Let's consider simple singleton implementation:
var singleton = function (Constructor) {
var singleton;
return function () {
if (!singleton) {
singleton = new Constructor();
}
return singleton;
};
};
We could move declaration of singleton variable to arguments:
var singleton = function (Constructor, singleton) {
return function () {
if (!singleton) {
singleton = new Constructor();
}
return singleton;
};
};
So I simply curious about side effects.
One more example:
var counter = (function (i) {
return function () {
i = (i || 0) + 1;
return i;
};
}());
We could move declaration of singleton variable to arguments
First, let's make it possible to talk about this without tying outselves up in knots by using the same symbol (singleton) for two completely different things within the same few lines of code.
Here's your seecond example renamed:
var singleton = function (Constructor, instance) {
return function () {
if (!instance) {
instance = new Constructor();
}
return instance;
};
};
If you did that, then calling the singleton function with two arguments would specify instance, making passing Constructor in pointless — Constructor would never be called by singleton (again, if you passed in two args [and the second arg was truthy]). So it would be a bit odd to do that.
But you asked about side-effects. There are no external effects involved if you assign to a formal argument within the function. It doesn't, for instance, have any effect outside the function:
function foo(arg) {
arg = 67;
}
var a = 42;
foo(a);
console.log(a); // Still 42
However, assigning to an argument rather than to a local variable does, in non-strict mode, have a small cost, because there's overhead involved: It relates to the magic arguments pseudo-array. In non-strict mode, there's a link between the formal arguments of a function and the arguments pseudo-array:
function foo(a, b) {
console.log(a);
console.log(b);
b = 42;
console.log(arguments[1]);
}
If we call that like this:
foo(1, 2);
we see
1
2
42
Note the magic: Assigning to the formal argument b updated the pseudo-array arguments. (It works the other way, too.) That link has a small but real runtime cost.
(In strict mode, there is no link, for this very reason.)
As far as closure goes, there is no difference in the two implementations. However, in first example, singleton is not settable by the caller.
I will choose between the two implementation, just based on whether singleton could have been created outside these example functions or not. If it can be, use model 2 otherwise use model 1
Related
I am curious why methods cannot call other methods or themselves in javascript. For example, this produces a Reference error saying add is not defined.
class sum {
add(x, amt) {
if(amt == 0) return x
return add(x+1, amt-1)
}
}
summer = new sum()
console.log(summer.add(5,5))
You must use this.add() instead.
Now I understand that the methods get translated to functions on the prototype but I don't see how that explains this limitation I'm pointing out?
Couldn't one reason that when add is defined it could have references to itself or other methods with a closure capture.
Why is it this way?
I was in the middle of illustrating this shortcoming when #briosheje made the comment:
function add() {
console.log('called the wrong add!');
}
class sum {
add(x, amt) {
if(amt == 0) return x
return add(x+1, amt-1)
}
}
summer = new sum()
console.log(summer.add(5,5))
I will try to make it "slightly" more theorical, without going too deep in the documentations and so on.
The main answer to your question can be found by transpiling your code to plain javascript. To accomplish that, you can either use babel online or the typescript online playground.
Either case, the transpiled code will look like this:
"use strict";
var sum = /** #class */ (function () {
function sum() {
}
sum.prototype.add = function (x, amt) {
if (amt == 0)
return x;
return add(x + 1, amt - 1);
};
return sum;
}());
var summer = new sum();
console.log(summer.add(5, 5));
As you can see, the add method belongs to the sum prototype, which is a function. Therefore, you can guess that accessing add whithin the add scope just can't implicitly lead to invoking the sum.prototype.add function.
Differently, if you look at the correct code:
class sum {
add(x, amt) {
if(amt == 0) return x
return this.add(x+1, amt-1)
}
}
var summer = new sum()
console.log(summer.add(5,5))
You will see that the transpiled code will invoke the this method:
"use strict";
var sum = /** #class */ (function () {
function sum() {
}
sum.prototype.add = function (x, amt) {
if (amt == 0)
return x;
return this.add(x + 1, amt - 1);
};
return sum;
}());
var summer = new sum();
console.log(summer.add(5, 5));
This is not really matter of being ambiguous, it's rather that, in javascript, such kind of invocation is allowed, because the add method is implicitly available from the global scope. Being able to access the global scope in your function scope (because remember that, whatever happens, a class in javascript is always transpiled to a function) allows to inherit the standard behavior of a function, which is having access to its parent, having its own scope and granting access to the global scope.
A little curiosity: if you actually could access this.add using add, you wouldn't be able to use undefined, since it's a global variable, hence you wouldn't be able to access it and use it, because it would implicitly be this.undefined.
So, once again, it's not about ambiguity, it's about how javascript functions works.
