Why try functional programming, specifically with FC++?
Some of the advantages that functional programming has over other programming paradigms, such as OOP, are:
 Conciseness of code
 Programming that's free of side effects (no global/static variables manipulated by endless set/get routines)
 Fast prototyping
 FC++ provides a wealth of syntax and library functions that help make the transition smooth for Haskell programmers.
You get around the fact that C++ does not have any functional programming constructs by using libraries. FC++ is the best available open source implementation of a C++ based functional programming library that you can plug in with legacy C++ code. FC++ has been used in projects such as BSFC++, which is a library for functional bulk synchronous parallel programming in C++.
Download and installation
FC++ is available for download from SourceForge (see Resources). Unpacking the installed compressed (.zip) file reveals a collection of header files. Including the prelude.h
header in the user application sources is all you need to do to get started. Listing 1 shows you how to compile sources that use FC++ code. Note that this is a headerdependent installation, and no other libraries are involved.
Listing 1. Compiling sources that use FC++ code
g++ user_source1.cpp –I<path to FC++ installation>
Note: All code in this article was tested using FC++ 1.5 with g++ 3.4.4.
Understanding CFunType
The functional programming paradigm lets functions accept other functions as
arguments. Clearly, the base versions of C/C++ don't allow for such syntax. To
circumvent this problem, FC++ functions are expressed as instances of classes that
follow certain coding conventions, and this is where CFunType
comes into play. C++ function objects are characterized by
the presence of the operator ( )
in the class definition.
Listing 2 below is an example.
Listing 2. Typical use of C++ function objects
struct square { int operator( ) (int x) { return x * x; } }; square sqr1; int result = sqr1(5);
The problem with the implementation in Listing 2 is that, in mathematical terms, the function type for sqr1
is int —> int
, but the C++ type for sqr1
is struct square
. FC++ introduces the template CFunType
, which is used for encoding the type signature information. The last argument in CFunType
is the return type of the function, and the rest are input type information in the same order they appear in the function prototype. Listing 3 shows how square looks using CFunType
.
Listing 3. Using CFunType
to encode function signature for square operation
#include “prelude.h” struct square : public CFunType<int, int> { int operator( ) (int x) { return x * x; } }; square sqr1; int result = sqr1(5);
Listing 4 is another example that inserts an integer into a list and returns the updated list.
Listing 4. Using CFunType
to encode function signature for list manipulation
#include “prelude.h” struct Insert : public CFunType<int,List<int>,List<int> > { List<int> operator()( int x, const List<int>& l ) const { // code for determining where to insert the data goes here }
Note: The List
data type in Listing 4 is a predefined FC++ type described later in this article.
Transforming functions into objects
For functions to accept functions as input arguments, the functions must be somehow transformed into objects. FC++ defines the FunN
category of classes that are built on top of CFunType
and the ptr_to_fun
routine, which actually carries out the transformation. Take a look at Listing 5.
Listing 5. Using ptr_to_fun
to convert function to FC++ function object
int multiply(int m, int n) { return m*n; } Fun2<int, int, int> mult1 = ptr_to_fun (&multiply); int result = mult1(8, 9); // result equals 72
As in CFunType
, signature for Fun2
implies that this object represents a function that accepts two integer inputs and returns an integer. Likewise, you may have Fun3<int, double, double, string>
, which represents a function that accepts one integer, two doubles and returns a string.
List and laziness basics
List manipulation is at the heart of functional programming. FC++ defines its own list data type, which is different from Standard Template Library (STL) list. FC++ lists are lazy. You can create lists with infinite elements in FC++, but they are only evaluated on a need basis. Listing 6 demonstrates what this means.
Listing 6. Defining and using lazy lists
List<int> numbers = enumFrom (33); List<int> even_and_greater_than_33 = filter (even, numbers); assert (take(4, even_and_greater_than_33)) = list_with (34, 36, 38, 40);
The enumFrom
, filter
, even
, take
, and list_with
elements are part of predefined functionality in FC++. In
Listing 6 above, enumFrom
returns an infinite list of numbers starting from 33. The filter
routine returns another infinite list with numbers that are even and greater than 33. Finally, the take
routine actually extracts the first four elements from this list. Clearly, none of the lists store an infinite list of numbers—the evaluation is strictly on a need basis.
Table 1 describes some of the typical functions used with lists in FC++.
Table 1. Functions that are used in conjunction with the FC++
Function  Description 

