Sebastian Redl wrote:
Eric Niebler wrote:
That only works on compilers that handle SFINAE. Probably the easiest thing would be to remove the char* overloads and modify the "primary" boost_range_end(T const &) overload, which currently only works when T is an STL container, to use is_pointer<T> to dispatch to str_end().
What if the non-template overload would be replaced by a template overload?
template<> const char* boost_range_end( const char* s )
Wouldn't that work? I know of a similar (but much less complex) case where this solved the overload resolution problem.
Did you try it? What you've shown is a function template specialization, not an overload, but it won't work, because it doesn't match any primary template. You could've written const char*&, but for completeness, you'd also have to provide overloads for char*&, char*const&, const char*const&, and all the wchar_t variants. That's a lot. Now that I think about it, we had discussed eliminating the inconsistent "boost::size(int[5])==5, but boost::size(char[5])==4" behavior. Bugs aside, why is it still there? Thorsten? -- Eric Niebler Boost Consulting www.boost-consulting.com