DEV Community

Adrian Matei for CodepediaOrg

Posted on • Originally published at codepedia.org

Javascript call by value or by reference, actually by sharing

Still confused about how passing variables in Javascript functions works? So was I, until recently. It took me some effort to understand, and I'd like to share my understanding with an example.

First try guessing the outcome of the following javascript snippet

Input

const number = 1983
const string = 'Adrian'
let obj1 = {
  value: 'obj1'
}
let obj2 = {
  value: 'obj2'
}
let obj3 = obj2;

let obj4 = ['a'];


function change(numberParam, stringParam, obj1Param, obj2Param, obj4Param) {
    console.log('\nSTART logs in function');

    numberParam = numberParam * 10;
    stringParam = 'Ionut';
    console.log('numberParam - ', numberParam);
    console.log('stringParam - ', stringParam);

    console.log('obj1Param.value in function before obj1Param = obj2Param assignment - ', obj1Param.value);
    obj1Param = obj2Param;
    console.log('obj1Param.value in function after obj1Param = obj2Param assignment - ', obj1Param.value);

    console.log('obj2Param.value in function before obj2Param.value change - ', obj2Param.value);
    obj2Param.value = 'changed'; // obj1Param.value = 'changed'; would yield the same result
    console.log('obj1Param.value in function after obj2Param.value change - ', obj1Param.value);
    console.log('obj2Param.value in function after obj2Param.value change - ', obj2Param.value);

    //obj4Param = ['b'];
    obj4Param.push('b');
    console.log('obj4Parma - ', obj4Param);

    console.log('END logs in function \n');
}

change(number, string, obj1, obj2, obj4);

console.log('number final - ', number);
console.log('string final - ', string);
console.log('obj1.value final - ', obj1.value);
console.log('obj2.value final - ', obj2.value);
console.log('obj3.value final - ', obj3.value);
console.log('obj4 final - ', obj4);
Enter fullscreen mode Exit fullscreen mode

Before you read the output I suggest you read the explanation of Call by sharing on Wikipedia, which is the best explanation on the topic I found.

Output

START logs in function
numberParam -  19830
stringParam -  Ionut
obj1Param.value in function before obj1Param = obj2Param assignment -  obj1
obj1Param.value in function after obj1Param = obj2Param assignment -  obj2
obj2Param.value in function before obj2Param.value change -  obj2
obj1Param.value in function after obj2Param.value change -  changed
obj2Param.value in function after obj2Param.value change -  changed
obj4Parma -  ["b"]
END logs in function

number final -  1983
string final -  Adrian
obj1.value final -  obj1
obj2.value final -  changed
obj3.value final -  changed
obj4 final -  ["a"]
Enter fullscreen mode Exit fullscreen mode

Ok, so what's going on?

  • number and string primitives are "boxed"1 in Number and String objects2 before passing. Boxed objects are always a copy of the value object, hence new objects (Number and String) are created in memory with the same primitive values. In the function execution (scope) they get "unboxed", their value is changed and placed in the new memory space, but once the function is over the new space in memory is cleared, with the original remaining unaffected.
  • a copy reference to obj1 and obj2 is passed to the function, pointing to the same address of the "original" objects in memory (call by sharing)3. With the obj1Param = obj2Param assignment in the function, both obj1Param and obj2Param to the original obj2 object in memory, so when changing its property obj2Param.value = 'changed' it will also be visible outside of the function scope, after it is terminated. obj1Param.value = 'changed' would have had the same effect after the assignment.
  • what about obj4? obj4param is also a copy reference to the obj4 object (remember in Javascript arrays are objects), but with the obj4Param = ['b'] assignment it's pointing now to a newly created object (the ['b'] array object), which is visible only in the function's scope and is destroyed when the function is over. Thus, it has no effect on the original object. On the other hand, a statement like obj4param.push('b') would have changed the original array and would display a  ["a", "b"] value.

Shared with love from Codever. Use the Copy to mine functionality to copy this snippet to your own personal collection and easy manage your code snippets.


  1. https://en.wikipedia.org/wiki/Object_type_(object-oriented_programming)#Boxing 

  2. https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects 

  3. https://en.wikipedia.org/wiki/Evaluation_strategy#Call_by_sharinghttps://en.wikipedia.org/wiki/Evaluation_strategy#Call_by_sharing 

Top comments (7)

Collapse
 
peerreynders profile image
peerreynders • Edited

number and string primitives are "boxed" in Number and String objects before passing.