Fundamentally OOP in Javascript is very different to java and other languages. There's not actually a strong concept of "classes". The classic OOP concepts like classes and instances in Javascript don't really come from "classes", but from the rules surrounding property lookups and the this and new keywords. Everything else is just functions, objects and properties—some of which are "special", like prototype.
E.g. you can assemble a "class" like this:
function Foo() {
this.bar = 42;
}
function baz() {
return this.bar;
}
Foo.prototype.baz = baz;
let foo = new Foo;
console.log(foo.baz());
There's no real cohesion here, baz doesn't intrinsically belong to any class, yet assembled like this they act like one. And the class syntax is just a thin veneer over this mechanism. So, if you write add inside a function, it follows variable scoping rules and will look up a variable in some surrounding scope, it's not going to find a method on a prototype.
In want to define a function-constructor inside a namespace. The way in which I defined the constructor till now was a simple constructor function without NS, combined with prototypal inheritance.
The code looked kind of like:
function mySuperObject() {
var self = this;
self.p1 = "";
self.p2 = "jkejie";
// and others
}
mySuperObject.prototype.func1 = function(){...}
// and others
Introducing namespace:
After reading many articles I decided to start up with a very simple way to define a namespace, maybe the most simplest.
Basically it is just about defining a variable which points to an object-literal and the content is the object (the "mySuperObject" in the code-snippet above). The constructor function is following: mySuperObjectInNS.
The code of the object:
var MYNAMESPACE = {
//some variable outside the object
file : "ConstructorInNamespace_obj.js: ",
//Defining contructor function inside a namespace
mySuperObjectInNS : function(propOne, propTwo){
var self = this;
self.objectName = "mySuperObject";
self.propertyOne = propOne;
self.propertyTwo = propTwo;
self.doSomething = function(){
console.log(file + " doSomething called - function of object");
};
///many more functions and attributes
}
}
MYNAMESPACE.mySuperObjectInNS.prototype.specialFunction = function(){
console.log(file + " specialFunction called - prototypical inheritance defined in file of object, outside of namespace");
};
///many more functions and attributes
In another file it is possible to intantiate the object, as follows:
...
var objOne = new MYNAMESPACE.mySuperObjectInNS("param11", "40");
//following line works just fine
objOne.doSomething();
....
Questions:
It seems to me that this all is about defining an Object-Literal and
I will come into trouble the latest I am trying to define "private"
properties of that object. Is this correct?
Is that mySuperObjectInNS still a constructor function? (For me it
seems it is something else,even if I can intantiate objects from it.
Is is a very bad very bad way of namespacing or is kind of ok?
It seems to me that this all is about defining an Object-Literal and I will come into trouble the latest I am trying to define "private" properties of that object. Is this correct?
"Private properties" have nothing to do with using an object for namespacing. In fact, originally when answering this question, I read that as "private functions" because that would be relevant.
There are lots of ways to do private and semi-private properties in JavaScript, but they relate to how you create the constructor function and the methods it gives the object, not how you expose the constructor function. "Namespace" objects are about how you expose the constructor.
One common pattern for creating "private" properties is to define methods that need to access them within the constructor, and make the "properties" local variables within the constructor (so they aren't really properties at all), like this:
function SuperObject() {
var privateInformation;
this.method = function() {
// This can access `privateInformation`, which is completely
// hidden from outside the constructor
};
}
It doesn't matter at all if you do that within a "namespacing" pattern or on its own.
Private functions, on the other hand, affect the pattern. I'll show both below.
A fairly common variant that provides for private functions is to use a function to create the object, which also gives you the opportunity to create private functions:
var TheNamespace = function() {
function privateFunction() {
}
function SuperObject() {
var privateInformation;
this.method = function() {
// This can access `privateInformation`, which is completely
// hidden from outside the constructor
};
}
SuperObject.prototype.otherMethod = function() {
// Can't access `privateInformation`, but has the advantage
// that the function is shared between instances
};
return {
SuperObject: SuperObject
};
}();
// usage
var s = new TheNamespace.SuperObject();
Is that mySuperObjectInNS still a constructor function? (For me it seems it is something else,even if I can intantiate objects from it.
Yes. A constructor function is any function that expect you to use new with it.
Is is a very bad very bad way of namespacing or is kind of ok?
Using objects as pseudo-namespaces is common practice. You might also consider various asynchronous module definition (AMD) technologies, which largely make "namespace" objects largely unnecessary.
Re your comment:
You defined a self-invoking function....which returns an an object?
It's not a self-invoking function, it's an inline-invoked function, but yes, it's a function that returns an object.
(if so I think parantheses are missing)
No, we don't need any parens that aren't there because the only reason for the outer parens other places you've seen this are to tell the parser that the word function starts an expression rather than declaration; we don't need that in the above because we're already on the right-hand side of an assignment, so there's no ambiguity when function is encountered.
Due to you proposed this way, is it a better way to define the ns?
