CS100 Lecture 19

Operator Overloading

Contents

• Basics
• Example: Rational
  • Arithmetic and relational operators
  • Increment and decrement operators (++, --)
  • IO operators (<<, >>)
• Example: Dynarray
  • Subscript operator ([])
• Example: WindowPtr
  • Dereference (indirection) operator (*)
  • Member access through pointer (->)
• User-defined type conversions

Basics

Operator overloading: Provide the behaviors of operators for class types.

We have already seen some:

• The copy assignment operator and the move assignment operator are two special overloads for operator=.
• The IOStream library provides overloaded operator<< and operator>> to perform input and output.
• The string library provides operator+ for concatenation of strings, and <, <=, >, >=, ==, != for comparison in lexicographical order.
• Standard library containers and std::string have operator[].
• Smart pointers have operator* and operator->.

Basics

Overloaded operators can be defined in two forms:

• as a member function, in which the leftmost operand is bound to this:

  • a[i] a.operator[](i)
  • a = b a.operator=(b)
  • *a a.operator*()
  • f(arg1, arg2, arg3, ...) f.operator()(arg1, arg2, arg3, ...)

• as a non-member function:

  • a == b operator==(a, b)
  • a + b operator+(a, b)

Basics

Some operators cannot be overloaded:

obj.mem, ::, ?:, obj.*memptr (not covered in CS100)

Some operators can be overloaded, but are strongly not recommended:

cond1 && cond2, cond1 || cond2

• Reason: Since x && y would become operator&&(x, y), there is no way to overload && (or ||) that preserves the short-circuit evaluation property.

Basics

• At least one operand should be a class type. Modifying the behavior of operators on built-in types is not allowed.

  int operator+(int, int);   // Error.
  MyInt operator-(int, int); // Still error.
  

• Inventing new operators is not allowed.

  double operator**(double x, double exp); // Error.
  

• Overloading does not modify the associativity, precedence and the operands' evaluation order.

  std::cout << a + b; // Equivalent to `std::cout << (a + b)`.
  

Example: Rational

A class for rational numbers

class Rational {
  int m_num;        // numerator
  unsigned m_denom; // denominator
  void simplify() { // Private, because this is our implementation detail.
    int gcd = std::gcd(m_num, m_denom); // std::gcd in <numeric> (since C++17)
    m_num /= gcd; m_denom /= gcd;
  }
public:
  Rational(int x = 0) : m_num{x}, m_denom{1} {} // Also a default constructor.
  Rational(int num, unsigned denom) : m_num{num}, m_denom{denom} { simplify(); }
  double to_double() const {
    return static_cast<double>(m_num) / m_denom;
  }
};

We want to have arithmetic operators supported for Rational.

Rational: arithmetic operators

A good way: define operator+= and the unary operator-, and then define other operators in terms of them.

class Rational {
  friend Rational operator-(const Rational &); // Unary `operator-` as in `-x`.
public:
  Rational &operator+=(const Rational &rhs) {
    m_num = m_num * static_cast<int>(rhs.m_denom) // Be careful with `unsigned`!
            + static_cast<int>(m_denom) * rhs.m_num;
    m_denom *= rhs.m_denom;
    simplify();
    return *this; // `x += y` should return a reference to `x`.
  }
};
Rational operator-(const Rational &x) {
  return {-x.m_num, x.m_denom};
  // The above is equivalent to `return Rational(-x.m_num, x.m_denom);`.
}

Rational: arithmetic operators

Define the arithmetic operators in terms of the compound assignment operators.

class Rational {
public:
  Rational &operator-=(const Rational &rhs) {
    // Makes use of `operator+=` and the unary `operator-`.
    return *this += -rhs;
  }
};
Rational operator+(const Rational &lhs, const Rational &rhs) {
  return Rational(lhs) += rhs; // Makes use of `operator+=`.
}
Rational operator-(const Rational &lhs, const Rational &rhs) {
  return Rational(lhs) -= rhs; // Makes use of `operator-=`.
}

[Best practice] Avoid repetition.

class Rational {
public:
  Rational &operator+=(const Rational &rhs) {
    m_num = m_num * static_cast<int>(rhs.m_denom)
           + static_cast<int>(m_denom) * rhs.m_num;
    m_denom *= rhs.m_denom;
    simplify();
    return *this;
  }
};

The arithmetic operators for Rational are simple yet requires carefulness.

