[Boost-users] Re: MPL iterator category

From: David Abrahams (dave_at_[hidden])
Date: 2005-02-21 10:40:11

Aleksey Gurtovoy <agurtovoy_at_[hidden]> writes:

> Writing an algorithm for _bidirectional_ iterators and making sure
> that it gets selected for random-access ones as well is somewhat more
> tricky:
>
>     template< typename Iterator > struct algorithm
>         : if_c<
>               iterator_category<Iterator>::type::value 
>                   >= bidirectional_iterator_tag::value
>             , bidirectional_algorithm<Iterator>
>             , forward_algorithm<Iterator>
>             >
>     {
>     };
>
> but try to write this up assuming that _there is_ an implicit
> conversion between 'random_access_iterator_tag' and
> 'bidirectional_iterator_tag', and you'll see that you'll end up with
> something at least as complex as the above.

If there was a _form_ relationship among MPL iterator tags just like
the _inheritance_ relationship among the STL ones, it would be easier
to do this with partial specialization.  E.g.,

  typedef tag<> forward_iterator_tag;
  typedef tag<tag<> > bidirectional_iterator_tag;
  typedef tag<tag<tag<> > > random_access_iterator_tag;

Unfortunately, I don't know how to make this look right.  A
specialization on:

  template <class T, class Category>
  struct algorithm<
        T
      , tag<tag<Category> >     // Bidirectional
  >
  {
    ...
  };

doesn't really communicate well.  Maybe some built-in SFINAE
facilities would help?

  template <class Iterator, class Enable = void>
  struct algorithm;

  template<class Iterator> 
  struct algorithm<Iterator, typename bidirectional_iterator_tag::enable_if<Iterator>::type>
  {
    ...
  };

-- 
Dave Abrahams
Boost Consulting
www.boost-consulting.com