"Better" is a subjective term. It gives you a scope in which you can define private functions, which you'd asked about.
Whereas I often also saw the option: var = {} | someNSName; What is this all about?
If you have several files that will add things to the "namespace" (as is common), then you frequently see this in each of them:
var TheNamespace = TheNamespace || {};
What that does is declare the variable if it hasn't been declared before, and assign an empty object to it if it doesn't already have one. In the first file that gets loaded, this happens:
The var is processed and creates a new variable, TheNamespace, with the value undefined.
The TheNameSpace = TheNameSpace || {} assignment is processed: Since undefined is falsey, the curiously-powerful || operator results in the new {}, which gets assigned to TheNamespace.
When the next file loads, this happens:
The var is a no-op, because the variable already exists.
The TheNameSpace = TheNameSpace || {} assignment is processed: Since TheNamespace has a non-null object reference, it's truthy, and the curiously-powerful || operator results in a reference to the object TheNamespace refers to.
That is, it has no effect at all.
This is used so you can load the files in any order, or load just one file in isolation.
Here's an example:
thingy.js:
var TheNamespace = TheNamespace || {};
TheNamespace.Nifty = function() {
function privateFunction() {
}
function Nifty() {
var privateInformation;
this.method = function() {
// Can access `privateInformation` here
};
}
Nifty.prototype.otherMethod = function() {
// ...
};
return Nifty;
}();
thingy.js:
var TheNamespace = TheNamespace || {};
TheNamespace.Thingy = function() {
function privateFunction() {
}
function Thingy() {
var privateInformation;
this.method = function() {
// Can access `privateInformation` here
};
}
Thingy.prototype.otherMethod = function() {
// ...
};
return Thingy;
}();
There are lots of variations on that basic pattern, particularly if one file may add multiple things to TheNamespace. Here's one that supports doing so fairly concisely:
var TheNamespace = function(exports) {
function privateFunction() {
}
function Nifty() {
var privateInformation;
this.method = function() {
// Can access `privateInformation` here
};
}
Nifty.prototype.otherMethod = function() {
// ...
};
exports.Nifty = Nifty;
function Thingy() {
var privateInformation;
this.method = function() {
// Can access `privateInformation` here
};
}
Thingy.prototype.otherMethod = function() {
// ...
};
exports.Thingy = Thingy;
}(TheNamespace || {});
So I have this code:
function fn(a){
var f=(new Function("return a"));
return f();
}
fn(7)//ReferenceError: a is not defined
Same problems with local variables:
function fn(){
var a=7;
var f=new Function("return a");
return f();
}
fn(7)//ReferenceError: a is not defined
I want it to return a but the new function cant see a,
it can only see global a
var a=1;
function fn(a){
var f=(new Function("return a"));
return f();
}
fn(7)//1
With the normal initialization the function can see the argument.
function fn(a){
var f=function(){return a};
return f();
}
fn(7)//7
I need to call the basic constructor in my project and can't use global variables.
I know that i could solve this by giving arguments to the newly created function and call it with that like this:
function fn(a){
var f=(new Function('a',"return a"));
return f(a);
}
fn(7)//7
And also could use some parsing function and some stupidly long method to make incoming arguments reachable like this:
function parsargs(funct){
//some parsing methodes giving back argument name list from funct.toString()
return "['a','b']";//like this
}
function fn(a,b){
var arrgstr,retfunc;
arrgstr="";
for(var i in arguments)
{
if(i<arguments.length-1)
arrgstr+=arguments[i]+",";
else
arrgstr+=arguments[i];
}
//return arrgstr;
retfunc="var f=new Function("+parsargs()+",'return b*a');return f("+arrgstr+")";
return (new Function(retfunc))();
}
fn(7,4)//28
But there must be an easier way which reaches local variables and functions as well...
Any suggestions?
PS:
i am trying to replace eval() in the project
Here is a simplified version of my original problem:
fiddle
The answer is NO...
Your exact question isn't clear but, supposing you can use arguments.callee (i.e. non strict mode) and that you want to be able to have any arguments name in fn, you "may" do this:
function fn(a,b){
var strargs = arguments.callee.toString().match(/\(([^\)]*)\)/)[1];
return (new Function(strargs.split(","),"return a+b")).apply(null,arguments);
}
console.log(fn(7, 3)) // 10
But I have a strong feeling this is a XY question and that we could have given a more useful answer knowing the real original problem to solve.
You could call your new Function with a context that references the local variables that you need:
function f(a) {
var b = 30;
return new Function("return this.a + this.b").call({ a: a, b: b })
}
f(10) // 40
The reason why this isn't working as easy as you would like is that JavaScript doesn't blindly put variables in a scope. Instead, it parses the function body (as good as it can) and determines which variables the code will need. It will then look up the variables in the outer scopes, create references for them and attach those references to the new function. That's what keeps variables from the outer scope accessible inside of the function when it eventually will be executed.
