继承的运算符重载不适用于派生类

Question

I wrote some code that manages to mix class templates, class inheritance, and operator overloading and I'm lost as to how to solve an issue regarding operator use. I have a base class with the bulk of the code (particularly, operator overloading implementation and data holders):

template <typename _type> 
class baseMatrix {
public:
    baseMatrix();
    ~baseMatrix();

    //operators 
    baseMatrix<_type>& operator= (baseMatrix<_type> _mat);
    template <typename _input_type> baseMatrix<_type>& operator*= (_input_type _val);
    template <typename _input_type> baseMatrix<_type>& operator/= (_input_type _val); 
    template <typename _input_type> baseMatrix<_type>& operator+= (_input_type _val); 
    template <typename _input_type> baseMatrix<_type>& operator-= (_input_type _val);
    template <typename _input_type> baseMatrix<_type>& operator*= (const baseMatrix<_input_type>& _mat);
    template <typename _input_type> baseMatrix<_type>& operator/= (const baseMatrix<_input_type>& _mat);
    template <typename _input_type> baseMatrix<_type>& operator+= (const baseMatrix<_input_type>& _mat);
    template <typename _input_type> baseMatrix<_type>& operator-= (const baseMatrix<_input_type>& _mat);        

protected:
    std::vector<_type> data;
};

/* ... */
template <typename _type>
template <typename _input_type> 
baseMatrix<_type>& baseMatrix<_type>::operator*=(_input_type _val) { 
    for (int i = 0; i < data.size(); ++i) data[i]*=_val; 
    return *this;
};
template <typename _type>
template <typename _input_type> 
baseMatrix<_type>& baseMatrix<_type>::operator*=(const baseMatrix<_input_type>& _mat) { 
    for (int i = 0; i < data.size(); ++i) data[i]*=_mat.data[i]; 
    return *this;
};
/* remaining operator overload functions */

I overloaded operators for both scalar and class arguments. I then have and additional class matrix2D that inherits those operators from baseMatrix :

template <typename _type> 
class matrix2D : public baseMatrix<_type> {
public:
    matrix2D(int _rows, int _cols);
    matrix2D(int _rows, int _cols, _type _val);
    ~matrix2D();

    _type& operator()(int _r, int _c); 
    _type& at(int _r, int _c);

protected:
    int nRows,nCols;
    using baseMatrix<_type>::data;
};

However, when instantiating these classes I'm only able to call the scalar operators, as using eg *= with two matrix2D objects results in a compilation error:

In file included from test.cpp:1:
baseMatrix.hpp: In instantiation of ‘baseMatrix<_type>& baseMatrix<_type>::operator*=(_input_type) [with _input_type = matrix2D<float>; _type = float]’:
test.cpp:29:6:   required from here
baseMatrix.hpp:56:47: error: no match for ‘operator*=’ (operand types are ‘__gnu_cxx::__alloc_traits<std::allocator<float>, float>::value_type’ {aka ‘float’} and ‘matrix2D<float>’)
for (int i = 0; i < data.size(); ++i) data[i]*=_val;

If, on the other hand, I instantiate a baseMatrix object, it compiles OK (fails at runtime for other reasons, ie unitialized data):

int main(int argc, char const *argv[]){
    matrix2D<float> M1(5,5,0.0);
    matrix2D<float> M2(3,3,6.0);
    baseMatrix<float> testM;

    M2*=0.47;   // works
    M2*=M1;     // does not compile
    M2*=testM  // runtime error (segfault)

}

So apparently the operator overload is not working for derived classes, what would be the correct syntax?

EDIT: I've realized that the problem is with multiple operator overloads. For some reason it is able to compile if I only declare the operators to take a baseMatrix object as argument, or vice-versa.

Answer 1

So for M1*=M2 , overload resolution checks to see whether operator*=(M1, M2) and/or M1.operator*=(M2) are possible. There are no good candidates for non-member operator*= . Since M1 has type matrix2D<float> which inherits baseMatrix<float> , the compiler sees that M1 has member functions:

template <typename _input_type> baseMatrix<float>& operator*= (_input_type _val); // #1
template <typename _input_type> baseMatrix<float>& operator*= (const baseMatrix<_input_type>& _mat); // #2

The next step is to attempt to deduce template arguments for each template in the overload set. Both succeed: For template #1, it finds a valid specialization by just taking _input_type = matrix2D<float> :

baseMatrix<float>& baseMatrix<float>::operator*=<matrix2D<float>>(matrix2D<float> _val); // #3

For template #2, it finds a specialization by using the base class of the M2 argument type and determining _input_type = float :

baseMatrix<float>& baseMatrix<float>::operator*=<float>(const baseMatrix<float>& _val); // #4

Then these function template specializations are compared. The trouble here is that you meant to use #4, but #3 is considered a better match than #4, since #3 uses the exact type of argument M2 , but #4 needs a derived-to-base conversion. Then the instantiation of #3 contains the statement data[i]*=_val; which doesn't make sense for data[i] a float and _val a matrix2D<float> , causing the error.

One solution is to use a SFINAE technique to make sure template #1 can't be used with a matrix type:

#include <utility>
#include <type_traits>

template <typename Type>
class baseMatrix {
private:
    template <typename ValType>
    static constexpr std::true_type is_matrix_ptr(const baseMatrix<ValType>*);
    static constexpr std::false_type is_matrix_ptr(const void*);

public:
    // ...
    template <typename ScalarType,
              std::enable_if_t<!decltype(is_matrix_ptr(std::declval<ScalarType*>()))
                  ::value>* = nullptr>
    baseMatrix& operator*=(ScalarType val);
    // ...
};