Re: [boost] [range] SinglePassIterator concept check is too strict

Subject: Re: [boost] [range] SinglePassIterator concept check is too strict
From: Dave Abrahams (dave_at_[hidden])
Date: 2012-01-10 04:11:37

on Mon Jan 09 2012, Nathan Ridge <zeratul976-AT-hotmail.com> wrote:

> Hello,
>
> It seems to me the concept checking performed by Boost.Range for the
> SinglePassIterator concept is too strict.
>
> In boost/range/concepts.hpp, SinglePassIteratorConcept checks for the
> following usage (line 157 in trunk):
>
> BOOST_DEDUCED_TYPENAME boost::detail::iterator_traits<Iterator>::reference r1(*i);
>
> This requires that the return type of operator* be convertible to the
> iterator's reference type.
>
> However, looking at the specification for single pass iterators [1], 
> I don't see any requirement that the return type of operator* be
> convertible to the reference type - it just needs to be convertible to
> the value type.
>
> An example of an iterator for which the return type of operator* is
> not convertible to the iterator's reference type is 
> std::istreambuf_iterator<charT>, whose reference type is charT&, but
> the return type of its operator* is charT. (Both of these are specified
> explicitly in section 24.5.3 of the C++03 standard.) 

I agree with your analysis.

> As a result, code such as the following fails to compile with GCC or
> MSVC in C++03 mode:
>
> #include <boost/range/concepts.hpp>
> #include <boost/range/iterator_range.hpp>
> #include <iterator>
> #include <istream>
>
> int main()
> {
> Â Â Â  BOOST_CONCEPT_ASSERT
> Â Â Â  (( 
> Â Â Â Â Â Â Â  boost::SinglePassRangeConcept<
> Â Â Â Â Â Â Â Â Â Â Â  boost::iterator_range<
> Â Â Â Â Â Â Â Â Â Â Â Â Â Â Â  std::istreambuf_iterator<char> > >
> Â Â Â  ));
> }
>
> (In C++11, things changed - a requirement was added that the return type
>
> of operator* be the same as the iterator's reference type, and
>
> correspondingly std::istreambuf_iterator<charT>'s reference type was
>
> changed to be charT).
>
> What can be done about this?

The code could be fixed?  Submit a patch I suppose.

-- 
Dave Abrahams
BoostPro Computing
http://www.boostpro.com