I have detected that the standard declares "distance" template at 24.3.4, as taking two "InputIterator" as parameters.
IMHO, the parameters shoudl be "const InputIterator" as they should not be changed by that function. In fact, the version I have seen of STLPORT does implement it like that ("const InputIterator") violating the standard, while DINKUMWARE respects the standard defintion.
It makes no difference: the function distance takes it's parameters by value so the two forms: template <class InputIterator> typename iterator_traits<InputIterator>::difference_type distance(InputIterator first, InputIterator last); and template <class InputIterator> typename iterator_traits<InputIterator>::difference_type distance(InputIterator const first, InputIterator const last); are actually exactly the same function (the const declaration only affects the mutability of the arguments within the scope of the function, they have no effect on the function's type).
This has given me some problems when trying to use BOOST libraries with BDS 2006 (which comes with Dinkumware library).
The function "size" in boost::iterator_range is declared "const", so that when it calls "std::distance", it will do so with "const IteratorT" parameters. With STLPORT, there will be no problem, as the undelying iterato_traits do not change.
But in the case of DINKUMWARE, the underlying iterator_traits change. This is specially evident when IteratorT is [const] char *, which is a valid random access iterator. "size" will call "std::distance" with "[const] char * const" parameters, which are non-valid random iterators.
That's unfortunately a compiler bug: the compiler is deducing the template parameters for distance as "T const" rather than "T", I've worked around this in the past by explicitly casting away the constantness of the arguments to distance, but it's a pain to get right all over the place, just to keep one broken compiler happy.... John.