an internal constructor() function, that function would be the same class body as well, and would be what is called when the class is called, hence a class is always a constructor.
Constructors require the use of the new operator to create a new instance, as such invoking a class without the new operator results in an error, as it's required for the class constructor to create a new instance.
The error message is also quite specific, and correct
TypeError: Class constructors cannot be invoked without 'new'
You could:
either use a regular function instead of a class1.
Always call the class with new.
Call the class inside a wrapping regular function, always using new, that way you get the benefits of classes, but the wrapping function can still be called with and without the new operator2.
1)
function Foo(x) {
if (!(this instanceof Foo)) return new Foo(x);
this.x = x;
this.hello = function() {
return this.x;
}
}
2)
class Foo {
constructor(x) {
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
var _old = Foo;
Foo = function(...args) { return new _old(...args) };

As others have pointed out, ES2015 spec strictly states that such call should throw TypeError, but at the same time, it provides feature that can be used to achieve exactly the desired result, namely Proxies.
Proxies allows us to virtualize over a concept of an object. For instance, they can be used to change some behaviour of particular object without affecting anything else.
In your specific use case, class Foo is Function object which can be called -- this normally means that body of this function will be executed. But this can be changed with Proxy:
const _Foo = new Proxy(Foo, {
// target = Foo
apply (target, thisArg, argumentsList) {
return new target(...argumentsList);
}
});
_Foo("world").hello();
const f = _Foo("world");
f instanceof Foo; // true
f instanceof _Foo; // true
(Note that _Foo is now the class you want to expose, so identifiers should probably be the other way round)
If run by browsers that support Proxies, calling _Foo(...) will now execute apply trap function instead of the original constructor.
At the same time, this "new" _Foo class is indistinguishable from original Foo (apart from being able to call it as a normal function). Similarly, there is no difference by which you can tell object created with Foo and _Foo.
The biggest downside of this is that it cannot be transpiled or polyfilled, but still it's viable solution for having Scala-like class applied in JS in the future.

Here's a pattern I've come across that really helps me. It doesn't use a class, but it doesn't require the use of new either. Win/Win.
const Foo = x => ({
x,
hello: () => `hello ${x}`,
increment: () => Foo(x + 1),
add: ({x: y}) => Foo(x + y)
})
console.log(Foo(1).x) // 1
console.log(Foo(1).hello()) // hello 1
console.log(Foo(1).increment().hello()) // hello 2
console.log(Foo(1).add(Foo(2)).hello()) // hello 3

i just made this npm module for you ;)
https://www.npmjs.com/package/classy-decorator
import classy from "classy-decorator";
#classy()
class IamClassy {
constructor() {
console.log("IamClassy Instance!");
}
}
console.log(new IamClassy() instanceof IamClassy()); // true
console.log(IamClassy() instanceof IamClassy()); // true

No, this is not possible. Constructors that are created using the class keyword can only be constructed with new, if they are [[call]]ed without they always throw a TypeError1 (and there's not even a way to detect this from the outside).
1: I'm not sure whether transpilers get this right
You can use a normal function as a workaround, though:
class Foo {
constructor(x) {
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
{
const _Foo = Foo;
Foo = function(...args) {
return new _Foo(...args);
};
Foo.prototype = _Foo.prototype;
}
Disclaimer: instanceof and extending Foo.prototype work as normal, Foo.length does not, .constructor and static methods do not but can be fixed by adding Foo.prototype.constructor = Foo; and Object.setPrototypeOf(Foo, _Foo) if required.
For subclassing Foo (not _Foo) with class Bar extends Foo …, you should use return Reflect.construct(_Foo, args, new.target) instead of the new _Foo call. Subclassing in ES5 style (with Foo.call(this, …)) is not possible.

class MyClass {
constructor(param) {
// ...
}
static create(param) {
return new MyClass(param);
}
doSomething() {
// ...
}
}
MyClass.create('Hello World').doSomething();
Is that what you want?
If you need some logic when creating a new instance of MyClass, it could be helpful to implement a "CreationStrategy", to outsorce the logic (for example complex builder logic with validation)
Edit: As discussed in the comments It does not make sense to create some sort of Builder Pattern with a separate class in JavaScript. Removed related example.

Here's a where you can use a 'scope safe constructor'
Observe this code:
function Student(name) {
if(this instanceof Student) {
this.name = name;
} else {
return new Student(name);
}
}
Now you can create a Student object without using new as follows:
var stud1 = Student('Kia');

Dug up this one in the draft
Constructors defined using class definition syntax throw when called as functions
So I guess that's not possible with classes.

Call class constructor manually can be usefull when refactoring code (having parts of the code in ES6, other parts beeing function & prototype definition)
I ended up with a small, yet usefull boilerplate, slicing the constructor into another function. Period.
class Foo {
constructor() {
//as i will not be able to call the constructor, just move everything to initialize
this.initialize.apply(this, arguments)
}
initialize() {
this.stuff = {};
//whatever you want
}
}
function Bar () {
Foo.prototype.initialize.call(this);
}
Bar.prototype.stuff = function() {}

I had problems extending classes converted with the transformation function mentioned in some other answers. The issue seems to be that node (as of v9.4.0) doesn't properly support the argument spread operator ((...args) =>).
This function based on the transpiled output of the classy-decorator (mentioned in another answer) works for me and doesn't require support for decorators or the argument spread operator.
// function that calls `new` for you on class constructors, simply call
// YourClass = bindNew(YourClass)
function bindNew(Class) {
function _Class() {
for (
var len = arguments.length, rest = Array(len), key = 0;
key < len;
key++
) {
rest[key] = arguments[key];
}
return new (Function.prototype.bind.apply(Class, [null].concat(rest)))();
}
_Class.prototype = Class.prototype;
return _Class;
}
Usage:
class X {}
X = bindNew(X);
// or
const Y = bindNew(class Y {});
const x = new X();
const x2 = X(); // woohoo
x instanceof X; // true
x2 instanceof X; // true
class Z extends X {} // works too
As a bonus, TypeScript (with "es5" output) seems to be fine with the old instanceof trick (well, it won't typecheck if used without new but it works anyhow):
class X {
constructor() {
if (!(this instanceof X)) {
return new X();
}
}
}
because it compiles it down to:
var X = /** #class */ (function () {
function X() {
if (!(this instanceof X)) {
return new X();
}
}
return X;
}());

Alright I have another answer here, and I think this one is pretty innovative.
Basically, the problem with doing something similar to Naomik's answer is that you create functions each and every time you chain methods together.
EDIT: This solution shares the same problem, however, this answer is being left up for educational purposes.
So here I'm offering a way to merely bind new values to your methods--which are basically just independent functions. This offer the additional benefit of being able to import functions from different modules into the newly constructed object.
Okay, so here it goes.
const assoc = (prop, value, obj) =>
Object.assign({},obj,{[prop]: value})
const reducer = ( $values, accumulate, [key,val] ) => assoc( key, val.bind( undefined,...$values ), accumulate )
const bindValuesToMethods = ( $methods, ...$values ) =>
Object.entries( $methods ).reduce( reducer.bind( undefined, ...$values), {} )
const prepareInstance = (instanceMethods, staticMethods = ({}) ) => Object.assign(
bindValuesToMethods.bind( undefined, instanceMethods ),
staticMethods
)
// Let's make our class-like function
const RightInstanceMethods = ({
chain: (x,f) => f(x),
map: (x,f) => Right(f(x)),
fold: (x,l,r) => r(x),
inspect: (x) => `Right(${x})`
})
const RightStaticMethods = ({
of: x => Right(x)
})
const Right = prepareInstance(RightInstanceMethods,RightStaticMethods)
Now you can do
Right(4)
.map(x=>x+1)
.map(x=>x*2)
.inspect()
You can also do
Right.of(4)
.map(x=>x+1)
.map(x=>x*2)
.inspect()
You also have the added benefit of being able to export from modules as such
export const Right = prepareInstance(RightInstanceMethods,RightStaticMethods)
While you don't get ClassInstance.constructor you do have FunctorInstance.name (note, you may need to polyfill Function.name and/or not use an arrow function for export for browser compatibility with Function.name purposes)
export function Right(...args){
return prepareInstance(RightInstanceMethods,RightStaticMethods)(...args)
}
PS - New name suggestions for prepareInstance welcomed, see Gist.
https://gist.github.com/babakness/56da19ba85e0eaa43ae5577bc0064456

