[boost] Re: Template deduction problems on Sun

From: Christopher Currie (christopher_at_[hidden])
Date: 2003-11-21 11:55:40

Peter Dimov wrote:
> That's odd. This is the line that fails:
> 
>     BOOST_TEST( bind(&X::hash, _1)(x) == 23558 );
> 
> and the error message seems to indicate that the overload
> 
> template<class F, class A1>
>     _bi::bind_t<_bi::unspecified, F, typename _bi::list_av_1<A1>::type>
>     BOOST_BIND(F f, A1 a1)
> 
> is (incorrectly) preferred over
> 
> template<class R, class T, class A1>
> _bi::bind_t< R const &, _mfi::dm<R, T>, typename _bi::list_av_1<A1>::type >
>     BOOST_BIND(R T::*f, A1 a1)

That helps. Here's the test case that makes it fail:

struct unspecified {};

template<class R> class bind_t { };

template<class R, class F>
     bind_t<R>
     bind(F f);

template<class F>
     bind_t<unspecified>
     bind(F f);

template<class R, class T>
     bind_t<R const &>
     bind(R T::* pm)
{
   return bind_t<R const &>();
}

struct X { int i; };

int main() { bind(&X::i); }

The Sun compiler is selecting the first partial specialization over the 
second. If I comment out the first specialization, the code compiles 
fine, but for some reason it thinks the first is a better match, so it 
won't link.

Is this something we can work around, or is this something the user will 
just have to deal with? I can specify "bind<int, X>(&X::i)" in main, and 
it will compile.

Christopher