Please refer to the code below:
#include <algorithm>
namespace N
{
template <typename T>
class C
{
public:
void SwapWith(C & c)
{
using namespace std; // (1)
//using std::swap; // (2)
swap(a, c.a);
}
private:
int a;
};
template <typename T>
void swap(C<T> & c1, C<T> & c2)
{
c1.SwapWith(c2);
}
}
namespace std
{
template<typename T> void swap(N::C<T> & c1, N::C<T> & c2)
{
c1.SwapWith(c2);
}
}
As written above, the code doesn't compile on Visual Studio 2008/2010. The error is:
'void N::swap(N::C<T> &,N::C<T> &)' : could not deduce template argument for 'N::C<T> &' from 'int'.
However, if I comment out (1) and uncomment (2), it will compile OK. What is the difference between using namespace std
and using std::swap
that explains this behavior?
The first case is a using directive (using namespace X
), and what it means is that the names from namespace X
will be available for regular lookup, in the first common namespace of X
and the current scope. In this case, the first common namespace ancestor of ::N
and ::std
is ::
, so the using directive will make std::swap
available only if lookup hits ::
.
The problem here is that when lookup starts it will look inside the function, then inside the class, then inside N
and it will find ::N::swap
there. Since a potential overload is detected, regular lookup does not continue to the outer namespace ::
. Because ::N::swap
is a function the compiler will do ADL (Argument dependent lookup), but the set of associated namespaces for fundamental types is empty, so that won't bring any other overload. At this point lookup completes, and overload resolution starts. It will try to match the current (single) overload with the call and it will fail to find a way of converting from int
to the argument ::N::C
and you get the error.
On the other hand a using declaration (using std::swap
) provides the declaration of the entity in the current context (in this case inside the function itself). Lookup will find std::swap
immediately and stop regular lookup with ::std::swap
and will use it.