Re: [boost] [Fusion] begin/end ADL issues with C++0x range-based for

Subject: Re: [boost] [Fusion] begin/end ADL issues with C++0x range-based for
From: Jeffrey Lee Hellrung, Jr. (jhellrung_at_[hidden])
Date: 2010-12-20 12:53:06

On 12/20/2010 9:21 AM, Christopher Schmidt wrote:
[...]
> Unfortunately this still leaves us with ADL ambiguity in some cases:
>
> https://svn.boost.org/trac/boost/ticket/4028#comment:6
>
> -Christopher

That boost::fusion::begin/end could be found via ADL in range-based for 
loops is a bit worrying, and, as mentioned in the ticket, SFINAE'ing out 
boost::fusion::begin/end for only Boost.Fusion sequences is only a 
partial solution, as it doesn't help for structures having both range 
and Boost.Fusion interfaces.

I don't know if I fully understand the problem, and I don't know if the 
following is a viable solution or what all the implications are (yes, 
that's a lot of qualifications), but maybe we could put the Boost.Fusion 
data structures (and anything else that would cause boost::fusion to be 
included in the set of "associated namespaces" when using ADL for 
unqualified begin/end) into another namespace, and bring them into the 
boost::fusion namespace with using declarations.  Then maybe 
boost::fusion::begin/end wouldn't be found via ADL, but one would still 
refer to them as boost::fusion::vector, etc.  E.g., the following 
compiles on MSVC9:

namespace fusion_data_structures
{

struct X { };

} // namespace fusion_data_structures

namespace fusion
{

template< class T > void f(T) { }

using fusion_data_structures::X;

} // namespace fusion

template< class T > void f(T) { }

int main(int argc, char* argv[])
{
     fusion::X x; f(x);
     return 0;
}

The fusion_data_structures namespace would be an implementation-detail, 
never to be referred to outside of the Boost.Fusion library.

Just a thought,

- Jeff