Question

I’m trying to write a message processing thingy. The main components are a receiver, a processor, a filter, a translator and a sender. I would like to have a single point of failure at the “process” function of the processor component, for catching exceptions, logging and metrics.

Directly below is the console output, showing the bit of over-riding I am getting wrong.

BASE
A
B
A


---FILTER---
FILTERED BASE
DID NOT FILTER A
Looks like a B     <------THIS IS WHAT I WANT
DID NOT FILTER B


---PROCESS---
FILTERED BASE
DID NOT FILTER A
DID NOT FILTER B   <-----THIS MAKES ME SAD

These are the class definitions

#include <string>
#include <iostream>
#include <memory>
//**************MESSAGES ************************
class Message
{
public:
    virtual std::string getOut() const {return std::string("BASE");};
};

class MsgA : public Message
{
public:
    std::string getOut() const {return std::string("A");};
};

class MsgB : public Message
{
public:
    std::string getOut() const {return std::string("B");};
};


//**************FILTERS ************************
class MsgFilter
{
public:
    void filter(const std::tr1::shared_ptr<Message> msg)
    {
        if( msg->getOut().compare("BASE") == 0 )
        {
            std::cout << "FILTERED BASE" << std::endl;
        }
        else
        {
            std::cout << "DID NOT FILTER " << msg->getOut() <<  std::endl;
        }
    }
    void filter(const std::tr1::shared_ptr<MsgB> msg)
    {
        std::cout << "Looks like a B" << std::endl;
        filter((std::tr1::shared_ptr<Message>)msg);
    }
};

//**************PROCESSORS ************************
class MsgProcessor
{
public:
    MsgProcessor():myFilt(MsgFilter()){}
    MsgFilter myFilt;
    virtual void process(std::tr1::shared_ptr<Message> msg)
    {
        myFilt.filter(msg);
    }
};

Below is the main function

#include "AllClassDefs.h"

int main(int argc, char *argv[])
{
    std::tr1::shared_ptr<Message> msg(new Message());
    std::tr1::shared_ptr<MsgA> msgA(new MsgA());
    std::tr1::shared_ptr<MsgB> msgB(new MsgB());

    std::cout << msg->getOut() << std::endl;
    std::cout << msgA->getOut() << std::endl;
    std::cout << msgB->getOut() << std::endl;

    std::tr1::shared_ptr<Message> msgAPtr(msgA);
    std::cout << msgAPtr->getOut() << std::endl;

    std::cout <<"\n\nFILTER---\n";

    std::tr1::shared_ptr<MsgFilter> flt(new MsgFilter());
    flt->filter(msg);
    flt->filter( (std::tr1::shared_ptr<Message>) msgA);
    flt->filter(msgB);

    std::cout <<"\n\nPROCESS---\n";

    std::tr1::shared_ptr<MsgProcessor> prc(new MsgProcessor());
    prc->process(msg);
    prc->process(msgA);
    prc->process(msgB);
}

So… I want the process function to call the Filter(MsgB) function and not the Filter(Message) function. By using the process(Message) function it looks like Filter treats the MsgB as a Message regardless. How can I make it work nicely?

I know I could use templates, but that seems a bit over overkill… plus in a system with a couple of hundred messages classes it might lead to a big exe. Should I use double dispatch? Or is there an easier solution to this?

Answer 1

Your problem is that overload resolution is performed at compile time, so this function

virtual void process(std::tr1::shared_ptr<Message> msg)

will always call the same overload

void filter(const std::tr1::shared_ptr<Message> msg)

because that matches the static type of the parameter.

It doesn’t matter if msg sometimes happens to point to a derive class.