Boost C++ Libraries

...one of the most highly regarded and expertly designed C++ library projects in the world. Herb Sutter and Andrei Alexandrescu, C++ Coding Standards

Click here to view the latest version of this page.
PrevUpHomeNext

Quick Start

I assume the reader is already familiar with tuples (Boost.Tuple) and its ancestor std::pair. The tuple is a generalization of std::pair for multiple heterogeneous elements (triples, quadruples, etc.). The tuple is more or less a synonym for fusion's vector.

For starters, we shall include all of Fusion's Sequence(s) [2]:

#include <boost/fusion/sequence.hpp>
#include <boost/fusion/include/sequence.hpp>

Let's begin with a vector [3]:

vector<int, char, std::string> stuff(1, 'x', "howdy");
int i = at_c<0>(stuff);
char ch = at_c<1>(stuff);
std::string s = at_c<2>(stuff);

Just replace tuple for vector and get for at_c and this is exactly like Boost.Tuple. Actually, either names can be used interchangeably. Yet, the similarity ends there. You can do a lot more with Fusion vector or tuple. Let's see some examples.

Print the vector as XML

First, let's include the algorithms:

#include <boost/fusion/algorithm.hpp>
#include <boost/fusion/include/algorithm.hpp>

Now, let's write a function object that prints XML of the form <type>data</type> for each member in the tuple.

struct print_xml
{
    template <typename T>
    void operator()(T const& x) const
    {
        std::cout
            << '<' << typeid(x).name() << '>'
            << x
            << "</" << typeid(x).name() << '>'
            ;
    }
};

Now, finally:

for_each(stuff, print_xml());

That's it! for_each is a fusion algorithm. It is a generic algorithm similar to STL's. It iterates over the sequence and calls a user supplied function. In our case, it calls print_xml's operator() for each element in stuff.

[Caution] Caution

The result of typeid(x).name() is platform specific. The code here is just for exposition. Of course you already know that :-)

for_each is generic. With print_xml, you can use it to print just about any Fusion Sequence.

Print only pointers

Let's get a little cleverer. Say we wish to write a generic function that takes in an arbitrary sequence and XML prints only those elements which are pointers. Ah, easy. First, let's include the is_pointer boost type trait:

#include <boost/type_traits/is_pointer.hpp>

Then, simply:

template <typename Sequence>
void xml_print_pointers(Sequence const& seq)
{
    for_each(filter_if<boost::is_pointer<_> >(seq), print_xml());
}

filter_if is another Fusion algorithm. It returns a filter_view, a conforming Fusion sequence. This view reflects only those elements that pass the given predicate. In this case, the predicate is boost::is_pointer<_>. This "filtered view" is then passed to the for_each algorithm, which then prints the "filtered view" as XML.

Easy, right?

Associative tuples

Ok, moving on...

Apart from vector, fusion has a couple of other sequence types to choose from. Each sequence has its own characteristics. We have list, set, map, plus a multitude of views that provide various ways to present the sequences.

Fusion's map associate types with elements. It can be used as a cleverer replacement of the struct. Example:

namespace fields
{
    struct name;
    struct age;
}

typedef map<
    fusion::pair<fields::name, std::string>
  , fusion::pair<fields::age, int> >
person;

map is an associative sequence. Its elements are Fusion pairs which differ somewhat from std::pair. Fusion pairs only contain one member, with the type of their second template parameter. The first type parameter of the pair is used as an index to the associated element in the sequence. For example, given a a_person of type, person, you can do:

using namespace fields;
std::string person_name = at_key<name>(a_person);
int person_age = at_key<age>(a_person);

Why go through all this trouble, you say? Well, for one, unlike the struct, we are dealing with a generic data structure. There are a multitude of facilities available at your disposal provided out of the box with fusion or written by others. With these facilities, introspection comes for free, for example. We can write one serialization function (well, two, if you consider loading and saving) that will work for all your fusion maps. Example:

struct saver
{
    template <typename Pair>
    void operator()(Pair const& data) const
    {
        some_archive << data.second;
    }
};

template <typename Stuff>
void save(Stuff const& stuff)
{
    for_each(stuff, saver());
}

The save function is generic and will work for all types of stuff regardless if it is a person, a dog or a whole alternate_universe.

Tip of the Iceberg

And... we've barely scratched the surface! You can compose and expand the data structures, remove elements from the structures, find specific data types, query the elements, filter out types for inspection, transform data structures, etc. What you've seen is just the tip of the iceberg.



[2] There are finer grained header files available if you wish to have more control over which components to include (see section Organization for details).

[3] Unless otherwise noted, components are in namespace boost::fusion. For the sake of simplicity, code in this quick start implies using directives for the fusion components we will be using.


PrevUpHomeNext