https://gcc.gnu.org/bugzilla/show_bug.cgi?id=103250
--- Comment #2 from Jonathan Wakely <redi at gcc dot gnu.org> ---
I don't think this is a bug. You've added an overload and it's not more
specialized than the function defined by the standard library:
operator/ (const valarray<T>&, const valarray<T>&)
The compiler is correct to reject it as ambiguous.
The unpreprocessed code is:
#include <valarray>
namespace qaz
{
namespace traits
{
template < typename T, typename Q1, typename Q2 > struct div_type { typedef
div_type type; };
template < typename T, typename Q > struct div_type<T,Q,Q> { typedef T
type; };
}
template < typename T, typename Q1, typename Q2 > using div_type = typename
traits::div_type<T,Q1,Q2>::type;
template < typename T, typename Q > struct num { };
template < typename T, typename Q1, typename Q2 > auto operator/( const
num<T,Q1> &, const num<T,Q2> & ) -> div_type<T,Q1,Q2>;
template < typename T, typename Q1, typename Q2 > auto operator/( const
std::valarray< num<T,Q1> > &x, const std::valarray< num<T,Q2> > &y ) ->
std::valarray< decltype(x[0]/y[0]) >;
}
template < typename T, typename Q >
auto f( size_t n )
{
std::valarray< qaz::num<T,Q> > x(n);
return x/x;
}
struct tag {};
template auto f< int, tag >( size_t );
You can fix it by adding another overload which will be more specialized:
template < typename T, typename Q1 > auto operator/( const std::valarray<
num<T,Q1> > &x, const std::valarray< num<T,Q1> > &y ) -> std::valarray<
decltype(x[0]/y[0]) >;