Zip Iterator

Author: David Abrahams, Thomas Becker
Contact: dave@boost-consulting.com, thomas@styleadvisor.com
Organization: Boost Consulting, Zephyr Associates, Inc.
Date: 2004-11-01
Copyright: Copyright David Abrahams and Thomas Becker 2003.
abstract:The zip iterator provides the ability to parallel-iterate over several controlled sequences simultaneously. A zip iterator is constructed from a tuple of iterators. Moving the zip iterator moves all the iterators in parallel. Dereferencing the zip iterator returns a tuple that contains the results of dereferencing the individual iterators.

Table of Contents

zip_iterator synopsis

template<typename IteratorTuple>
class zip_iterator
{  

public:
  typedef /* see below */ reference;
  typedef reference value_type;
  typedef value_type* pointer;
  typedef /* see below */ difference_type;
  typedef /* see below */ iterator_category;

  zip_iterator();
  zip_iterator(IteratorTuple iterator_tuple);

  template<typename OtherIteratorTuple>
  zip_iterator(
        const zip_iterator<OtherIteratorTuple>& other
      , typename enable_if_convertible<
              OtherIteratorTuple
            , IteratorTuple>::type* = 0     // exposition only
  );

  const IteratorTuple& get_iterator_tuple() const;

private:
  IteratorTuple m_iterator_tuple;     // exposition only
};

template<typename IteratorTuple> 
zip_iterator<IteratorTuple> 
make_zip_iterator(IteratorTuple t);

The reference member of zip_iterator is the type of the tuple made of the reference types of the iterator types in the IteratorTuple argument.

The difference_type member of zip_iterator is the difference_type of the first of the iterator types in the IteratorTuple argument.

The iterator_category member of zip_iterator is convertible to the minimum of the traversal categories of the iterator types in the IteratorTuple argument. For example, if the zip_iterator holds only vector iterators, then iterator_category is convertible to boost::random_access_traversal_tag. If you add a list iterator, then iterator_category will be convertible to boost::bidirectional_traversal_tag, but no longer to boost::random_access_traversal_tag.

zip_iterator requirements

All iterator types in the argument IteratorTuple shall model Readable Iterator.

zip_iterator models

The resulting zip_iterator models Readable Iterator.

The fact that the zip_iterator models only Readable Iterator does not prevent you from modifying the values that the individual iterators point to. The tuple returned by the zip_iterator's operator* is a tuple constructed from the reference types of the individual iterators, not their value types. For example, if zip_it is a zip_iterator whose first member iterator is an std::vector<double>::iterator, then the following line will modify the value which the first member iterator of zip_it currently points to:

zip_it->get<0>() = 42.0;

Consider the set of standard traversal concepts obtained by taking the most refined standard traversal concept modeled by each individual iterator type in the IteratorTuple argument.The zip_iterator models the least refined standard traversal concept in this set.

zip_iterator<IteratorTuple1> is interoperable with zip_iterator<IteratorTuple2> if and only if IteratorTuple1 is interoperable with IteratorTuple2.

zip_iterator operations

In addition to the operations required by the concepts modeled by zip_iterator, zip_iterator provides the following operations.

zip_iterator();

Returns:An instance of zip_iterator with m_iterator_tuple default constructed.

zip_iterator(IteratorTuple iterator_tuple);

Returns:An instance of zip_iterator with m_iterator_tuple initialized to iterator_tuple.
template<typename OtherIteratorTuple>
zip_iterator(
      const zip_iterator<OtherIteratorTuple>& other
    , typename enable_if_convertible<
            OtherIteratorTuple
          , IteratorTuple>::type* = 0     // exposition only
);
Returns:An instance of zip_iterator that is a copy of other.
Requires:OtherIteratorTuple is implicitly convertible to IteratorTuple.

const IteratorTuple& get_iterator_tuple() const;

Returns:m_iterator_tuple

reference operator*() const;

Returns:A tuple consisting of the results of dereferencing all iterators in m_iterator_tuple.

zip_iterator& operator++();

Effects:Increments each iterator in m_iterator_tuple.
Returns:*this

zip_iterator& operator--();

Effects:Decrements each iterator in m_iterator_tuple.
Returns:*this
template<typename IteratorTuple> 
zip_iterator<IteratorTuple> 
make_zip_iterator(IteratorTuple t);
Returns:An instance of zip_iterator<IteratorTuple> with m_iterator_tuple initialized to t.
template<typename IteratorTuple> 
zip_iterator<IteratorTuple> 
make_zip_iterator(IteratorTuple t);
Returns:An instance of zip_iterator<IteratorTuple> with m_iterator_tuple initialized to t.

Examples

There are two main types of applications of the zip_iterator. The first one concerns runtime efficiency: If one has several controlled sequences of the same length that must be somehow processed, e.g., with the for_each algorithm, then it is more efficient to perform just one parallel-iteration rather than several individual iterations. For an example, assume that vect_of_doubles and vect_of_ints are two vectors of equal length containing doubles and ints, respectively, and consider the following two iterations:

std::vector<double>::const_iterator beg1 = vect_of_doubles.begin();
std::vector<double>::const_iterator end1 = vect_of_doubles.end();
std::vector<int>::const_iterator beg2 = vect_of_ints.begin();
std::vector<int>::const_iterator end2 = vect_of_ints.end();

std::for_each(beg1, end1, func_0());
std::for_each(beg2, end2, func_1());

These two iterations can now be replaced with a single one as follows:

std::for_each(
  boost::make_zip_iterator(
    boost::make_tuple(beg1, beg2)
    ),
  boost::make_zip_iterator(
    boost::make_tuple(end1, end2)
    ),
  zip_func()
  );

A non-generic implementation of zip_func could look as follows:

struct zip_func : 
  public std::unary_function<const boost::tuple<const double&, const int&>&, void>
{
  void operator()(const boost::tuple<const double&, const int&>& t) const
  {
    m_f0(t.get<0>());
    m_f1(t.get<1>());
  }

private:
  func_0 m_f0;
  func_1 m_f1;
};

The second important application of the zip_iterator is as a building block to make combining iterators. A combining iterator is an iterator that parallel-iterates over several controlled sequences and, upon dereferencing, returns the result of applying a functor to the values of the sequences at the respective positions. This can now be achieved by using the zip_iterator in conjunction with the transform_iterator.

Suppose, for example, that you have two vectors of doubles, say vect_1 and vect_2, and you need to expose to a client a controlled sequence containing the products of the elements of vect_1 and vect_2. Rather than placing these products in a third vector, you can use a combining iterator that calculates the products on the fly. Let us assume that tuple_multiplies is a functor that works like std::multiplies, except that it takes its two arguments packaged in a tuple. Then the two iterators it_begin and it_end defined below delimit a controlled sequence containing the products of the elements of vect_1 and vect_2:

typedef boost::tuple<
  std::vector<double>::const_iterator,
  std::vector<double>::const_iterator
  > the_iterator_tuple;

typedef boost::zip_iterator<
  the_iterator_tuple
  > the_zip_iterator;

typedef boost::transform_iterator<
  tuple_multiplies<double>,
  the_zip_iterator
  > the_transform_iterator;

the_transform_iterator it_begin(
  the_zip_iterator(
    the_iterator_tuple(
      vect_1.begin(),
      vect_2.begin()
      )
    ),
  tuple_multiplies<double>()
  );

the_transform_iterator it_end(
  the_zip_iterator(
    the_iterator_tuple(
      vect_1.end(),
      vect_2.end()
      )
    ),
  tuple_multiplies<double>()
  );