Nice article, but it still doesn't answer the question I've always had about anonymous functions: why are they more "important" (read: why is everyone making such a big deal about them) than pointers-to-functions in C? I've written C/C++ code that pass functions as arguments to other functions by using pointers. I could just as easily write code that returns a pointer to a function. The only difference I see here is that anonymous functions are, well, anonymous.
So someone clarify for me why this is any better. What am I missing?
This isn't necessarily any better, it just has more syntax support. However, it isn't equivalent to a function ptr, it's more equivalent to a structure with a fptr and associated data. A C++ function object or a C fptr pointing to a trampoline fn has exactly this functionality.
However, the functionality isn't the only thing. There are two notational conveniences that typically come along for the ride:
1) You can declare the function nested within another function. This is a necessary prerequisite to...
2) The associated structure is automatically created by the compiler based on what data the function uses in the surrounding context.
3) Just like the struct/fptr pair you might have created manually on the heap, you can return this function from another function with the data it captured coming along for the ride. That's generally were all the fun stuff happens :-)
To clarify this: See the recent "lambda" proposal for C++0x to see that same feature in the context of a Cish language, with the attendant modifications for manual memory management, stack based capture, etc.
The interresting part is that you can define on the spot a single usen, throwaway function to be used as a parameter of a function that accepts functions as arguments.
Another property of lambdas is nested defines with the possibility to return an inner function (be it anonymous or not) from an outer one.
At last their lexical scope allows, in the inner/outer functions described above, to keep the scope of the outer function alive as long as the returned inner function is referenced. This allows to make them work as lightweight objects (or even to define an object system if you want to).
Pointers can break easily. It's entirely possible to add or subtract from a pointer, which drops you in a different memory area. This isn't possible with anonymous functions, making them more safe.
It's the same with references vs. pointers. You can't do any calculations with references, but you can with pointers.
Not much. The other difference is that closures can access the local variables that are in scope at the point where the function is defined. I guess you could hack this together in C by packing and unpacking the variables into a struct together with the function pointer. It's a bit long winded that way though.
In smalltalk and ruby closures are used extensively to define new control structures using the block syntax.
So someone clarify for me why this is any better. What am I missing?