head(<list>)  Returns the first element of the list 
tail(<list>)  Returns a list with the same elements as <list> other than the first 
cons(<element >, <list>)  Returns a list with <element> added to the head of the list 
NIL  Signifies an empty list 
list_with(<element1, element2>,…, <elementN>)  Creates a list with N elements 
enumFrom(<element1>)  Creates an infinite list starting with element1 
compose(<func1>, <func2>)  Compose (f, g) = f(g(x)), where f(x) and g(x) are two functions 
filter(<func1>, <list>)  Returns a list of elements from <list> filtered using the <func1> function 
take(<N>, <list>)  Returns a list with the first N elements from <list> 
map(<function>, <list>)  Applies the first <function> function to each element of first <list> 
Listing 7 is another example that shows how to create and display the contents of a list.
Listing 7. Creating a list, checking its contents, and displaying data
#include <iostream> #include “prelude.h” int main( ) { int x=1, y=2, z=3; List<int> li = cons(x,cons(y,cons(z,NIL))); // head also removes the 1st element from the list assert( head(li) == 1 ); // tail returns whatever is left of in the list, and list_with is // used to define small sized list assert( tail(li) == list_with(2,3) ); while( li ) { std::cout << li.head() << " "; li = li.tail(); } return 0; }
Note: In the creation of the li
list, the cons
routine adds elements to the front of a list; z, y and x are added in that order to create the final list.
Faster list implementation
FC++ 1.5 provides an additional variant of the List
data
structure called OddList
, which is defined in list.h. OddList
s have exactly the same interface as List
s, but they are faster. All FC++ routines that operate on List
operate on OddList
, too. The efficiency in OddList
is gained by caching the next node in the list. Listing 8 sums up some of the subtler aspects of using OddList
.
Listing 8. Subtler aspects of using OddList
OddList<int> odd1 = enumFrom (1); List<int> list1 = odd1.tail ( ); // always returns List<int>!! OddList<int> odd2 = enumFrom (1); List<int> list2 = odd2.delay ( ); // create a List<int> with same data as odd2 List<int> list3 = enumFrom (1); OddList<int> odd3 = list3.force ( ); // creates an OddList<int> with same data as list3
OddList
s don't have support for STL style iterators that exist for Lists
. See Resources for further detail on OddList
implementation.
Creating your own filters
If you want to create your own filter in Listing 6 (for instance
all numbers that are divisible by 100 and greater than 33), all you need to do is define your own filter function and then call ptr_to_fun
to convert it into a function object. Listing 9 shows you how.
Listing 9. Using CFunType
to encode function signature for list manipulation
bool div_by_100 (int n) { return n % 100 ? false : true; } List<int> num = enumFrom(34); List<int> my_nums = filter( ptr_to_fun(&div_by_100), num);
Note that FC++ List
s and filter
s are completely generic in nature and can accommodate any data type.
Next, look into two fundamental functional techniques: currying and composition.
Currying
Currying is a functional programming technique that binds a subset of some function's arguments to fixed values, thus creating new functions. Listing 10 is an example that curries the f
function.
Listing 10. Using currying to create new functions
int multiply(int m, int n) { return m * n; } Fun2<int, int, int> f2 = ptr_to_fun (&multiply); Fun1<int, int> f1 = curry2 (f2, 9); std::cout << f1(4) << std::endl; // equivalent to multiply(9, 4) Fun1<int, int> f1_implicit = f2(9); std::cout << f1_implicit(4) << std::endl; // same as f1(4)
The predefined curry2
routine binds the first argument of
f2
to 9
. FC++ 1.5 provides
curry1
, curry2
, and curry3
operators that fix the first N
arguments to specific values. Additionally, FC++ also defines the bind routines to create new functions that prefix values to specific arguments of existing functions. For example, bind2and3of3 (f, 8, 9)
is equivalent to f(x, 8, 9)
where f(x, y, z) is a 3input function. Yet another interesting way of specializing arguments is to use an underscore (_
). For instance, greater (_, 10)
is the same as f(x) = (x > 10). Note that greater is predefined in FC++. Listing 11 provides some more examples of currying.
Listing 11. More currying examples
List<int> integers = enumFrom (1); List<int> int_gt_100 = filter(greater(_, 100), integers); // This list will add 3 to all elements of integers. List<int> plus_3 = map (plus(3), integers);
Listing 12 shows a code snippet that displays all the factors of a number, including the number itself.
Listing 12. Displaying all the factors of a number
#include "prelude.h" using namespace fcpp; #include <iostream> using namespace std; bool divisible( int x, int y ) { return x%y==0; } struct Factors : public CFunType<int,OddList<int> > { OddList<int> operator()( int x ) const { return filter( curry2(ptr_to_fun(&divisible),x), enumFromTo(1,x) ); } } factors; int main() { OddList<int> odd = factors(20); while (odd) { cout << head(odd) << endl; odd = tail(odd); } return 0; }
The key to understanding Listing 12 lies in this snippet: return filter( curry2(divisible,x), enumFromTo(1,x) );
. You are creating a filter for the list returned by enumFrom(1, 20)
such that all numbers that perfectly divide 20 form a part of the final list. The curry2
routine binds 20 to the first argument of the divisible
function. Note that ptr_to_fun
makes divisible
a function object that can be passed as an argument to curry2
.
Composition
Functional programming produces new functionality by combining existing code. The compose ( )
operator composes two unary functions, f(x)
and g(x)
, to yield a new function, h(x)
, such that h(x) = f(g(x)). For example, compose (head, tail)
on a list returns the second element in the list. This is functional coding in its proper sense; g(x)
serves as the argument to f(x)
. Listing 13, obtained from "Functional Programming with the FC++ Library" (see Resources), is an example that uses composition.
Listing 13. Using compose
and tail
to obtain the second element of a list
std::string s=”foo”, t=”bar”, u=”qux”; List<std::string> ls = cons(s, cons(t, cons(u, NIL))); ls = compose(tail, tail) (ls); // tail(tail(ls)); assert (head(ls) == ”qux”); // s, t are removed
Listing 14 is another example that increments all elements of a list by two.
Listing 14. Using compose
to increment list elements
List<int> integers = enumFrom (1); map (compose(inc, inc), integers); // this modifies integers to an infinite list [3, 4, 5 ...]
Lambda functions
Any discussion about functional programming is incomplete without a mention of lambda functions. Lambda abstraction is used for defining anonymous functions. This is useful when you don't want to define separate functions for small pieces of code. To use lambda functionality in code, you need to define the FCPP_ENABLE_LAMBDA
macro. Listing 15 succinctly defines new mathematical and logical functions from existing code. Notice how factorial
is defined.
Listing 15. Defining lambda functions
// a new function where f(x) = 3*x+1 lambda(X)[ plus[multiplies[3,X],1] ] // a new function where f(x) = x! (factorial x) lambda(X)[ l_if[equal[X,0],1,multiplies[X,SELF[minus[X,1]]]] ]
The code in Listing 15 is selfexplanatory. Routines plus
, multiplies
, and so on are defined as part of the FC++ library, and you use the lambda
operator to create new functionality from existing code.
Conclusion
FC++ provides:
 Objects of type
CFunType
, which you can easily extend to serve functional programming needs  Implementation of lazy lists that can potentially hold infinite sequences
 Several functional programming operators such as
head
,tail
,map
,filter
,ptr_to_fun
, and so on  The ability to create new functions from existing functions using currying operators,
lambda
, orcompose
Probably the singular drawback of FC++ is the lack of standardized documentation that describes the functions defined in its headers. This article introduced the most useful ones: compose
, curry
, bind
, take
, map
, ptr_to_fun
, and filter
.
Resources
Learn
 Wikipedia provides an interesting introduction for beginners to functional programming.
 Getting started with FC++ provides informal documentation for readers who are familiar with C++ but new to functional programming.
 For a more complete overview, read "Functional Programming with the FC++ Library" by Brian McNamara and Yannis Smaragdakis.