As pointed out by you and others
Foo("world").hello();
fails with an error because it is an error,
according to rules of ES6 syntax.
Others pointed out that
(new Foo("world")).hello();
works but is clunky because
It needs the 'new' AND
It needs the extra parenthesis.
I agree it is clunky. So I'm often using
this solution instead:
In your class Foo, create a static method
named 'new':
static new (...args)
{ return new this (...args);
}
Use it like this:
Foo.new("world").hello();
This way I hide the "clunkiness" inside
this static method 'new()'.
Note that this method new() is generic,
it will work as is also
when inherited to sub-classes. If you need
to customize it in a subclass you can first call:
super.new(...args)
and then add any other stuff you need in the
method in a subclass, before returning its result.

A recapped working "one-line" solution for ES6: explained
The answer posted above by Bergi is basically correct.
TLDR; skip to the end 😎 for the one-liner solution
Bergi's answer may seem a unclear when reading it. So, here is a more expanded code-sample that illustrates TWO new ES6 features to achieve the desired goals.
Together, they let a single function C (below) provide the dual-role of a factory and new-able fn; which constructs a B inst that derives from a A.
The B constructor utilizes super handling to invoke the A constructor with initialization arguments. In our final #3 - #4 examples constructed by C.
The A constructor demonstrates the semantics of the new.target psuedo-var to discover new was actually invoked with B.
First, we will make use of ES6 new.target psuedo-var that gives us the RHS of a new RHS() expression.
Technically, we could have gotten new.target as this?.__proto__?.constructor; they are equivalent.
Second, we will make use of ES6 Reflect.construct. Which is crucial to working around the ES6 class constructor invocation constraints; if we are bound and determined to not use new RHS(...).
Test the following and see for yourself its output (also provided in #1-4 below).
class A {
constructor(...a) {
const descendentType = new.target;
console.log(`A's constructor seeing 'new' invoked on ${descendentType?.name} with args: %o`,a);
}
}
class B extends A {
constructor(...a) {
super(...a);
}
}
// C is our DUAL mode Factory
function C(...a) {
console.log(`C's new.target => ${new.target?.name}`);
const inst = new.target ? Reflect.construct(B, a) : new B(...a);
console.log(`C has constructed a ${inst.__proto__.constructor.name} inst`);
return inst;
}
Which we can then invoke it in the following ways:
new A('NEW-A()')
output => "A's constructor seeing 'new' invoked on A with args: ['NEW-A()']"
new B('NEW-B()')
output => "A's constructor seeing 'new' invoked on B with args: ['NEW-B()']"
new C('NEW-C()')
output => "C's new.target => C"
output => "A's constructor seeing 'new' invoked on B with args: ['NEW-C()']"
output => "C has constructed a B inst"
C('PLAIN-C()')
output => "C's new.target => undefined"
output => "A's constructor seeing 'new' invoked on B with args: ['PLAIN-C()']"
output => "C has constructed a B inst"
Where #3 and #4 achieve the originally desired goals.
The simplified `C` looks like:
function C(...a) {return Reflect.construct(B, a);}
OR - if 3rd arg of Reflect.construct not utilized for init.
function C(...a) {return new B(...a);}
Beware: C must be a function not a class for this to both be allowed, and to work returning an alternate this on a new C() invocation, etc.
Also to circumvent strict mode rules for arguments.callee requires using a closure (smalltalk-block. Illustrated below:
class B extends A {
// embedding within a class and generically referencing it requires =>
static C = (() => {
const $class = this; return function(...a) {
return Reflect.construct($class, a);}})();
// Read more on `Reflect.construct` 3rd argument to see more capabilities
// for why it does MORE than just `new $class(...a)` would do.
}
exports.C = B.C;
⛐⚠️⛐ You could do some awful things like fiddle the __proto__ on the resulting inst and change out its constructor and name. Which would make it look and feel like a real subclass C of B depending on how far you want to go to manipulate the object-model. The subtleties abound in what happens with getters/setters, super and # privates. But for much of that you can STAY ES6 CLEAN and get clever with using extends and providing a template superclass flattened mixin tree; which I do a lot of in efekt for supporting tiny-but-complete µhtml reactive custom-elements parts and related PWA app models and responsive dynamic just-in-time versioned code bundling from EdgeS ESS backend servers. As in ... const M = $class => class extends $class {...}.
My motivations...
I posted this to help explain the semantics and a working ES6 solution, which is what I use to support subclassing Promise to provide FutureValue with better workflow handling capabilities in my github efekt library (EdgeS Front End Kit library).

In 2022, with ES6 onwards you can do it with the static method that can be called before the instance of the class is created, to create a instance of the class.
So the code should look something like this:
class Foo {
constructor(x) {
this.x = x;
}
//static class
static Init(x) {
return new Foo(x)
}
sayHello() {
return `hello ${this.x}`;
}
}
//so if i call
Foo.Init('world').sayHello();
//it prints: hello world
But if you are doing all this to make a chain of method you can also look at the following construct:
function MyName(name) {
if (this instanceof MyName) {
this.name = name,
this.prepend = function(n) {
this.name = `${n} ${this.name}`;
return this;
}
,
this.append = function(n) {
this.name = `${this.name} ${n} `;
return this;
}
,
this.show = function() {
return this.name;
}
} else {
return new MyName(name);
}
}
//Call
MyName('vinod').prepend('dev').append('hacks').show();
//prints: dev vinod hacks
The method above returns this at the end of each method which makes the object, properties & method avaialble.
The good part is these methods can be used again & again to create a sentence as
MyName('vinod').prepend('dev').append('hacks')
.prepend("the").append('javascript').append('for Stackoverflow').show();
I have used it as a stringBuilder or to generate xml dynamically.

Calling the class constructor without the new keyword is not possible.
The error message is quite specific.
See a blog post on 2ality and the spec:
However, you can only invoke a class via new, not via a function call (Sect. 9.2.2 in the spec):
> Point()
TypeError: Classes can’t be function-called

I'm adding this as a follow up to a comment by naomik and utilizing on the method illustrated by Tim and Bergi. I'm also going to suggest an of function to use as a general case.
To do this in a functional way AND utilize the efficiency of prototypes (not re-create all method each time a new instance is created), one could use this pattern
const Foo = function(x){ this._value = x ... }
Foo.of = function(x){ return new Foo(x) }
Foo.prototype = {
increment(){ return Foo.of(this._value + 1) },
...
}
Please note that this is consistent with fantasy-land JS specs
https://github.com/fantasyland/fantasy-land#of-method
I personally feel that it is cleaner to use the ES6 class syntax
class Foo {
static of(x) { new Foo(x)}
constructor(x) { this._value = x }
increment() { Foo.of(this._value+1) }
}
Now one could wrap this in a closure as such
class Foo {
static of(x) { new _Foo(x)}
constructor(x) { this._value = x }
increment() { Foo.of(this._value+1) }
}
function FooOf (x) {
return Foo.of(x)
}
Or rename FooOf and Foo as desired, ie the class could be FooClass and the function just Foo, etc.
This is better than place the class in the function because creating new instances doesn't burden us with creating new classes as well.
Yet another way is to create a an of function
const of = (classObj,...args) => (
classObj.of
? classObj.of(value)
: new classObj(args)
)
And then do something like of(Foo,5).increment()