You want to build a function from a string body. JavaScript can't tell which variables the body will use and hence, it won't make those variables available.
To make it work for function arguments, use apply()::
function fn(a){
var argref = arguments; // create reference to the arguments of fn() and hence to a or any other arguments
var func = new Function("return arguments[0]");
var wrapper = function() {
return func.apply(null, argref);
};
return wrapper;
}
Note that you still can't reference arguments by name since the code never specifies what the names of the arguments of func are - JavaScript can't magically read you mind and do what you want. If you want to access the arguments by name, you need to tell the interpreter the names.
This question has some code how to determine the names from a function reference: How to get function parameter names/values dynamically from javascript
I don't see a way to make local variables available without passing them to fn() as arguments. Even if you used them inside of fn(), they would be out of scope when func() is eventually executed.
A simple solution would be to pass an object to fn():
function fn(conf) {
var func = new Function('conf', "return conf.a");
var wrapper = function(conf) {
return func.apply(null, conf);
};
return wrapper;
}
fn({a:7});
new Function doesn't create a closure context, but eval does.
Here's a low-tech way to build a function with an arbitrary evaluated body and access to the calling scope:
function f(a) {
return eval("(function() { return a })")()
}
f(10) // 10
Following are two ways to define BW.Timer. Can someone tell me what the difference is? I am not certain the first is even valid, but if it is valid, what is different about using the myfunc=(function(){}()) syntax?
BW.Timer = (function () {
return {
Add: function (o) {
alert(o);
},
Remove: function (o) {
alert(o);
}
};
} ());
And...
BW.Timer = function () {
return {
Add: function (o) {
alert(o);
},
Remove: function (o) {
alert(o);
}
};
};
The first is the return-value of the immediately-invoked function. The second is a function. It essentially comes down to what the difference is between these:
var f = (function() { return 0; })();
var f = function() { return 0; };
Since the first function is called immediately, the value of 0 is given to the variable f. The first f is not a function. However, the second f we must call in order to get the value:
f(); // 0
It follows that in your example, the first BW.Timer is the object literal itself and the second is a function returning an object literal. You must call the function in order to get to the object:
BW.Timer().Add(x);
Why use the first then?
You might ask yourself why one would use a syntax like a = (function() { return {}; })() instead of a = {}, but there's a good reason. An IIFE (Immeditately-Invoked Function Expression), unlike a regular function allows the emulation of static variables (variables that maintain their value through a single instance). For example:
var module = (function() {
var x = 0;
return { get: function() { return x },
set: function(n) { x = n }
};
})();
The above a text-book example of the Module Pattern. Since the function is called right away, the variable x is instantiated and the return value (the object) is given to module. There's no way we can get to x other than by using the get and set methods provided for us. Therefore, x is static, meaning its variable won't be overridden each time you use module.
module.set(5);
module.get(); // 5
On the other hand, let's see an example where module is declared as a function instead:
// assume module was made as a function
module().set(5);
module().get(); // 0
When we call module() the x variable is overridden each time. So we're effectively using different instances of module and x each time we call module.
The difference is rather large.
In the first case, BW.Timer is executed when it is first encountered, and that is the static version assigned to BW.Timer. In that instance, BW.Timer.Add(1) may be used. Each call to BW.Timer will be the same object.
In the second case, BW.Timer is not executed when first encountered, and instead is a function referece which must be invoked BW.Timer(). For Add to be used, this must be the case BW.Timer().Add(1). Also, you can issue var timer = new BM.Timer();. Each instance of BW.Timer() will be unique here.
In the first example BW.Timer references an object that the self-executing function returns, while in the second example it references a function object, in other words it's a function that can be executed BW.Timer().
Something like this works for the global Function.prototype.
Function.prototype.aaa = 1
(function () {}).aaa // => 1
But is there a way to put inherited properties of functions without changing Function.prototype?
function MyFunction () { return function () {} }
MyFunction.prototype.bbb = 2
// Can I have (new MyFunction).bbb ?
Your only other choice is to create your own Function factory function, since you've ruled out modifying the mechanism used by the standard Function factory. And of course, that's unlikely to do what you want it to do, since everyone would have to use your factory. :-)
Modifying the Function.prototype isn't necessarily evil. Prototype's been doing it for years, mostly to good effect.
function Wrap() {
var f = new Function;
/* for (var k in Wrap.prototype) {
f[k] = Wrap.prototype[k];
}
return f; */
// return $.extend(f, Wrap.prototype);
return _.extend(f, Wrap.prototype);
}
Wrap.prototype.foo = 42;
(Wrap()).foo === 42; // true
Create a new function then extend that function with properties from your Wrap.prototype.
Either use jQuery, underscore or a simple for loop.