Dynamic dispatching of template functions?

A typical ‘trick’ to bridge compile time and runtime when dealing with templates is visiting a variant type. That’s what the Generic Image Library (available as Boost.GIL or standalone) does for instance. It typically takes the form of:

typedef boost::variant<T, U, V> variant_type;
variant_type variant = /* type is picked at runtime */
boost::apply_visitor(visitor(), variant);

where visitor is a polymorphic functor that simply forwards to the template:

struct visitor: boost::static_visitor<> {
    template<typename T>
    void
    operator()(T const& t) const
    { foo(t); } // the real work is in template<typename T> void foo(T const&);
};

This has the nice design that the list of types that the template will/can be instantiated with (here, the variant_type type synonym) is not coupled to the rest of the code. Metafunctions like boost::make_variant_over also allows computations over the list of types to use.

Since this technique is not available to non-type parameters, you need to ‘unroll’ the visitation by hand, which unfortunately means the code is not as readable/maintainable.

void
bar(int i) {
    switch(i) {
        case 0: A<0>::f(); break;
        case 1: A<1>::f(); break;
        case 2: A<2>::f(); break;

        default:
            // handle
    }
}

The usual way to deal with the repetition in the above switch is to (ab)use the preprocessor. An (untested) example using Boost.Preprocessor:

#ifndef LIMIT
 #define LIMIT 20 // 'reasonable' default if nothing is supplied at build time
#endif
#define PASTE(rep, n, _) case n: A< n >::f(); break;

void
bar(int i) {
    switch(i) {
        BOOST_PP_REPEAT(LIMIT, PASTE, _)

        default:
            // handle
    }
}

#undef PASTE
#undef LIMIT

Better find good, self-documenting names for LIMIT (wouldn’t hurt for PASTE either), and limit the above code-generation to just one site.


Building from David’s solution and your comments:

template<int... Indices>
struct indices {
    typedef indices<Indices..., sizeof...(Indices)> next;
};

template<int N>
struct build_indices {
    typedef typename build_indices<N - 1>::type::next type;
};

template<>
struct build_indices<0> {
    typedef indices<> type;
};

template<int... Indices>
void
bar(int i, indices<Indices...>)
{
    static void (*lookup[])() = { &A<Indices>::f... };
    lookup[i]();
}

then to call bar: bar(i, typename build_indices<N>::type()) where N would be your constant-time constant, sizeof...(something). You can add a layer to hide the ‘ugliness’ of that call:

template<int N>
void
bar(int i)
{ bar(i, typename build_indices<N>::type()); }

which is called as bar<N>(i).

Leave a Comment