Is it possible to iterate an mpl::vector at run time without instantiating the types in the vector?

Question

Generally, I would use boost::mpl::for_each<>() to traverse a boost::mpl::vector, but this requires a functor with a template function declared like the following:

template<typename T> void operator()(T&){T::staticCall();}

My problem with this is that I don't want the object T to be instantiated by for_each<>. I don't need the T parameter in the operator() at all. Is there a way to accomplish this, or an alternative to for_each<> that doesn't pass an object of type T to the template function?

Optimally, I would like the operator() definition to look like this:

template<typename T> void operator()(){T::staticCall();}

And of course, I don't want T to be instantiated at all prior to the call. Any other tips/suggestions are also welcome.

Answer 1

Just encountered the same situation and provided different solution to the problem which I would like to share. It's not as that obvious, but it uses an existing algorithm. The idea is to use pointers instead.

typedef boost::mpl::vector<type1*, type2*> container;

struct functor
{
    template<typename T> void operator()(T*)
    {
        std::cout << "created " << typeid(T).name() << std::endl;
    }
};

int main()
{
    boost::mpl::for_each<container>(functor());
}

so here we get a null pointers, but we don't care as we are not going to use them.

As I stated before that is not obvious in the code and would probably require some additional comments, but it still solves the question without writing any additional code.

added

I think Diego Sevilla suggested something similar.

Answer 2

Interesting question! As far as I can tell, Boost.MPL does not seem to provide such an algorithm. However, writing your own should not be too difficult using iterators.

Here is a possible solution:

#include <boost/mpl/begin_end.hpp>
#include <boost/mpl/next_prior.hpp>
#include <boost/mpl/vector.hpp>

using namespace boost::mpl;


namespace detail {

template < typename Begin, typename End, typename F >
struct static_for_each
{
    static void call( )
    {
        typedef typename Begin::type currentType;

        F::template call< currentType >();
        static_for_each< typename next< Begin >::type, End, F >::call();
    }
};


template < typename End, typename F >
struct static_for_each< End, End, F >
{
    static void call( )
    {
    }
};

} // namespace detail


template < typename Sequence, typename F >
void static_for_each( )
{
    typedef typename begin< Sequence >::type begin;
    typedef typename end< Sequence >::type   end;

    detail::static_for_each< begin, end, F >::call();
}

[The naming may not be very well chosen, but well...]

Here is how you would use this algorithm:

struct Foo
{
    static void staticMemberFunction( )
    {
        std::cout << "Foo";
    }
};


struct Bar
{
    static void staticMemberFunction( )
    {
        std::cout << "Bar";
    }
};


struct CallStaticMemberFunction
{
    template < typename T >
    static void call()
    {
        T::staticMemberFunction();
    }
};


int main()
{
    typedef vector< Foo, Bar > sequence;

    static_for_each< sequence, CallStaticMemberFunction >(); // prints "FooBar"
}