Still finding interesting ways to use instanceof without relying on new or class keywords. In this example program, we compute the 100,000th fibonacci number in less than one second. The result is over 20,000 digits long -
const fib = x =>
Loop // <- no `new`
( (n, a, b) =>
n <= 0n
? String(a) // <- no `new`
: Recur(n - 1n, b, a + b) // <- no `new`
, BigInt(x) // <- no `new`
, 0n
, 1n
)
function Loop (f, ...init)
{ let r = f(...init)
while (r instanceof Recur) // <- instanceof works
r = f(...r)
return r
}
function Recur (...v)
{ return Object.create // <- not a class, but works
( Recur.prototype // <- set prototype
, { constructor: { value: Recur } // <- set constructor
, [Symbol.iterator]: { value: _ => v.values() } // <- whatever you want
}
)
}
document.body.textContent = fib(100000)
body { overflow-wrap: anywhere; }
I don't know why I haven't thought of this before -
function atom (T, v)
{ return Object.assign
( Object.create
( T.prototype
, { constructor: { value: T } }
)
, v
)
}
function pair (car, cdr)
{ return atom(pair, { car, cdr }) }
const p =
pair(1, 2)
console.log(p)
console.log(p instanceof pair)
Output -
{
"car": 1,
"cdr": 2
}
true

I wrote a small helper function which solves this problem. It effectively converts an ES6 class into an older ES5 constructor function which isn't subject to the same ruleset. This way you can create constructors which don't need new. You can also overload constructors in a similar way to the builtin Number, String etc.
function callableConstructor(c, f) {
function ret(...args) {
if(new.target) {
return new c(...args)
}
return f(...args)
}
ret.prototype = c.prototype
ret.prototype.constructor = ret
return ret
}
Test it below:
function callableConstructor(c, f) {
function ret(...args) {
if(new.target) {
return new c(...args)
}
return f(...args)
}
ret.prototype = c.prototype
ret.prototype.constructor = ret
return ret
}
// Usage
class Foo {
constructor(a, b) {
this.a = a
this.b = 2 * b
}
f() {
return this.a + this.b
}
}
Foo = callableConstructor(Foo, (...args) => new Foo(...args))
let foo = new Foo(2, 3)
console.log(foo) // Foo { a: 2, b: 6 }
console.log(foo.f()) // 8
console.log(foo instanceof Foo) // true
foo = Foo(2, 3)
console.log(foo) // Foo { a: 2, b: 6 }
console.log(foo.f()) // 8
console.log(foo instanceof Foo) // true

I came at this issue because I encountered the no-new "do not use new for side effects" eslint rule - which turns out it's a bad practice to use new for an object that is immediately discarded.
I still wanted to use the class syntax because I like it, but I agree that a regular class with new keyword for something that does not produce an object can be confusing.
The solution for me was simple. Define an unexported class in a module and export a function that instatinates it.
class SideEffects {
constructor() {
}
// ...
}
export function addSideEffects() {
// eslint-disable-next-line no-new
new SideEffects();
}
Yes, we are still using the new keyword, but it's used internally in the module and it's obvious from reading the module file that it's not a regular class - and the exported function also makes it clear that it does not create an object.

This might be a little contrived, but it works
function Foo(x){
"use strict"
class Bar {
constructor(x) {
if (!(this instanceof Bar)) return new Bar(x);
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
return new Bar(x)
}
Foo("world").hello()

You can't use a class without the new constructor, in my case I didn't want to use the new constructor any time I wanted to use my class, so what you can do is to wrap your class as follows (in my case it's a Dates utils library):
const defaultOptions = {
defaultFormatOptions: 'dd/MM/yyyy'
}
class DatesClass {
constructor(date = new Date(), options) {
this.date = date
this.options = { ...defaultOptions, ...options }
}
get value() {
return this.date
}
add() {}
...
}
export default (date, options) => new DateClass(date, options)
// then you can use it as follow
import dates from 'path/to/yourClass/from/above'
dates(new Date()).add({ unit: 'day', qty: 2}).value

Unable to have get and set method in Object Constructor (not in class) in JavaScript? [duplicate]

I recently read about the fact that there is a possibility of defining getters/setters in JavaScript. It seems extremely helpful - the setter is a kind of 'helper' which can parse the value to be set first, before actually setting it.
For example, I currently have this code:
var obj = function(value) {
var test = !!value; // 'test' has to be a boolean
return {
get test() { return test },
set test(value) { test = !!value }
};
};
var instance = new obj(true);
This code always converts value to a boolean. So if you code instance.test = 0, then instance.test === false.
However, for this to work you have to actually return an object, which means that the new instance is not of type obj but just is a plain object. This means that changing the prototype of obj has no effect on instances. For example, this does not work - instance.func is undefined:
obj.prototype.func = function() { console.log(this.value); };
because instance is not of type obj. To get the prototype functions work, I guess I should not return a plain object, but rather not return anything so that instance would just be of type obj, like a regular constructor works.
The problem then is how to implement getters/setters? I can only find articles describing how to add these to an object, not as being part of the constructor of a custom type.
So how do I implement getters/setters in the constructor so as to be able to both use getters/setters and extending the prototype?

You can't do that.
You can set setter/getters for properties of objects though. I advice you use ES5 Object.defineProperties though. of course this only works in modern browsers.
var obj = function() {
...
Object.defineProperties(this, {
"test": {
"get": function() { ... },
"set": function() { ... }
}
});
}
obj.prototype.func = function() { ... }
var o = new obj;
o.test;
o.func();

