Answer by Daniel Pryden for What's the difference between passing by reference vs. passing by value?

Many answers here (and in particular the most highly upvoted answer) are factually incorrect, since they misunderstand what "call by reference" really means. Here's my attempt to set matters straight.

TL;DR

In simplest terms:

• call by value means that you pass values as function arguments
• call by reference means that you pass variables as function arguments

In metaphoric terms:

• Call by value is where I write down something on a piece of paper and hand it to you. Maybe it's a URL, maybe it's a complete copy of War and Peace. No matter what it is, it's on a piece of paper which I've given to you, and so now it is effectively your piece of paper. You are now free to scribble on that piece of paper, or use that piece of paper to find something somewhere else and fiddle with it, whatever.
• Call by reference is when I give you my notebook which has something written down in it. You may scribble in my notebook (maybe I want you to, maybe I don't), and afterwards I keep my notebook, with whatever scribbles you've put there. Also, if what either you or I wrote there is information about how to find something somewhere else, either you or I can go there and fiddle with that information.

What "call by value" and "call by reference" don't mean

Note that both of these concepts are completely independent and orthogonal from the concept of reference types (which in Java is all types that are subtypes of Object, and in C# all class types), or the concept of pointer types like in C (which are semantically equivalent to Java's "reference types", simply with different syntax).

The notion of reference type corresponds to a URL: it is both itself a piece of information, and it is a reference (a pointer, if you will) to other information. You can have many copies of a URL in different places, and they don't change what website they all link to; if the website is updated then every URL copy will still lead to the updated information. Conversely, changing the URL in any one place won't affect any other written copy of the URL.

Note that C++ has a notion of "references" (e.g. int&) that is not like Java and C#'s "reference types", but is like "call by reference". Java and C#'s "reference types", and all types in Python, are like what C and C++ call "pointer types" (e.g. int*).

OK, here's the longer and more formal explanation.

Terminology

To start with, I want to highlight some important bits of terminology, to help clarify my answer and to ensure we're all referring to the same ideas when we are using words. (In practice, I believe the vast majority of confusion about topics such as these stems from using words in ways that to not fully communicate the meaning that was intended.)

To start, here's an example in some C-like language of a function declaration:

void foo(int param) {  // line 1
  param += 1;
}

And here's an example of calling this function:

void bar() {
  int arg = 1;  // line 2
  foo(arg);     // line 3
}

Using this example, I want to define some important bits of terminology:

• foo is a function declared on line 1 (Java insists on making all functions methods, but the concept is the same without loss of generality; C and C++ make a distinction between declaration and definition which I won't go into here)
• param is a formal parameter to foo, also declared on line 1
• arg is a variable, specifically a local variable of the function bar, declared and initialized on line 2
• arg is also an argument to a specific invocation of foo on line 3

There are two very important sets of concepts to distinguish here. The first is value versus variable:

• A value is the result of evaluating an expression in the language. For example, in the bar function above, after the line int arg = 1;, the expression arg has the value 1.
• A variable is a container for values. A variable can be mutable (this is the default in most C-like languages), read-only (e.g. declared using Java's final or C#'s readonly) or deeply immutable (e.g. using C++'s const).

The other important pair of concepts to distinguish is parameter versus argument:

• A parameter (also called a formal parameter) is a variable which must be supplied by the caller when calling a function.
• An argument is a value that is supplied by the caller of a function to satisfy a specific formal parameter of that function

Call by value

In call by value, the function's formal parameters are variables that are newly created for the function invocation, and which are initialized with the values of their arguments.

This works exactly the same way that any other kinds of variables are initialized with values. For example:

int arg = 1;
int another_variable = arg;

Here arg and another_variable are completely independent variables -- their values can change independently of each other. However, at the point where another_variable is declared, it is initialized to hold the same value that arg holds -- which is 1.

Since they are independent variables, changes to another_variable do not affect arg:

int arg = 1;
int another_variable = arg;
another_variable = 2;

assert arg == 1; // true
assert another_variable == 2; // true

This is exactly the same as the relationship between arg and param in our example above, which I'll repeat here for symmetry:

void foo(int param) {
  param += 1;
}

void bar() {
  int arg = 1;
  foo(arg);
}

It is exactly as if we had written the code this way:

// entering function "bar" here
int arg = 1;
// entering function "foo" here
int param = arg;
param += 1;
// exiting function "foo" here
// exiting function "bar" here

That is, the defining characteristic of what call by value means is that the callee (foo in this case) receives values as arguments, but has its own separate variables for those values from the variables of the caller (bar in this case).

Going back to my metaphor above, if I'm bar and you're foo, when I call you, I hand you a piece of paper with a value written on it. You call that piece of paper param. That value is a copy of the value I have written in my notebook (my local variables), in a variable I call arg.

(As an aside: depending on hardware and operating system, there are various calling conventions about how you call one function from another. The calling convention is like us deciding whether I write the value on a piece of my paper and then hand it to you, or if you have a piece of paper that I write it on, or if I write it on the wall in front of both of us. This is an interesting subject as well, but far beyond the scope of this already long answer.)

Call by reference

In call by reference, the function's formal parameters are simply new names for the same variables that the caller supplies as arguments.

Going back to our example above, it's equivalent to:

// entering function "bar" here
int arg = 1;
// entering function "foo" here
// aha! I note that "param" is just another name for "arg"
arg /* param */ += 1;
// exiting function "foo" here
// exiting function "bar" here

Since param is just another name for arg -- that is, they are the same variable, changes to param are reflected in arg. This is the fundamental way in which call by reference differs from call by value.

Very few languages support call by reference, but C++ can do it like this:

void foo(int& param) {
  param += 1;
}

void bar() {
  int arg = 1;
  foo(arg);
}

In this case, param doesn't just have the same value as arg, it actually is arg (just by a different name) and so bar can observe that arg has been incremented.

Note that this is not how any of Java, JavaScript, C, Objective-C, Python, or nearly any other popular language today works. This means that those languages are not call by reference, they are call by value.

Addendum: call by object sharing

If what you have is call by value, but the actual value is a reference type or pointer type, then the "value" itself isn't very interesting (e.g. in C it's just an integer of a platform-specific size) -- what's interesting is what that value points to.

If what that reference type (that is, pointer) points to is mutable then an interesting effect is possible: you can modify the pointed-to value, and the caller can observe changes to the pointed-to value, even though the caller cannot observe changes to the pointer itself.

To borrow the analogy of the URL again, the fact that I gave you a copy of the URL to a website is not particularly interesting if the thing we both care about is the website, not the URL. The fact that you scribbling over your copy of the URL doesn't affect my copy of the URL isn't a thing we care about (and in fact, in languages like Java and Python the "URL", or reference type value, can't be modified at all, only the thing pointed to by it can).

Barbara Liskov, when she invented the CLU programming language (which had these semantics), realized that the existing terms "call by value" and "call by reference" weren't particularly useful for describing the semantics of this new language. So she invented a new term: call by object sharing.

When discussing languages that are technically call by value, but where common types in use are reference or pointer types (that is: nearly every modern imperative, object-oriented, or multi-paradigm programming language), I find it's a lot less confusing to simply avoid talking about call by value or call by reference. Stick to call by object sharing (or simply call by object) and nobody will be confused. :-)