The Wikipedia reference is talking about Java - not JavaScript.

MDN: Primitive values:

All types except objects define immutable values.

The mental model you are presenting is way too complicated.

Have a look at this recent discussion.

Given that all primitive values are immutable it makes sense to unify the perspective of "how variables work".

// Bind immutable number `1983` to name `number`
let number = 1983;

// Bind immutable string `Adrian` to name `string`
let string = 'Adrian'

// Bind object instance `obj1` to name `obj1`
const obj1 = {
  value: 'obj1'
}

// Bind object instance `obj2` to name `obj2`
const obj2 = {
  value: 'obj2'
}

// Bind object instance `obj2` to name `obj3`;
const obj3 = obj2;

// Bind array instance `obj4` to name `obj4` 
const obj4 = ['a'];

// Bind an entirely new immutable number that 
// is the result of `1983 * 10` to name `number`
//
// The immutable number `1983` continues to exist 
// but without a name referring to it
// it will be eventually garbage collected
//
number = number * 10;

// Bind an entirely new immutable string 'Ionut' 
// to name `string`
//
// The immutable string `Adrian` continues to exist
// but without a name referring to it
// it will be eventually garbage collected
//
string = 'Ionut';

// Bind the immutable string 'changed' to 
// the `value` property on 
// the `obj2` object instance (via the `obj3` name)
// 
// The immutable string 'obj2' continues to exist 
// but without a name referring to it
// it will be eventually garbage collected
//
obj3.value = 'changed';

// Push immutable string 'b' onto array instance `obj4` (via `obj4` name)
obj4.push('b')
Enter fullscreen mode Exit fullscreen mode

At this point it should become clear that function arguments upon invocation simply work like this:

// Bind immutable number currently bound to `number` to name `numberParam`;
let numberParam = number;

// Bind immutable string currently bound to `string` to name `stringParam`;
let stringParam = string;

// Bind object instance `obj1` to name `obj1Param`;
let obj1Param = obj1;

// Bind object instance `obj2` to name `obj2Param`;
let obj2Param = obj2;

// Bind object instance `obj4` to name `obj4Param`;
let obj4Param = obj4;

// Bind an entirely new immutable number that 
// is the result of `1983 * 10` to name `numberParam`
//
// The immutable number `1983` continues to exist
// and the name `number` still refers to it.
//
numberParam = numberParam * 10;

// Bind an entirely new immutable string 'Ionut' 
// to name `stringParam`
//
// The immutable string `Adrian` continues to exist
// and the name `string` still refers to it.
//
stringParam = 'Ionut';

// Bind the immutable string 'changed' to 
// the `value` property on 
// the `obj2` object instance (via the `obj2Param` name)
// 
// The immutable string 'obj2' continues to exist 
// but without a name referring to it
// it will be eventually garbage collected
//
obj2Param.value = 'changed';

// Push immutable string 'b' onto array instance `obj4` (via `obj4Param` name)
obj4Param.push('b')
Enter fullscreen mode Exit fullscreen mode

That's it!

Note - arguments vs. parameters - MDN: Parameter:

  • Function parameters are the names listed in the function's definition.
  • Function arguments are the real values passed to the function.

Bonus: const simply means that the binding is permanent. Primitive values are already immutable - (internally) objects (and arrays which are objects) however are not immutable.

MDN: const:

The const declaration creates a read-only reference to a value. It does not mean the value it holds is immutable—just that the variable identifier cannot be reassigned.

Collapse
 
ama profile image
Adrian Matei • Edited

The Wikipedia entry does mention Javascript too -

Call by sharing (also known as "call by object" or "call by object-sharing") is an evaluation strategy first noted by Barbara Liskov in 1974 for the CLU language.[8] It is used by languages such as Python,[9] Java (for object references), Ruby, JavaScript, Scheme, OCaml, AppleScript, and many others.

I have found this good article JavaScript, Ruby and C are not call by reference... will correct and update if this is the case...

Collapse
 
peerreynders profile image
peerreynders • Edited

The core message of that article is:

call-by-sharing is call by value, but the value is always a reference.

It's always pass by value, but the value of the variable is a reference.

but the article also points out that the technical details around what a reference exactly is in any given language can be a bit prickly.

JavaScript isn't a low level language so the language spec simply describes how ECMAScript has to behave but does not specify how it has to be implemented. Engines such as V8 or SpiderMonkey can implement the runtime in whatever way they wish as long as the observable behaviour conforms to the specification.

