The random access tag is in my opinion a strongest point of the iterator, you can get a performance boost in many applications (specially when implementing containers). I'm out of ideas about how to solve this. Can't we live breaking the law? Should we create new categories? Just let it as a "poor" input iterator?
We already have new categories. It can report that it has random access traversal. It's just not allowed to have random_access_iterator_tag. See the "new iterator categories" document.
Sorry, I really wanted to say "random access traversal" instead of random access iterator. Taking a snippet of the proposed implementation: template <class T, class Difference = std::ptrdiff_t> class range_from_ref_iterator : public boost::iterator_facade < range_from_ref_iterator<T, Difference> , T , boost::random_access_traversal_tag , T & , Difference> Correct me if I'm wrong but what I want is to simulate a source of N identical objects without modifying them, so I think this is a Readable Iterator and a Random Access Traversal Iterator. I would now propose something like: template <class T, class Difference = std::ptrdiff_t> class repeat_read_iterator : public boost::iterator_facade < repeat_read_iterator<T, Difference> , const T , boost::random_access_traversal_tag , const T & , Difference> so dereferencing "repeat_read_iterator<T>" would return const T & with random_access_traversal capability to be able move fast inside that "virtual read-only array". Am I missing something? Do you see this read only repeated sequence simulation useful to include it the iterator library? Regards, Ion