Usually you want class methods. The answer by #Raynos on May 7, 2011 gets the job done, but it defines an instance method, not a class method.
The following illustrates a class definition with a the getter and setter being part of the class. This definition is a lot like the answer by #Raynos, but with two differences in the code: (1) The "defineProperties()" action has been moved out of the constructor. (2) The argument to "defineProperties()"as been changed from the instance object "this", to the constructor's prototype object.
function TheConstructor(side) {
this.side = side;
}
Object.defineProperties(TheConstructor.prototype, {
area: {
get: function() { return this.side * this.side; }
,set: function(val) { this.side = Math.sqrt(val); }
}
});
// Test code:
var anInstance = new TheConstructor(2);
console.log("initial Area:"+anInstance.area);
anInstance.area = 9;
console.log("modified Area:"+anInstance.area);
Which produces these results:
initial Area:4
modified Area:9
Although usually the distinction between class versus instance
definition is just a matter of style, there is a purpose to
good style, and there is a case where the distinction matters:
the memoized getter. The purpose for a memoized getter is
described here: Smart/self-overwriting/lazy getters
Define the getter at the class level when the memoized value is to
pertain to the entire class. For example, a configuration file
should be read only once; the resulting values should then apply
for the duration of the program. The following sample code
defines a memoized getter at the class level.
function configureMe() {
return 42;
}
Object.defineProperties(TheConstructor.prototype, {
memoizedConfigParam: {
get: function() {
delete TheConstructor.prototype.memoizedConfigParam;
return TheConstructor.prototype.memoizedConfigParam = configureMe();
}
,configurable: true
}
});
// Test code:
console.log("memoizedConfigParam:"+anInstance.memoizedConfigParam);
Produces:
memoizedConfigParam:42
As can be seen in the example, memoized getters have the
characteristic that the getter function deletes itself,
then replaces itself with a simple value that
(presumably) will never change.
Note that 'configurable' must be set to 'true'.
Define the getter at the instance level when the memoized value
depends upon the contents of instance. The definition moves
inside the constructor, and the object of attention is 'this'.
function TheConstructorI(side) {
this.side = side;
Object.defineProperties(this, {
memoizedCalculation: {
get: function() {
delete this.memoizedCalculation;
return this.memoizedCalculation = this.expensiveOperation();
}
,configurable: true
}
});
}
TheConstructorI.prototype.expensiveOperation = function() {
return this.side * this.side * this.side;
}
//Test code:
var instance2 = new TheConstructorI(2);
var instance3 = new TheConstructorI(3);
console.log("memoizedCalculation 2:"+instance2.memoizedCalculation);
console.log("memoizedCalculation 3:"+instance3.memoizedCalculation);
Produces:
memoizedCalculation 2:8
memoizedCalculation 3:27
If you want to guarantee (rather than presume) that the memoized
value will never be changed, the 'writable' attribute needs to
be changed. That makes the code a bit more complicated.
function TheConstructorJ(side) {
this.side = side;
Object.defineProperties(this, {
memoizedCalculation: {
get: function() {
delete this.memoizedCalculation;
Object.defineProperty( this, 'memoizedCalculation'
,{ value : this.expensiveOperation()
,writable : false
});
return this.memoizedCalculation;
}
,configurable: true
}
});
}
TheConstructorJ.prototype.expensiveOperation = function() {
return this.side * this.side * this.side;
}
//Test code:
var instanceJ = new TheConstructorJ(2);
console.log("memoizedCalculation:"+instanceJ.memoizedCalculation);
instanceJ.memoizedCalculation = 42; // results in error
Produces:
memoizedCalculation:8
>Uncaught TypeError: Cannot assign to read only property 'memoizedCalculation' of object '#<TheConstructorJ>'
The OP's original question, from March 7, 2011, presented basic
getter and setter syntax, noted that it worked on an object but
not on 'this', and asked how to define getters and setters within
a constructor. In addition to all the examples above, there is
also a "cheap-shot" way of doing it: create a new object within
the constructor, like the OP did, but then assign the object to
be a member within 'this'. So, the original code would look like
this:
var MyClass = function(value) {
var test = !!value; // 'test' has to be a boolean
this.data = {
get test() { return test },
set test(value) { test = !!value }
};
};
var instance = new MyClass(true);
// But now 'data' is part of the access path
instance.data.test = 0;
console.log(instance.data.test);
Produces:
false
Believe it or not, I have actually run into situations where
this "cheap-shot" is the best solution. Specifically, I used this
technique when I had records from several tables encapsulated within
a single class, and wanted to present a unified view as though
they were a single record called 'data'.
Have fun.
IAM_AL_X

Update for ES6 -- have a look at section 19.3.1 of Alex Rauschmayer's book Exploring ES6 http://exploringjs.com/es6/ch_maps-sets.html#sec_weakmaps-private-data which demonstrates how to use WeakMaps with getters and setters to hold private data. Combining with section 16.2.2.3 http://exploringjs.com/es6/ch_classes.html#leanpub-auto-getters-and-setters would result in something like
# module test_WeakMap_getter.js
var _MyClassProp = new WeakMap();
class MyClass {
get prop() {
return _MyClassProp.get( this );
}
set prop(value) {
_MyClassProp.set( this, value );
}
}
var mc = new MyClass();
mc.prop = 5 ;
console.log( 'My value is', mc.prop );
$ node --use_strict test_WeakMap_getter.js
My value is 5

function Obj(value){
this.value = !!value;
}
Obj.prototype = {
get test () {
return this.value;``
},
set test (value) {
this.value = !!this.value;
}
};
var obj = new Obj(true);

I know this might be extremely late but I figured out a different way to accomplish what you want and for the sake of people, like myself, googling for an answer to this here it is.
function Constructor(input){
this.input = input;
}
Object.__defineGetter__.call(Constructor.prototype, "value", function(){
return this.input * 2;
});
var test = new Constructor(5);
alert(test.value) // 10
I've tested this in chrome, safari, mobile safari, firefox and they all work (latest versions of course)

#Alex I see it as more option and more power, programming is art, #Nat share his finding with us, and for that I thank him. Maybe someone want to do it that way.
I'm sure the setter version is the same but just changing that g to a s.
i.g:
function Constructor(input){
this.input = input;
}
Object.__defineGetter__.call(Constructor.prototype, "value", function(){
return this.input * 2;
});
Object.__defineSetter__.call(Constructor.prototype, "bar", function(foo){
return this.input *= foo;
});
var test = new Constructor(5);
console.log(test.value); // 10
test.bar = 5;
console.log(test.input); //25
With that said, this feature is deprecated, advices to not to use in production coding.

Make JS class callable without side effects [duplicate]

Given a simple class
class Foo {
constructor(x) {
if (!(this instanceof Foo)) return new Foo(x);
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
Is it possible to call the class constructor without the new keyword?
Usage should allow
(new Foo("world")).hello(); // "hello world"
Or
Foo("world").hello(); // "hello world"
But the latter fails with
Cannot call a class as a function

Classes have a "class body" that is a constructor.
If you use an internal constructor() function, that function would be the same class body as well, and would be what is called when the class is called, hence a class is always a constructor.
Constructors require the use of the new operator to create a new instance, as such invoking a class without the new operator results in an error, as it's required for the class constructor to create a new instance.
The error message is also quite specific, and correct
TypeError: Class constructors cannot be invoked without 'new'
You could:
either use a regular function instead of a class1.
Always call the class with new.
Call the class inside a wrapping regular function, always using new, that way you get the benefits of classes, but the wrapping function can still be called with and without the new operator2.
1)
function Foo(x) {
if (!(this instanceof Foo)) return new Foo(x);
this.x = x;
this.hello = function() {
return this.x;
}
}
2)
class Foo {
constructor(x) {
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
var _old = Foo;
Foo = function(...args) { return new _old(...args) };

As others have pointed out, ES2015 spec strictly states that such call should throw TypeError, but at the same time, it provides feature that can be used to achieve exactly the desired result, namely Proxies.
Proxies allows us to virtualize over a concept of an object. For instance, they can be used to change some behaviour of particular object without affecting anything else.
In your specific use case, class Foo is Function object which can be called -- this normally means that body of this function will be executed. But this can be changed with Proxy:
const _Foo = new Proxy(Foo, {
// target = Foo
apply (target, thisArg, argumentsList) {
return new target(...argumentsList);
}
});
_Foo("world").hello();
const f = _Foo("world");
f instanceof Foo; // true
f instanceof _Foo; // true
(Note that _Foo is now the class you want to expose, so identifiers should probably be the other way round)
If run by browsers that support Proxies, calling _Foo(...) will now execute apply trap function instead of the original constructor.
At the same time, this "new" _Foo class is indistinguishable from original Foo (apart from being able to call it as a normal function). Similarly, there is no difference by which you can tell object created with Foo and _Foo.
The biggest downside of this is that it cannot be transpiled or polyfilled, but still it's viable solution for having Scala-like class applied in JS in the future.