• Integers with different signed-ness need careful treatment.
• Remember to simplify().

Fortunately, we only need to pay attention to these things in operator+=. Everything will be right if operator+= is right.

[Best practice] Avoid repetition.

The code would be very error-prone if you implement every function from scratch!

class Rational {
public:
  Rational &operator+=(const Rational &rhs) {
    m_num = m_num * static_cast<int>(rhs.m_denom)
           + static_cast<int>(m_denom) * rhs.m_num;
    m_denom *= rhs.m_denom;
    simplify();
    return *this;
  }
  Rational &operator-=(const Rational &rhs) {
    m_num = m_num * static_cast<int>(rhs.m_denom)
           - static_cast<int>(m_denom) * rhs.m_num;
    m_denom *= rhs.m_denom;
    simplify();
    return *this;
  }
  friend Rational operator+(const Rational &,
                            const Rational &);
  friend Rational operator-(const Rational &,
                            const Rational &);
};

Rational operator+(const Rational &lhs,
                   const Rational &rhs) {
  return {
    lhs.m_num * static_cast<int>(rhs.m_denom)
        + static_cast<int>(lhs.m_denom) * rhs.lhs,
    lhs.m_denom * rhs.m_denom
  };
}
Rational operator-(const Rational &lhs,
                   const Rational &rhs) {
  return {
    lhs.m_num * static_cast<int>(rhs.m_denom)
        - static_cast<int>(lhs.m_denom) * rhs.lhs,
    lhs.m_denom * rhs.m_denom
  };
}

Rational: arithmetic operators

Exercise: Define operator* (multiplication) and operator/ (division) as well as operator*= and operator/= for Rational.

Rational: arithmetic and relational operators

What if we define them (say, operator+) as member functions?

class Rational {
public:
  Rational(int x = 0) : m_num{x}, m_denom{1} {}
  Rational operator+(const Rational &rhs) const {
    return {
      m_num * static_cast<int>(rhs.m_denom)
          + static_cast<int>(m_denom) * rhs.m_num,
      m_denom * rhs.m_denom
    };
  }
};

Rational: arithmetic and relational operators

What if we define them (say, operator+) as member functions?

class Rational {
public:
  Rational(int x = 0) : m_num{x}, m_denom{1} {}
  Rational operator+(const Rational &rhs) const {
    // ...
  }
};

Rational r = some_value();
auto s = r + 0; // OK, `r.operator+(0)`, effectively `r.operator+(Rational(0))`
auto t = 0 + r; // Error! `0.operator+(r)` ???

Rational: arithmetic and relational operators

To allow implicit conversions on both sides, the operator should be defined as non-member functions.

Rational r = some_value();
auto s = r + 0; // OK, `operator+(r, 0)`, effectively `operator+(r, Rational(0))`
auto t = 0 + r; // OK, `operator+(0, r)`, effectively `operator+(Rational(0), r)`

[Best practice] The "symmetric" operators, whose operands are often exchangeable, often should be defined as non-member functions.

Rational: relational operators

Define < and ==, and define others in terms of them. (Before C++20)

• Since C++20: Define == and <=>, and the compiler will generate others.

A possible way: Use to_double and compare the floating-point values.

bool operator<(const Rational &lhs, const Rational &rhs) {
  return lhs.to_double() < rhs.to_double();
}

• This does not require operator< to be a friend.
• However, this is subject to floating-point errors.

