Question

I was wondering what the difference (of any kind) is when using a vector of vectors to represent a 2D matrix or make a class like:

template < class T > 
class Matrix2D {
public:
    Matrix2D( unsigned m, unsigned n ) : m( m ), n( n ), x( m * n ) {} ;
    Matrix2D( const Matrix2D<T> &matrix ) : m( matrix.m ), n( matrix.n) x( matrix.x ) {} ;
    Matrix2D& operator= ( const Matrix2D<T> &matrix ) ;
    T& operator ()( unsigned i, unsigned j ) ;
    void resize( int nx, int ny ) ;
private:
    unsigned m, n ;
    std::vector< T > x ;         
} ;


template <class T>
T& Matrix2D<T>::operator ()( unsigned i, unsigned j ) {
    return x[ j + n * i ] ;
}

template <class T>
Matrix2D<T>& Matrix2D<T>::operator= ( const Matrix2D<T> &matrix ) {
    m = matrix.m ;
    n = matrix.n ;
    x = matrix.x ;
    return *this ;
}

template <class T>
void Matrix2D<T>::resize( int nx, int ny ) {
    m = nx ;
    n = ny ;
    x.resize( nx * ny ) ;
}

Edit: Ignore the resize method, as Erik pointed out it would not keep original data place. I only added for a specific task where I didn’t mind. The basic class is just the ctor and the () operator.

Answer 1

• – .resize() will not keep existing data in the original positions.
• – Syntax differences, operator() vs operator[]
• – No iterators, and no using e.g. std:: algorithms
• + Better locality, backing vector has contiguous memory
• + More understandable syntax for initialization
• + Guarantees that the array isn’t jagged

In short, the class is fine and likely better for specialized purposes, but it’s not doing too well on generic-purpose.