Here's a pattern I've come across that really helps me. It doesn't use a class, but it doesn't require the use of new either. Win/Win.
const Foo = x => ({
x,
hello: () => `hello ${x}`,
increment: () => Foo(x + 1),
add: ({x: y}) => Foo(x + y)
})
console.log(Foo(1).x) // 1
console.log(Foo(1).hello()) // hello 1
console.log(Foo(1).increment().hello()) // hello 2
console.log(Foo(1).add(Foo(2)).hello()) // hello 3

i just made this npm module for you ;)
https://www.npmjs.com/package/classy-decorator
import classy from "classy-decorator";
#classy()
class IamClassy {
constructor() {
console.log("IamClassy Instance!");
}
}
console.log(new IamClassy() instanceof IamClassy()); // true
console.log(IamClassy() instanceof IamClassy()); // true

No, this is not possible. Constructors that are created using the class keyword can only be constructed with new, if they are [[call]]ed without they always throw a TypeError1 (and there's not even a way to detect this from the outside).
1: I'm not sure whether transpilers get this right
You can use a normal function as a workaround, though:
class Foo {
constructor(x) {
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
{
const _Foo = Foo;
Foo = function(...args) {
return new _Foo(...args);
};
Foo.prototype = _Foo.prototype;
}
Disclaimer: instanceof and extending Foo.prototype work as normal, Foo.length does not, .constructor and static methods do not but can be fixed by adding Foo.prototype.constructor = Foo; and Object.setPrototypeOf(Foo, _Foo) if required.
For subclassing Foo (not _Foo) with class Bar extends Foo …, you should use return Reflect.construct(_Foo, args, new.target) instead of the new _Foo call. Subclassing in ES5 style (with Foo.call(this, …)) is not possible.

class MyClass {
constructor(param) {
// ...
}
static create(param) {
return new MyClass(param);
}
doSomething() {
// ...
}
}
MyClass.create('Hello World').doSomething();
Is that what you want?
If you need some logic when creating a new instance of MyClass, it could be helpful to implement a "CreationStrategy", to outsorce the logic (for example complex builder logic with validation)
Edit: As discussed in the comments It does not make sense to create some sort of Builder Pattern with a separate class in JavaScript. Removed related example.

Here's a where you can use a 'scope safe constructor'
Observe this code:
function Student(name) {
if(this instanceof Student) {
this.name = name;
} else {
return new Student(name);
}
}
Now you can create a Student object without using new as follows:
var stud1 = Student('Kia');

Dug up this one in the draft
Constructors defined using class definition syntax throw when called as functions
So I guess that's not possible with classes.

Call class constructor manually can be usefull when refactoring code (having parts of the code in ES6, other parts beeing function & prototype definition)
I ended up with a small, yet usefull boilerplate, slicing the constructor into another function. Period.
class Foo {
constructor() {
//as i will not be able to call the constructor, just move everything to initialize
this.initialize.apply(this, arguments)
}
initialize() {
this.stuff = {};
//whatever you want
}
}
function Bar () {
Foo.prototype.initialize.call(this);
}
Bar.prototype.stuff = function() {}

I had problems extending classes converted with the transformation function mentioned in some other answers. The issue seems to be that node (as of v9.4.0) doesn't properly support the argument spread operator ((...args) =>).
This function based on the transpiled output of the classy-decorator (mentioned in another answer) works for me and doesn't require support for decorators or the argument spread operator.
// function that calls `new` for you on class constructors, simply call
// YourClass = bindNew(YourClass)
function bindNew(Class) {
function _Class() {
for (
var len = arguments.length, rest = Array(len), key = 0;
key < len;
key++
) {
rest[key] = arguments[key];
}
return new (Function.prototype.bind.apply(Class, [null].concat(rest)))();
}
_Class.prototype = Class.prototype;
return _Class;
}
Usage:
class X {}
X = bindNew(X);
// or
const Y = bindNew(class Y {});
const x = new X();
const x2 = X(); // woohoo
x instanceof X; // true
x2 instanceof X; // true
class Z extends X {} // works too
As a bonus, TypeScript (with "es5" output) seems to be fine with the old instanceof trick (well, it won't typecheck if used without new but it works anyhow):
class X {
constructor() {
if (!(this instanceof X)) {
return new X();
}
}
}
because it compiles it down to:
var X = /** #class */ (function () {
function X() {
if (!(this instanceof X)) {
return new X();
}
}
return X;
}());

Alright I have another answer here, and I think this one is pretty innovative.
Basically, the problem with doing something similar to Naomik's answer is that you create functions each and every time you chain methods together.
EDIT: This solution shares the same problem, however, this answer is being left up for educational purposes.
So here I'm offering a way to merely bind new values to your methods--which are basically just independent functions. This offer the additional benefit of being able to import functions from different modules into the newly constructed object.
Okay, so here it goes.
const assoc = (prop, value, obj) =>
Object.assign({},obj,{[prop]: value})
const reducer = ( $values, accumulate, [key,val] ) => assoc( key, val.bind( undefined,...$values ), accumulate )
const bindValuesToMethods = ( $methods, ...$values ) =>
Object.entries( $methods ).reduce( reducer.bind( undefined, ...$values), {} )
const prepareInstance = (instanceMethods, staticMethods = ({}) ) => Object.assign(
bindValuesToMethods.bind( undefined, instanceMethods ),
staticMethods
)
// Let's make our class-like function
const RightInstanceMethods = ({
chain: (x,f) => f(x),
map: (x,f) => Right(f(x)),
fold: (x,l,r) => r(x),
inspect: (x) => `Right(${x})`
})
const RightStaticMethods = ({
of: x => Right(x)
})
const Right = prepareInstance(RightInstanceMethods,RightStaticMethods)
Now you can do
Right(4)
.map(x=>x+1)
.map(x=>x*2)
.inspect()
You can also do
Right.of(4)
.map(x=>x+1)
.map(x=>x*2)
.inspect()
You also have the added benefit of being able to export from modules as such
export const Right = prepareInstance(RightInstanceMethods,RightStaticMethods)
While you don't get ClassInstance.constructor you do have FunctorInstance.name (note, you may need to polyfill Function.name and/or not use an arrow function for export for browser compatibility with Function.name purposes)
export function Right(...args){
return prepareInstance(RightInstanceMethods,RightStaticMethods)(...args)
}
PS - New name suggestions for prepareInstance welcomed, see Gist.
https://gist.github.com/babakness/56da19ba85e0eaa43ae5577bc0064456

As pointed out by you and others
Foo("world").hello();
fails with an error because it is an error,
according to rules of ES6 syntax.
Others pointed out that
(new Foo("world")).hello();
works but is clunky because
It needs the 'new' AND
It needs the extra parenthesis.
I agree it is clunky. So I'm often using
this solution instead:
In your class Foo, create a static method
named 'new':
static new (...args)
{ return new this (...args);
}
Use it like this:
Foo.new("world").hello();
This way I hide the "clunkiness" inside
this static method 'new()'.
Note that this method new() is generic,
it will work as is also
when inherited to sub-classes. If you need
to customize it in a subclass you can first call:
super.new(...args)
and then add any other stuff you need in the
method in a subclass, before returning its result.

