Why do you need to invoke an anonymous function on the same line?

Drop the semicolon after the function definition.

(function (msg){alert(msg)})
('SO');

Above should work.

DEMO Page: https://jsfiddle.net/e7ooeq6m/

I have discussed this kind of pattern in this post:

jQuery and $ questions

EDIT:

If you look at ECMA script specification, there are 3 ways you can define a function. (Page 98, Section 13 Function Definition)

1. Using Function constructor

var sum = new Function('a','b', 'return a + b;');
alert(sum(10, 20)); //alerts 30

2. Using Function declaration.

function sum(a, b)
{
    return a + b;
}

alert(sum(10, 10)); //Alerts 20;

3. Function Expression

var sum = function(a, b) { return a + b; }

alert(sum(5, 5)); // alerts 10

So you may ask, what's the difference between declaration and expression?

From ECMA Script specification:

FunctionDeclaration : function Identifier ( FormalParameterListopt ){ FunctionBody }

FunctionExpression : function Identifieropt ( FormalParameterListopt ){ FunctionBody }

If you notice, 'identifier' is optional for function expression. And when you don't give an identifier, you create an anonymous function. It doesn't mean that you can't specify an identifier.

This means following is valid.

var sum = function mySum(a, b) { return a + b; }

Important point to note is that you can use 'mySum' only inside the mySum function body, not outside. See following example:

var test1 = function test2() { alert(typeof test2); }

alert(typeof(test2)); //alerts 'undefined', surprise! 

test1(); //alerts 'function' because test2 is a function.

Live Demo

Compare this to

 function test1() { alert(typeof test1) };

 alert(typeof test1); //alerts 'function'

 test1(); //alerts 'function'

Armed with this knowledge, let's try to analyze your code.

When you have code like,

    function(msg) { alert(msg); }

You created a function expression. And you can execute this function expression by wrapping it inside parenthesis.

    (function(msg) { alert(msg); })('SO'); //alerts SO.

It's called a self-invoked function.

What you are doing when you call (function(){}) is returning a function object. When you append () to it, it is invoked and anything in the body is executed. The ; denotes the end of the statement, that's why the 2nd invocation fails.

An anonymous function is not a function with the name "". It is simply a function without a name.

Like any other value in JavaScript, a function does not need a name to be created. Though it is far more useful to actually bind it to a name just like any other value.

But like any other value, you sometimes want to use it without binding it to a name. That's the self-invoking pattern.

Here is a function and a number, not bound, they do nothing and can never be used:

function(){ alert("plop"); }
2;

So we have to store them in a variable to be able to use them, just like any other value:

var f = function(){ alert("plop"); }
var n = 2;

You can also use syntatic sugar to bind the function to a variable:

function f(){ alert("plop"); }
var n = 2;

But if naming them is not required and would lead to more confusion and less readability, you could just use them right away.

(function(){ alert("plop"); })(); // will display "plop"
alert(2 + 3); // will display 5

Here, my function and my numbers are not bound to a variable, but they can still be used.

Said like this, it looks like self-invoking function have no real value. But you have to keep in mind that JavaScript scope delimiter is the function and not the block ({}).

So a self-invoking function actually has the same meaning as a C++, C# or Java block. Which means that variable created inside will not "leak" outside the scope. This is very useful in JavaScript in order not to pollute the global scope.

It's just how JavaScript works. You can declare a named function:

function foo(msg){
   alert(msg);
}

And call it:

foo("Hi!");

Or, you can declare an anonymous function:

var foo = function (msg) {
    alert(msg);
}

And call that:

foo("Hi!");

Or, you can just never bind the function to a name:

(function(msg){
   alert(msg);
 })("Hi!");

Functions can also return functions:

function make_foo() {
    return function(msg){ alert(msg) };
}

(make_foo())("Hi!");

It's worth nothing that any variables defined with "var" in the body of make_foo will be closed over by each function returned by make_foo. This is a closure, and it means that the any change made to the value by one function will be visible by another.

This lets you encapsulate information, if you desire:

function make_greeter(msg){
    return function() { alert(msg) };
}

var hello = make_greeter("Hello!");

hello();

It's just how nearly every programming language but Java works.

The code you show,

(function (msg){alert(msg)});
('SO');

consist of two statements. The first is an expression which yields a function object (which will then be garbage collected because it is not saved). The second is an expression which yields a string. To apply the function to the string, you either need to pass the string as an argument to the function when it is created (which you also show above), or you will need to actually store the function in a variable, so that you can apply it at a later time, at your leisure. Like so:

var f = (function (msg){alert(msg)});
f('SO');

Note that by storing an anonymous function (a lambda function) in a variable, your are effectively giving it a name. Hence you may just as well define a regular function:

function f(msg) {alert(msg)};
f('SO');

This answer is not strictly related to the question, but you might be interested to find out that this kind of syntax feature is not particular to functions. For example, we can always do something like this:

alert(
    {foo: "I am foo", bar: "I am bar"}.foo
); // alerts "I am foo"

Related to functions. As they are objects, which inherit from Function.prototype, we can do things like:

Function.prototype.foo = function () {
    return function () {
        alert("foo");
    };
};

var bar = (function () {}).foo();

bar(); // alerts foo

And you know, we don't even have to surround functions with parenthesis in order to execute them. Anyway, as long as we try to assign the result to a variable.

var x = function () {} (); // this function is executed but does nothing

function () {} (); // syntax error

One other thing you may do with functions, as soon as you declare them, is to invoke the new operator over them and obtain an object. The following are equivalent:

var obj = new function () {
    this.foo = "bar";
};

var obj = {
    foo : "bar"
};

When you did:

(function (msg){alert(msg)});
('SO');

You ended the function before ('SO') because of the semicolon. If you just write:

(function (msg){alert(msg)})
('SO');

It will work.

Working example: http://jsfiddle.net/oliverni/dbVjg/

The IIFE simply compartmentalizes the function and hides the msg variable so as to not "pollute" the global namespace. In reality, just keep it simple and do like below unless you are building a billion dollar website.

var msg = "later dude";
window.onunload = function(msg){
  alert( msg );
};

You could namespace your msg property using a Revealing Module Pattern like:

var myScript = (function() {
    var pub = {};
    //myscript.msg
    pub.msg = "later dude";
    window.onunload = function(msg) {
        alert(msg);
    };
    //API
    return pub;
}());