In my last posting about C++ Multiple Dispatch I wondered if it was really any different than function overloading. I now appreciate that it is something that needs to occur at runtime, not compile time.
With a little help from the Boost libraries, I threw together this example of how one could do Multiple Dispatch in C++. Note that in this version the first matching function is called, which means the generic “Thing, Thing” version must be registered last to get the appropriate behavior.
Also, note that this template version is reusable and could actually be put to use for something. It is, however, limited to only 2 parameters, but it would be fairly trivial to increase/decrease the number. Most of this code is reusable, I put a comment where the user bits begin.
#include <iostream>
#include <stdexcept>
#include <vector>
#include <boost/function.hpp>
#include <boost/shared_ptr.hpp>
template<typename ReturnType, typename T1, typename T2>
class Dispatcher{
public:
struct UnhandledDispatch : std::runtime_error
{
UnhandledDispatch()
: std::runtime_error("Uknown dispatch for given arguments")
{
}
};
template<typename I1, typename I2>
void addHandler(const boost::function<ReturnType (const I1 &, const I2 &)> &h)
{
myHandlers.push_back(boost::shared_ptr<Handler>(new HandlerImpl<I1, I2>(h)));
}
template<typename I1, typename I2>
void addHandler(ReturnType (*fp)(const I1&, const I2&))
{
myHandlers.push_back(
boost::shared_ptr<Handler>(
new HandlerImpl<I1, I2>(boost::function<ReturnType (const I1 &, const I2 &)>(fp))));
}
ReturnType operator()(const T1 &t1, const T2 &t2)
{
for (typename std::vector<boost::shared_ptr<Handler> >::iterator itr = myHandlers.begin();
itr != myHandlers.end();
++itr)
{
try {
return (*itr)->go(t1, t2);
} catch (std::bad_cast &) {
//The current function did not match!
//keep truckin and try the next one
}
}
throw UnhandledDispatch();
}
private:
struct Handler
{
virtual ReturnType go(const T1 &, const T2 &) = 0;
};
template<typename I1, typename I2>
struct HandlerImpl : Handler
{
HandlerImpl(const boost::function<ReturnType (I1, I2)> &f)
: myFunction(f)
{
}
virtual ReturnType go(const T1 &t1, const T2 &t2)
{
// this only succeeds if the dynamic_cast succeeds. Otherwise bad_cast is thrown
// and caught by Dispatcher::operator()
return myFunction(dynamic_cast<const I1 &>(t1), dynamic_cast<const I2 &>(t2));
}
private:
boost::function<ReturnType (I1, I2)> myFunction;
};
std::vector<boost::shared_ptr<Handler> > myHandlers;
};
// Everything above this line is implementation details, everything below is what the user
// of the dispatcher needs to know.
struct Thing {
virtual ~Thing() { }
};
struct Rock : Thing {};
struct Paper : Thing {};
struct Scissors : Thing {};
bool defeats(const Thing &, const Thing &) { return false; }
bool defeats(const Paper &, const Rock &) { return true; }
bool defeats(const Rock &, const Scissors &) { return true; }
bool defeats(const Scissors &, const Paper &) { return true; }
int main(int, char **){
Dispatcher<bool, Thing, Thing> myDispatcher;
// the dispatcher returns a bool and takes two "Thing" classes
myDispatcher.addHandler<Paper, Rock>(&defeats); //Add paper, rock version
myDispatcher.addHandler<Rock, Scissors>(&defeats); // add rock, scissors version
myDispatcher.addHandler<Scissors, Paper>(&defeats); // add scissors, paper version
myDispatcher.addHandler<Thing, Thing>(&defeats); // add generic version
Paper paper;
Rock rock;
Thing &thing1(paper);
Thing &thing2(rock);
std::cout << std::boolalpha; std::cout << myDispatcher(thing1, thing2) << std::endl; // decide (at runtime) which version to call
std::cout << myDispatcher(rock, paper) << std::endl;
};