on Mon Dec 10 2007, shunsuke <pstade.mb-AT-gmail.com> wrote:

Hi,

Assume a base iterator is an InputIterator with RandomAccessTraversalTag. boost::make_zip_iterator often returns such iterators.

boost::transform_iterator sometimes resurrects lvalue-ness of its base iterator. For example,

char & to_lvalue(boost::tuple<char &, int &> const &t) { return boost::get<0>(t); }

can return a lvalue from zip_iterator dereference.

Therefore, if boost::transform_iterator extracts "pure" traversal tag from its base iterator, it can be not an InputIterator but a RandomAccessIterator. The patch will be something like this:

template <class UnaryFunc, class Iterator, class Reference, class Value> struct transform_iterator_base { // ...

typedef iterator_adaptor< transform_iterator<UnaryFunc, Iterator, Reference, Value> , Iterator , cv_value_type - , use_default // Leave the traversal category alone + , typename pure_traversal_tag<typename iterator_traversal<Iterator>::type>::type , reference > type; }; }

Is there any pitfalls?

Looks like a good idea. Did you check to see if it breaks any tests? -- Dave Abrahams Boost Consulting http://www.boost-consulting.com