A recapped working "one-line" solution for ES6: explained
The answer posted above by Bergi is basically correct.
TLDR; skip to the end 😎 for the one-liner solution
Bergi's answer may seem a unclear when reading it. So, here is a more expanded code-sample that illustrates TWO new ES6 features to achieve the desired goals.
Together, they let a single function C (below) provide the dual-role of a factory and new-able fn; which constructs a B inst that derives from a A.
The B constructor utilizes super handling to invoke the A constructor with initialization arguments. In our final #3 - #4 examples constructed by C.
The A constructor demonstrates the semantics of the new.target psuedo-var to discover new was actually invoked with B.
First, we will make use of ES6 new.target psuedo-var that gives us the RHS of a new RHS() expression.
Technically, we could have gotten new.target as this?.__proto__?.constructor; they are equivalent.
Second, we will make use of ES6 Reflect.construct. Which is crucial to working around the ES6 class constructor invocation constraints; if we are bound and determined to not use new RHS(...).
Test the following and see for yourself its output (also provided in #1-4 below).
class A {
constructor(...a) {
const descendentType = new.target;
console.log(`A's constructor seeing 'new' invoked on ${descendentType?.name} with args: %o`,a);
}
}
class B extends A {
constructor(...a) {
super(...a);
}
}
// C is our DUAL mode Factory
function C(...a) {
console.log(`C's new.target => ${new.target?.name}`);
const inst = new.target ? Reflect.construct(B, a) : new B(...a);
console.log(`C has constructed a ${inst.__proto__.constructor.name} inst`);
return inst;
}
Which we can then invoke it in the following ways:
new A('NEW-A()')
output => "A's constructor seeing 'new' invoked on A with args: ['NEW-A()']"
new B('NEW-B()')
output => "A's constructor seeing 'new' invoked on B with args: ['NEW-B()']"
new C('NEW-C()')
output => "C's new.target => C"
output => "A's constructor seeing 'new' invoked on B with args: ['NEW-C()']"
output => "C has constructed a B inst"
C('PLAIN-C()')
output => "C's new.target => undefined"
output => "A's constructor seeing 'new' invoked on B with args: ['PLAIN-C()']"
output => "C has constructed a B inst"
Where #3 and #4 achieve the originally desired goals.
The simplified `C` looks like:
function C(...a) {return Reflect.construct(B, a);}
OR - if 3rd arg of Reflect.construct not utilized for init.
function C(...a) {return new B(...a);}
Beware: C must be a function not a class for this to both be allowed, and to work returning an alternate this on a new C() invocation, etc.
Also to circumvent strict mode rules for arguments.callee requires using a closure (smalltalk-block. Illustrated below:
class B extends A {
// embedding within a class and generically referencing it requires =>
static C = (() => {
const $class = this; return function(...a) {
return Reflect.construct($class, a);}})();
// Read more on `Reflect.construct` 3rd argument to see more capabilities
// for why it does MORE than just `new $class(...a)` would do.
}
exports.C = B.C;
⛐⚠️⛐ You could do some awful things like fiddle the __proto__ on the resulting inst and change out its constructor and name. Which would make it look and feel like a real subclass C of B depending on how far you want to go to manipulate the object-model. The subtleties abound in what happens with getters/setters, super and # privates. But for much of that you can STAY ES6 CLEAN and get clever with using extends and providing a template superclass flattened mixin tree; which I do a lot of in efekt for supporting tiny-but-complete µhtml reactive custom-elements parts and related PWA app models and responsive dynamic just-in-time versioned code bundling from EdgeS ESS backend servers. As in ... const M = $class => class extends $class {...}.
My motivations...
I posted this to help explain the semantics and a working ES6 solution, which is what I use to support subclassing Promise to provide FutureValue with better workflow handling capabilities in my github efekt library (EdgeS Front End Kit library).

In 2022, with ES6 onwards you can do it with the static method that can be called before the instance of the class is created, to create a instance of the class.
So the code should look something like this:
class Foo {
constructor(x) {
this.x = x;
}
//static class
static Init(x) {
return new Foo(x)
}
sayHello() {
return `hello ${this.x}`;
}
}
//so if i call
Foo.Init('world').sayHello();
//it prints: hello world
But if you are doing all this to make a chain of method you can also look at the following construct:
function MyName(name) {
if (this instanceof MyName) {
this.name = name,
this.prepend = function(n) {
this.name = `${n} ${this.name}`;
return this;
}
,
this.append = function(n) {
this.name = `${this.name} ${n} `;
return this;
}
,
this.show = function() {
return this.name;
}
} else {
return new MyName(name);
}
}
//Call
MyName('vinod').prepend('dev').append('hacks').show();
//prints: dev vinod hacks
The method above returns this at the end of each method which makes the object, properties & method avaialble.
The good part is these methods can be used again & again to create a sentence as
MyName('vinod').prepend('dev').append('hacks')
.prepend("the").append('javascript').append('for Stackoverflow').show();
I have used it as a stringBuilder or to generate xml dynamically.

Calling the class constructor without the new keyword is not possible.
The error message is quite specific.
See a blog post on 2ality and the spec:
However, you can only invoke a class via new, not via a function call (Sect. 9.2.2 in the spec):
> Point()
TypeError: Classes can’t be function-called

I'm adding this as a follow up to a comment by naomik and utilizing on the method illustrated by Tim and Bergi. I'm also going to suggest an of function to use as a general case.
To do this in a functional way AND utilize the efficiency of prototypes (not re-create all method each time a new instance is created), one could use this pattern
const Foo = function(x){ this._value = x ... }
Foo.of = function(x){ return new Foo(x) }
Foo.prototype = {
increment(){ return Foo.of(this._value + 1) },
...
}
Please note that this is consistent with fantasy-land JS specs
https://github.com/fantasyland/fantasy-land#of-method
I personally feel that it is cleaner to use the ES6 class syntax
class Foo {
static of(x) { new Foo(x)}
constructor(x) { this._value = x }
increment() { Foo.of(this._value+1) }
}
Now one could wrap this in a closure as such
class Foo {
static of(x) { new _Foo(x)}
constructor(x) { this._value = x }
increment() { Foo.of(this._value+1) }
}
function FooOf (x) {
return Foo.of(x)
}
Or rename FooOf and Foo as desired, ie the class could be FooClass and the function just Foo, etc.
This is better than place the class in the function because creating new instances doesn't burden us with creating new classes as well.
Yet another way is to create a an of function
const of = (classObj,...args) => (
classObj.of
? classObj.of(value)
: new classObj(args)
)
And then do something like of(Foo,5).increment()

Still finding interesting ways to use instanceof without relying on new or class keywords. In this example program, we compute the 100,000th fibonacci number in less than one second. The result is over 20,000 digits long -
const fib = x =>
Loop // <- no `new`
( (n, a, b) =>
n <= 0n
? String(a) // <- no `new`
: Recur(n - 1n, b, a + b) // <- no `new`
, BigInt(x) // <- no `new`
, 0n
, 1n
)
function Loop (f, ...init)
{ let r = f(...init)
while (r instanceof Recur) // <- instanceof works
r = f(...r)
return r
}
function Recur (...v)
{ return Object.create // <- not a class, but works
( Recur.prototype // <- set prototype
, { constructor: { value: Recur } // <- set constructor
, [Symbol.iterator]: { value: _ => v.values() } // <- whatever you want
}
)
}
document.body.textContent = fib(100000)
body { overflow-wrap: anywhere; }
I don't know why I haven't thought of this before -
function atom (T, v)
{ return Object.assign
( Object.create
( T.prototype
, { constructor: { value: T } }
)
, v
)
}
function pair (car, cdr)
{ return atom(pair, { car, cdr }) }
const p =
pair(1, 2)
console.log(p)
console.log(p instanceof pair)
Output -
{
"car": 1,
"cdr": 2
}
true

I wrote a small helper function which solves this problem. It effectively converts an ES6 class into an older ES5 constructor function which isn't subject to the same ruleset. This way you can create constructors which don't need new. You can also overload constructors in a similar way to the builtin Number, String etc.
function callableConstructor(c, f) {
function ret(...args) {
if(new.target) {
return new c(...args)
}
return f(...args)
}
ret.prototype = c.prototype
ret.prototype.constructor = ret
return ret
}
Test it below:
function callableConstructor(c, f) {
function ret(...args) {
if(new.target) {
return new c(...args)
}
return f(...args)
}
ret.prototype = c.prototype
ret.prototype.constructor = ret
return ret
}
// Usage
class Foo {
constructor(a, b) {
this.a = a
this.b = 2 * b
}
f() {
return this.a + this.b
}
}
Foo = callableConstructor(Foo, (...args) => new Foo(...args))
let foo = new Foo(2, 3)
console.log(foo) // Foo { a: 2, b: 6 }
console.log(foo.f()) // 8
console.log(foo instanceof Foo) // true
foo = Foo(2, 3)
console.log(foo) // Foo { a: 2, b: 6 }
console.log(foo.f()) // 8
console.log(foo instanceof Foo) // true

I came at this issue because I encountered the no-new "do not use new for side effects" eslint rule - which turns out it's a bad practice to use new for an object that is immediately discarded.
I still wanted to use the class syntax because I like it, but I agree that a regular class with new keyword for something that does not produce an object can be confusing.
The solution for me was simple. Define an unexported class in a module and export a function that instatinates it.
class SideEffects {
constructor() {
}
// ...
}
export function addSideEffects() {
// eslint-disable-next-line no-new
new SideEffects();
}
Yes, we are still using the new keyword, but it's used internally in the module and it's obvious from reading the module file that it's not a regular class - and the exported function also makes it clear that it does not create an object.

