Function pointers, Closures, and Lambda

Question

I am just now learning about function pointers and, as I was reading the K&R chapter on the subject, the first thing that hit me was, \"Hey, this is kinda like a closure

Answer 1

As someone who has written compilers for languages both with and without 'real' closures, I respectfully disagree with some of the answers above. A Lisp, Scheme, ML, or Haskell closure does not create a new function dynamically. Instead it reuses an existing function but does so with new free variables. The collection of free variables is often called the environment, at least by programming-language theorists.

A closure is just an aggregate containing a function and an environment. In the Standard ML of New Jersey compiler, we represented one as a record; one field contained a pointer to the code, and the other fields contained the values of the free variables. The compiler created a new closure (not function) dynamically by allocating a new record containing a pointer to the same code, but with different values for the free variables.

You can simulate all this in C, but it is a pain in the ass. Two techniques are popular:

1. Pass a pointer to the function (the code) and a separate pointer to the free variables, so that the closure is split across two C variables.

2. Pass a pointer to a struct, where the struct contains the values of the free variables and also a pointer to the code.

Technique #1 is ideal when you are trying to simulate some kind of polymorphism in C and you don't want to reveal the type of the environment---you use a void* pointer to represent the environment. For examples, look at Dave Hanson's C Interfaces and Implementations. Technique #2, which more closely resembles what happens in native-code compilers for functional languages, also resembles another familiar technique... C++ objects with virtual member functions. The implementations are almost identical.

This observation led to a wisecrack from Henry Baker:

People in the Algol/Fortran world complained for years that they didn't understand what possible use function closures would have in efficient programming of the future. Then the `object oriented programming' revolution happened, and now everyone programs using function closures, except that they still refuse to to call them that.