So while it is correct to say that JavaScript behaves in the manner of "Call by Sharing", the implementation details are up to the actual runtime engine.

For a more terse reference - Call by Sharing:

The caller and called routine communicate only through the argument and result objects; routines do not have access to any variables of the caller.

The remainder I would rephrase this way:

After the assignments of actual arguments to formal arguments, the caller and the called routine share values. If the called routine modifies a shared value, the modification is visible to the caller on return. The names used to denote the shared values are distinct in the caller and called routine; if a routine assigns a value to a formal argument variable, there is no effect on the caller. From the point of view of the invoked routine, the only difference between its formal argument variables and its other local variables is that the formals are initialized by its caller.

And in the case of JavaScript primitive values are immutable so:

If the called routine modifies a shared value, the modification is visible to the caller on return.

only applies to objects (aka compound data). For primitive values it is only possible to

assigns a value to a formal argument variable, there is no effect on the caller.


What hampers most people is the metaphor they use to visualise "variables".

In Java courses they tend to get told that "variables" are like boxes:

  • A primitive data box can hold one primitive value. You can "vary" by you putting a different value in the box.
  • An object box holds a remote control (reference) to the object. If you want to change something about the object, do it through the remote control. There may be multiple remote controls with a connection to the same object.

That metaphor doesn't work very well for JavaScript. That is why I typically don't use the terms "variable" and "assignment" but rather "name" and "bind".

A more useful metaphor for

let number = 1983;
Enter fullscreen mode Exit fullscreen mode

is writing "number" on a sticky note and sticking it the immutable number 1983. Then

number = number * 10;
Enter fullscreen mode Exit fullscreen mode

creates an entirely new immutable number 19830 and repositions the "number" sticky note to it.

On the other hand

let numberParam = number;
Enter fullscreen mode Exit fullscreen mode

simply writes "numberParam" on another sticky note and sticks it on the same immutable number 1983 leaving two separate sticky notes on it. Now

numberParam = numberParam * 10;
Enter fullscreen mode Exit fullscreen mode

creates an entirely new immutable number 19830 and repositions the "numberParam" sticky note to it while leaving the "number" sticky note on the original 1983.

So

const number = 1983;
Enter fullscreen mode Exit fullscreen mode

is writing "number" on a permanent label and sticking it on that immutable number 1983. That label cannot be "repositioned".

And the beauty of the "sticky note" metaphor is that works for objects and arrays as well.

const obj4 = ['a'];
Enter fullscreen mode Exit fullscreen mode

So while obj4 is the permanent label for that particular array, the "sticky note" obj4[0] (a separate (indexed) name) can still be "repositioned" (to a different value) and more "sticky notes" obj4.push('b') can still be added.

On a technical level these "sticky notes" (or "permanent labels") are just references - the JavaScript way.

Thread Thread
 
sleeplessbyte profile image
Derk-Jan Karrenbeld

^ this :)

Thread Thread
 
kingychiu profile image
Anthony Chiu • Edited

Thanks for this explanation here and also the "variables just point to values" mindset: in this post.

(Sorry for the long comment below in advance)

So let me try to summarize what I have learned here (hopefully is correct):

  1. For primitive types a=1, JS variable holds a reference (address) to the actual memory location that stores the value.
  2. For non-primitive types b=[1,2,3], the variable holds a reference (address) to the locations storing other references to the actual values of 1, 2, 3.
  3. For primitive types, the value are immutable, and the reassignment is copy(value) on write!! like a=1;b=a; b and a hold the same references/address to the same memory location. A new memory is allocated only if we try to assign b=2 for example. (what if we assign b=1? since 1 is immutable, will a=1;b=1; result in both a and b hold references tht point to the same place that storing that immutable 1?

It is shocking that even for primitive types, the JS variable only holds a reference (address in JS) to the actual value.

So basically, 90%+ of the materials out there are wrong, like this

A variable is a container for a value, like a number we might use in a sum, or a string that we might use as part of a sentence.

from developer.mozilla.org/en-US/docs/L...

I mainly have two questions here:
First: I learned most of the knowledge of a programming language internal from C++. Any reference materials that I can read about the virtual memory level of JavaScript? Like the stack vs heap in C++. For example in this blog, it describes let name = 'John'; stores the value in the stack frame of the stack, which is wrong based on the discussion here (right?)