This might be a little contrived, but it works
function Foo(x){
"use strict"
class Bar {
constructor(x) {
if (!(this instanceof Bar)) return new Bar(x);
this.x = x;
}
hello() {
return `hello ${this.x}`;
}
}
return new Bar(x)
}
Foo("world").hello()

You can't use a class without the new constructor, in my case I didn't want to use the new constructor any time I wanted to use my class, so what you can do is to wrap your class as follows (in my case it's a Dates utils library):
const defaultOptions = {
defaultFormatOptions: 'dd/MM/yyyy'
}
class DatesClass {
constructor(date = new Date(), options) {
this.date = date
this.options = { ...defaultOptions, ...options }
}
get value() {
return this.date
}
add() {}
...
}
export default (date, options) => new DateClass(date, options)
// then you can use it as follow
import dates from 'path/to/yourClass/from/above'
dates(new Date()).add({ unit: 'day', qty: 2}).value

How to define private constructors in javascript?

I have defined pure objects in JS which expose certain static methods which should be used to construct them instead of the constructor. How can I make a constructor for my class private in Javascript?
var Score = (function () {
// The private constructor
var Score = function (score, hasPassed) {
this.score = score;
this.hasPassed = hasPassed;
};
// The preferred smart constructor
Score.mkNewScore = function (score) {
return new Score(score, score >= 33);
};
return Score;
})();
Update: The solution should still allow me to test for x instanceof Score. Otherwise, the solution by #user2864740 of exposing only the static constructor works.

One can use a variable (initializing) inside a closure which can throw an error if the constructor was called directly instead of via a class method:
var Score = (function () {
var initializing = false;
var Score = function (score, hasPassed) {
if (!initializing) {
throw new Error('The constructor is private, please use mkNewScore.');
}
initializing = false;
this.score = score;
this.hasPassed = hasPassed;
};
Score.mkNewScore = function (score) {
intializing = true;
return new Score(score, score >= 33);
};
return Score;
})();

Is there a solution which will allow me to say x instanceof Score?
Yes. Conceptually, #user2864740 is right, but for instanceof to work we need to expose (return) a function instead of a plain object. If that function has the same .prototype as our internal, private constructor, the instanceof operator does what is expected:
var Score = (function () {
// the module API
function PublicScore() {
throw new Error('The constructor is private, please use Score.makeNewScore.');
}
// The private constructor
var Score = function (score, hasPassed) {
this.score = score;
this.hasPassed = hasPassed;
};
// Now use either
Score.prototype = PublicScore.prototype; // to make .constructor == PublicScore,
PublicScore.prototype = Score.prototype; // to leak the hidden constructor
PublicScore.prototype = Score.prototype = {…} // to inherit .constructor == Object, or
PublicScore.prototype = Score.prototype = {constructor:null,…} // for total confusion :-)
// The preferred smart constructor
PublicScore.mkNewScore = function (score) {
return new Score(score, score >= 33);
};
return PublicScore;
}());
> Score.mkNewScore(50) instanceof Score
true
> new Score
Error (…)

Simply don't expose the constructor function. The core issue with the original code is the "static method" is defined as a property of the constructor (which is used as a "class") as opposed a property of the module.
Consider:
return {
mkNewScore: Score.mkNewScore
// .. and other static/module functions
};
The constructor can still be accessed via .constructor, but .. meh. At this point, might as well just let a "clever user" have access.
return {
mkNewScore: function (score) {
var s = new Score(score, score >= 33);
/* Shadow [prototype]. Without sealing the object this can
be trivially thwarted with `del s.constructor` .. meh.
See Bergi's comment for an alternative. */
s.constructor = undefined;
return s;
}
};

In order to create a private constructor in JS, I like to create a private key that is only accessible in the class (function) file and provide a static factory function as the only allowed way to construct said class:
// in PrivateConstructorClass.js
// Use a Symbol as this will always be unique.
// If you don't have Symbol in your runtime,
// use a random string that nobody can reliably guess,
// such as the current time plus some other random values.
const PRIVATE_CONSTRUCTOR_KEY = Symbol()
class PrivateConstructorClass {
constructor(arg1, arg2, argN, constructorKey) {
if (constructorKey !== PRIVATE_CONSTRUCTOR_KEY) {
throw new Error('You must use the PrivateConstructorClass.create() to construct an instance.')
}
this.arg1 = arg1
this.arg2 = arg2
this.argN = argN
}
static create(arg1, arg2, argN) {
return new PrivateConstructorClass(arg1, arg2, argN, PRIVATE_CONSTRUCTOR_KEY)
}
}
// From Another JS File:
try {
const myFailedInstanceA = new PrivateConstructorClass('foo', 123, {
size: 'n'
})
} catch (err) {
console.error('Failed:', err.message)
}
const myFactoryInstance = PrivateConstructorClass.create('foo', 123, {
size: 'n'
})
console.log('Success:', myFactoryInstance)

Another possible simple approach is to use predicate function instead of instanceof. For typescript it can be a type guard and type synonym instead of a class can be exported:
// class is private
class _Score {
constructor() {}
}
export type Score = _Score
export function isScore(s): s is Score {
return s instanceof _Score
}

So to be fair the simplest answer is usually the best. An object literal is always a single instance. Not much reason for anything more complex other than, perhaps allocation of memory on demand.
That being said, here is a classical implementation of a singleton using ES6.
The instance "field" is "private". This really means we hide the instance as a property of the constructor. Somewhere not Constructor.prototype, which will be available to the instance through prototipical inheritance.
The constructor is "private". We really are just throwing an error when the caller is not the static getInstance method.
Also of note. It’s important to understand what the keyword this means in different contexts.
In the constructor, this points to the instance being created.
In the static getInstance method, this points to the left of the dot, Universe constructor function which, is an object like most things in JS and can hold properties.
class Universe {
constructor() {
if (!((new Error).stack.indexOf("Universe.getInstance") > -1)) {
throw new Error("Constructor is private. Use static method getInstance.");
}
this.constructor.instance = this;
this.size = 1;
}
static getInstance() {
if (this.instance) {
return this.instance;
}
return new this;
}
expand() {
this.size *= 2;
return this.size;
}
}
//console.log(Universe.getInstance())
//console.log(Universe.getInstance().expand())
//console.log(Universe.getInstance())
//console.log(new Universe())
const getInstance= () => { console.log('hi');
console.log("From singleton: ", Universe.getInstance()); return new Universe() };
console.log(getInstance())

Getter/setter in constructor

I recently read about the fact that there is a possibility of defining getters/setters in JavaScript. It seems extremely helpful - the setter is a kind of 'helper' which can parse the value to be set first, before actually setting it.
For example, I currently have this code:
var obj = function(value) {
var test = !!value; // 'test' has to be a boolean
return {
get test() { return test },
set test(value) { test = !!value }
};
};
var instance = new obj(true);
This code always converts value to a boolean. So if you code instance.test = 0, then instance.test === false.
However, for this to work you have to actually return an object, which means that the new instance is not of type obj but just is a plain object. This means that changing the prototype of obj has no effect on instances. For example, this does not work - instance.func is undefined:
obj.prototype.func = function() { console.log(this.value); };
because instance is not of type obj. To get the prototype functions work, I guess I should not return a plain object, but rather not return anything so that instance would just be of type obj, like a regular constructor works.
The problem then is how to implement getters/setters? I can only find articles describing how to add these to an object, not as being part of the constructor of a custom type.
So how do I implement getters/setters in the constructor so as to be able to both use getters/setters and extending the prototype?