For more detail on
OddList
implementation, see FC++ lazy list implementation on the Georgia College of Tech Computing website.  See Currying in FC++ for more information about currying.
 To learn more about lambda functionality, see FC++ lambda
 Stay current with developerWorks technical events and webcasts focused on a variety of IBM products and IT industry topics.
 Attend a free developerWorks Live! briefing to get uptospeed quickly on IBM products and tools as well as IT industry trends.
 Watch developerWorks ondemand demos ranging from product installation and setup demos for beginners, to advanced functionality for experienced developers.
Get products and technologies
 Evaluate XL C/C++ for AIX
 Download FC++.
 Download the FC++ client for further insight into its workings.
 Evaluate IBM products in the way that suits you best: Download a product trial, try a product online, use a product in a cloud environment, or spend a few hours in the SOA Sandbox learning how to implement Service Oriented Architecture efficiently.
Discuss
 Get involved in the My developerWorks community. Connect with other developerWorks users while exploring the developerdriven blogs, forums, groups, and wikis.
 Follow developerWorks on Twitter.
 Get involved in the My developerWorks community.
 Participate in the AIX and UNIX® forums:
Comments
Dig deeper into AIX and Unix on developerWorks
 Overview
 New to AIX and Unix
 Technical library (articles and more)
 Forums
 Community
 Downloads and products
 Open source projects
 Events

Bluemix Developers Community
Get samples, articles, product docs, and community resources to help build, deploy, and manage your cloud apps.

developerWorks Labs
Experiment with new directions in software development.

DevOps Services
Software development in the cloud. Register today to create a project.

IBM evaluation software
Evaluate IBM software and solutions, and transform challenges into opportunities.