Second: The second question I have is entirely the opposite. If I don't want to touch on any details about virtual memory, how can I learn or teach others about JavaScript? Since JavaScript is a high-level programming language, we should not care about memory (unless we need to optimize). But it seems like it is impossible to teach the correct concept without touching the memory and addresses.

I appreciate the new metaphor you suggested here for JavaScript, but now I worry if that variable metaphor idea will differ in different languages (Java vs JavaScript or even others). What is the point of learning that metaphor (I believe metaphor is an abstraction that should be helpful in quickly picking up different languages)? All in all, my second question is more about what is the right level of abstraction to learn/teach a programming language without touching the internals but learning/teaching it correctly?

Thread Thread
 
peerreynders profile image
peerreynders

storing other references to the actual values of 1, 2, 3.

That depends as the runtime can optimize. See Elements kinds in V8. Also there are Typed Arrays which hold the actual (uniformly typed) values.

It is shocking that even for primitive types, the JS variable only holds a reference (address in JS) to the actual value.

Remember that the runtime is free to optimize whichever way it seems fit as long as the observable behaviour aligns with the language specification. However it is dictated that the conceptual behaviour of a value of a primitive type is that of an "immutable datum represented directly at the lowest level of the language". Immutability is easily explained with Strings; you can't change a string, you can only create a new one, so on "assignment" the let name is essentially being rebound to the new string. But Boolean, Null Undefined, Number, BigInt, and Symbol values all behave the same way by virtue of being primitive types. I seem to remember coming across some V8 source code where primitive JavaScript values (at least in some context) were represented by C++ classes.

So basically, 90%+ of the materials out there are wrong

It's a simplified mental model which mostly works and is easy to grasp for beginners.

which is wrong based on the discussion here (right?)

Again the runtime implementation isn't dictated. V8 has the "Ignition" interpreter and the "Turbofan" compiler so actual JS execution could be accomplished in very different ways depending on the level of optimization.

if that variable metaphor idea will differ in different languages (Java vs JavaScript or even others).

My metaphor isn't original. Given that in JavaScript primitive values are immutable it's unfortunate that the term "assignment" is being used as that implies PLOP (PLace-Oriented Programming), i.e. (re)placing a value inside a location giving rise to the notion of a "variable" as the value inside the location can vary. In value-oriented programming (typically practiced in functional languages)

let number = 1983;
Enter fullscreen mode Exit fullscreen mode

is binding the name number to the value 1983 while

number = number * 10;
Enter fullscreen mode Exit fullscreen mode

is rebinding the name number to the result of the number * 10 expression. With that terminology it is the binding that changes, not the value.

I only stumbled upon this because I got annoyed at const—until I discovered:

"The const declaration creates a read-only reference to a value. It does not mean the value it holds is immutable—just that the variable identifier cannot be reassigned."

So people keep saying that const works differently for primitive values and objects (overlooking completely that primitive values are already immutable). This is a perspective that becomes necessary because of the assignment/variable oversimplification that people learn in the beginning. The truth is much simpler:

const doesn't act on the value but on the binding—it's the binding that is constant and unchangeable. As a result the const behaviour is absolutely identical for primitive values and objects as the binding being constant does not affect the internal mutability of objects (which includes arrays).

If I don't want to touch on any details about virtual memory

I hope I conveyed that going to that level isn't actually necessary. Once you eliminate the notion of variables and assignments and replace them entirely with names and bindings, everything behaves entirely consistently. It's just that beginners material insists on starting with variables and assignments and then never adjust that initial mental model.

Thread Thread
 
kingychiu profile image
Anthony Chiu

Thanks for your reply, let me study more with the given directions.

My imdidate response to this

Remember that the runtime is free to optimize whichever way it seems fit as long as the observable behaviour aligns with the language specification.

Yeah, it makes a lot of sense. For example, in C/C++, the "Optimization" I learned is like out-of-order execution (including speculation) and unfolding loops. I remember back then, I was shocked as well. Now I realize it is not just the JavaScript tutorials problem. One of the utimate problems is that the teaching materials don't separate what the language specifications are. But for non-specified behaviors, they are subjected to be changed/optimized. I think this is also a missing mental model puzzle here.

Back then, our university C++/Java course should also focus more on the language specifications first and make a clear separation between abstraction and implementation. Now I am confused about what I have learned (for all languages what their specifications/promises are).