You can't do that.
You can set setter/getters for properties of objects though. I advice you use ES5 Object.defineProperties though. of course this only works in modern browsers.
var obj = function() {
...
Object.defineProperties(this, {
"test": {
"get": function() { ... },
"set": function() { ... }
}
});
}
obj.prototype.func = function() { ... }
var o = new obj;
o.test;
o.func();

Usually you want class methods. The answer by #Raynos on May 7, 2011 gets the job done, but it defines an instance method, not a class method.
The following illustrates a class definition with a the getter and setter being part of the class. This definition is a lot like the answer by #Raynos, but with two differences in the code: (1) The "defineProperties()" action has been moved out of the constructor. (2) The argument to "defineProperties()"as been changed from the instance object "this", to the constructor's prototype object.
function TheConstructor(side) {
this.side = side;
}
Object.defineProperties(TheConstructor.prototype, {
area: {
get: function() { return this.side * this.side; }
,set: function(val) { this.side = Math.sqrt(val); }
}
});
// Test code:
var anInstance = new TheConstructor(2);
console.log("initial Area:"+anInstance.area);
anInstance.area = 9;
console.log("modified Area:"+anInstance.area);
Which produces these results:
initial Area:4
modified Area:9
Although usually the distinction between class versus instance
definition is just a matter of style, there is a purpose to
good style, and there is a case where the distinction matters:
the memoized getter. The purpose for a memoized getter is
described here: Smart/self-overwriting/lazy getters
Define the getter at the class level when the memoized value is to
pertain to the entire class. For example, a configuration file
should be read only once; the resulting values should then apply
for the duration of the program. The following sample code
defines a memoized getter at the class level.
function configureMe() {
return 42;
}
Object.defineProperties(TheConstructor.prototype, {
memoizedConfigParam: {
get: function() {
delete TheConstructor.prototype.memoizedConfigParam;
return TheConstructor.prototype.memoizedConfigParam = configureMe();
}
,configurable: true
}
});
// Test code:
console.log("memoizedConfigParam:"+anInstance.memoizedConfigParam);
Produces:
memoizedConfigParam:42
As can be seen in the example, memoized getters have the
characteristic that the getter function deletes itself,
then replaces itself with a simple value that
(presumably) will never change.
Note that 'configurable' must be set to 'true'.
Define the getter at the instance level when the memoized value
depends upon the contents of instance. The definition moves
inside the constructor, and the object of attention is 'this'.
function TheConstructorI(side) {
this.side = side;
Object.defineProperties(this, {
memoizedCalculation: {
get: function() {
delete this.memoizedCalculation;
return this.memoizedCalculation = this.expensiveOperation();
}
,configurable: true
}
});
}
TheConstructorI.prototype.expensiveOperation = function() {
return this.side * this.side * this.side;
}
//Test code:
var instance2 = new TheConstructorI(2);
var instance3 = new TheConstructorI(3);
console.log("memoizedCalculation 2:"+instance2.memoizedCalculation);
console.log("memoizedCalculation 3:"+instance3.memoizedCalculation);
Produces:
memoizedCalculation 2:8
memoizedCalculation 3:27
If you want to guarantee (rather than presume) that the memoized
value will never be changed, the 'writable' attribute needs to
be changed. That makes the code a bit more complicated.
function TheConstructorJ(side) {
this.side = side;
Object.defineProperties(this, {
memoizedCalculation: {
get: function() {
delete this.memoizedCalculation;
Object.defineProperty( this, 'memoizedCalculation'
,{ value : this.expensiveOperation()
,writable : false
});
return this.memoizedCalculation;
}
,configurable: true
}
});
}
TheConstructorJ.prototype.expensiveOperation = function() {
return this.side * this.side * this.side;
}
//Test code:
var instanceJ = new TheConstructorJ(2);
console.log("memoizedCalculation:"+instanceJ.memoizedCalculation);
instanceJ.memoizedCalculation = 42; // results in error
Produces:
memoizedCalculation:8
>Uncaught TypeError: Cannot assign to read only property 'memoizedCalculation' of object '#<TheConstructorJ>'
The OP's original question, from March 7, 2011, presented basic
getter and setter syntax, noted that it worked on an object but
not on 'this', and asked how to define getters and setters within
a constructor. In addition to all the examples above, there is
also a "cheap-shot" way of doing it: create a new object within
the constructor, like the OP did, but then assign the object to
be a member within 'this'. So, the original code would look like
this:
var MyClass = function(value) {
var test = !!value; // 'test' has to be a boolean
this.data = {
get test() { return test },
set test(value) { test = !!value }
};
};
var instance = new MyClass(true);
// But now 'data' is part of the access path
instance.data.test = 0;
console.log(instance.data.test);
Produces:
false
Believe it or not, I have actually run into situations where
this "cheap-shot" is the best solution. Specifically, I used this
technique when I had records from several tables encapsulated within
a single class, and wanted to present a unified view as though
they were a single record called 'data'.
Have fun.
IAM_AL_X

Update for ES6 -- have a look at section 19.3.1 of Alex Rauschmayer's book Exploring ES6 http://exploringjs.com/es6/ch_maps-sets.html#sec_weakmaps-private-data which demonstrates how to use WeakMaps with getters and setters to hold private data. Combining with section 16.2.2.3 http://exploringjs.com/es6/ch_classes.html#leanpub-auto-getters-and-setters would result in something like
# module test_WeakMap_getter.js
var _MyClassProp = new WeakMap();
class MyClass {
get prop() {
return _MyClassProp.get( this );
}
set prop(value) {
_MyClassProp.set( this, value );
}
}
var mc = new MyClass();
mc.prop = 5 ;
console.log( 'My value is', mc.prop );
$ node --use_strict test_WeakMap_getter.js
My value is 5

function Obj(value){
this.value = !!value;
}
Obj.prototype = {
get test () {
return this.value;``
},
set test (value) {
this.value = !!this.value;
}
};
var obj = new Obj(true);

I know this might be extremely late but I figured out a different way to accomplish what you want and for the sake of people, like myself, googling for an answer to this here it is.
function Constructor(input){
this.input = input;
}
Object.__defineGetter__.call(Constructor.prototype, "value", function(){
return this.input * 2;
});
var test = new Constructor(5);
alert(test.value) // 10
I've tested this in chrome, safari, mobile safari, firefox and they all work (latest versions of course)

#Alex I see it as more option and more power, programming is art, #Nat share his finding with us, and for that I thank him. Maybe someone want to do it that way.
I'm sure the setter version is the same but just changing that g to a s.
i.g:
function Constructor(input){
this.input = input;
}
Object.__defineGetter__.call(Constructor.prototype, "value", function(){
return this.input * 2;
});
Object.__defineSetter__.call(Constructor.prototype, "bar", function(foo){
return this.input *= foo;
});
var test = new Constructor(5);
console.log(test.value); // 10
test.bar = 5;
console.log(test.input); //25
With that said, this feature is deprecated, advices to not to use in production coding.

Develop Reference

JavaScript is the programming language of the Web.

javascript and memoized getters - javascript

No, there is no language-level support for memoized getters in JavaScript. In your second example, a new object would be created every time boo was accessed.

You can do it all on one line: class Foo() { get boo() { return this._boo = this._boo || new Boo(); } } This is easy to remember and isn't convoluted. Got to consider the maintenance factor, and try to keep things simple.

Related

Is it possible to instantiate a class without a `new` operator? [duplicate]

Unable to have get and set method in Object Constructor (not in class) in JavaScript? [duplicate]

Make JS class callable without side effects [duplicate]

How to define private constructors in javascript?

Getter/setter in constructor

Categories

Resources