Avoiding ambiguity in overload resolution

问题

This is a follow up to this question so if you need to see the Register class please refer to that question. Now based on the supplied answer I have written a function to do just that. I have 2 versions of the function one that will store the results back into the original and one that will return a copy. Here are my functions:

template<std::uint64_t N>
void reverseBitOrder( Register<N>& reg ) {
    auto str = reg.register_.to_string();
    std::reverse(str.begin(), str.end());
    auto x = vpc::Byte(str);
    reg.register_ = x;
}

// Extra unused parameter to avoid ambiguity
template<std::uint64_t N>
Register<N> reverseBitOrder(Register<N>& reg, bool _x_ /*not used*/ ) {
    auto str = reg.register_.to_string();
    std::reverse(str.begin(), str.end());
    auto x = vpc::Byte(str);
    Register<N> temp;
    temp.register_ = x;
    return temp;
}

The first one saves the value, the second returns a copy. My question is on the 2^nd function I ended up adding a second parameter that is not used in order to avoid ambiguity due to overload resolution as functions can not be resolved on return types alone. So when I call this function I would have to pass either 0, 1, true or false to the function which has no effect.

Overall this in itself is not a very big deal, however, it doesn't seem very clean and concise to me. Are there any other ways to achieve this? I prefer not to make this a function of the class. My Register class or struct is complete as is and any kind of operations done on a register will be done by functions that take a reference to one or more register objects.

回答1:

You can use std::optional to achieve this.

The return type of function template reverseBitOrder should be std::optional<vpc::Register<N>>.

The function template should be modified to:

template<std::uint64_t N>
std::optional<vpc::Register<N>> reverseBitOrder(vpc::Register<N>& reg, bool _x_ = false) {

    auto str = reg.register_.to_string();
    std::reverse(str.begin(), str.end());
    vpc::Register<N> temp;    

    if(_x_) //return copy
    {
        temp.register_ = vpc::Byte(str); //since register_ is a vpc::Byte. Generalize accordingly.
        return temp;
    }
    else //save in the same variable
    {
        reg.register_ = vpc::Byte(str);        
        return {};
    }
}

Live Demo here.

But you don't really need to use std::optional, since there is really no "failure" case in the function template.

回答2:

If I understand correctly, both functions compute x and store it in a Register<N>, but one returns said object by value while the other stores the result in the function's argument reg.

Technically, this can be done by overloading on constness by defining those two functions:

template<std::uint64_t N> void reverseBitOrder( Register<N>& reg );
template<std::uint64_t N> Register<N> reverseBitOrder( Register<N> const& reg );

While this technically answers your question, that would be terrible design. If I'm not mistaken, the real issue is that what you want is this:

// Behaviour 1: value stored in reg
reverseBitOrder(reg);

// Behaviour 2: value stored in val, reg left untouched
auto val = reverseBitOrder(reg);

The problem with this is that whether the returned value is used or not is not something you can detect from inside the function.

---

The proper way to have one function do two things here would be to have a function with this signature:

template<std::uint64_t N> void reverseBitOrder( Register<N> const& inputReg, Register<N>& outputReg );

That function would use inputReg to compute x then store the result in outputReg, meaning you would use it like this:

// Behaviour 1: value stored in reg
reverseBitOrder(reg, reg);

// Behaviour 2: value stored in val, reg leftuntouched
reverseBitOrder(reg, val);

---

Now, if that really doesn't do it for you, there is a way to get the syntaxic sugar you're looking for, at the cost of unneeded complexity and of adding a constructor to Register<N>. It would look roughly like this:

// Forward declaration
template<std::uint64_t N>
class Register;

// Proxy class
template<std::uint64_t N>
struct RegisterProxy
{
    Register<N>* reg;
    TypeOfX x;

    ~RegisterProxy()
    {
        if (reg)
        {
            reg->register = x;
        }
    }
};

// Modified Register class
template<std::uint64_t N>
class Register
{
    ...

    Register(RegisterProxy& proxy)
    {
        ...
        register = proxy.x;
        proxy.reg = nullptr;
        ...
    }

    ...

    // Define the matching assignment operator too

    ...
};

// Your function
template<std::uint64_t N>
RegisterProxy<N> reverseBitOrder( Register<N>& reg )
{
    auto str = reg.register_.to_string();
    std::reverse(str.begin(), str.end());
    auto x = vpc::Byte(str);
    return RegisterProxy{reg, x};
}

That allows you to do

// Behaviour 1: temporary object immediately destroyed and value stored in reg
reverseBitOrder(reg);

// Behaviour 2: constructor called, value stored in val
//              the temporary object's destructor doesn't do anything, reg left untouched
auto val = reverseBitOrder(reg);

I can't recommend doing that though, it's more trouble than it's worth. I suggest using the solution I called "the proper way". It the least complicated and the one which is the hardest to use wrong later

来源：https://stackoverflow.com/questions/55999515/avoiding-ambiguity-in-overload-resolution