Rational: ralational operators

Another way (possibly better):

class Rational {
  friend bool operator<(const Rational &, const Rational &);
  friend bool operator==(const Rational &, const Rational &);
};
bool operator<(const Rational &lhs, const Rational &rhs) {
  return static_cast<int>(rhs.m_denom) * lhs.m_num
        < static_cast<int>(lhs.m_denom) * rhs.m_num;
}
bool operator==(const Rational &lhs, const Rational &rhs) {
  return lhs.m_num == rhs.m_num && lhs.m_denom == rhs.m_denom;
}

If there are member functions to obtain the numerator and the denominator, these functions don't need to be friend.

Rational: relational operators

[Best practice] Avoid repetition.

Define others in terms of < and ==:

bool operator>(const Rational &lhs, const Rational &rhs) {
  return rhs < lhs;
}
bool operator<=(const Rational &lhs, const Rational &rhs) {
  return !(lhs > rhs);
}
bool operator>=(const Rational &lhs, const Rational &rhs) {
  return !(lhs < rhs);
}
bool operator!=(const Rational &lhs, const Rational &rhs) {
  return !(lhs == rhs);
}

Relational operators

Define relational operators in a consistent way:

• a != b should mean !(a == b)
• !(a < b) and !(a > b) should imply a == b

C++20 has devoted some efforts to the design of consistent comparison: P0515r3.

Relational operators

Avoid abuse of relational operators:

struct Point2d { double x, y; };
bool operator<(const Point2d &lhs, const Point2d &rhs) {
  return lhs.x < rhs.x; // Is this the unique, best behavior?
}
// Much better design: Use a named function.
bool less_in_x(const Point2d &lhs, const Point2d &rhs) {
  return lhs.x < rhs.x;
}

[Best practice] Operators should be used for operations that are likely to be unambiguous to users.

• If an operator has plausibly more than one interpretation, use named functions instead. Function names can convey more information.

std::string has operator+ for concatenation. Why doesn't std::vector have one?

++ and --

++ and -- are often defined as members, because they modify the object.

To differentiate the postfix version x++ and the prefix version ++x: The postfix version has a parameter of type int.

• The compiler will translate ++x to x.operator++(), x++ to x.operator++(0).

class Rational {
public:
  Rational &operator++() { ++m_num; simplify(); return *this; }
  Rational operator++(int) { // This `int` parameter is not used.
    // The postfix version is almost always defined like this.
    auto tmp = *this;
    ++*this; // Makes use of the prefix version.
    return tmp;
  }
};

++ and --

class Rational {
public:
  Rational &operator++() { ++m_num; simplify(); return *this; }
  Rational operator++(int) { // This `int` parameter is not used.
    // The postfix version is almost always defined like this.
    auto tmp = *this;
    ++*this; // Make use of the prefix version.
    return tmp;
  }
};

The prefix version returns reference to *this, while the postfix version returns a copy of *this before incrementation.

• Same as the built-in behaviors.

IO operators

Implement std::cin >> r and std::cout << r.

Input operator:

std::istream &operator>>(std::istream &, Rational &);

Output operator:

std::ostream &operator<<(std::ostream &, const Rational &);

• std::cin is of type std::istream, and std::cout is of type std::ostream.

• The left-hand side operand should be returned, so that we can write

  std::cin >> a >> b >> c; std::cout << a << b << c;
  

Rational: output operator

class Rational {
  friend std::ostream &operator<<(std::ostream &, const Rational &);
};
std::ostream &operator<<(std::ostream &os, const Rational &r) {
  return os << r.m_num << '/' << r.m_denom;
}

If there are member functions to obtain the numerator and the denominator, it don't have to be a friend.

std::ostream &operator<<(std::ostream &os, const Rational &r) {
  return os << r.get_numerator() << '/' << r.get_denominator();
}

Rational: input operator

Suppose the input format is a b for the rational number , where a and b are integers.

std::istream &operator>>(std::istream &is, Rational &r) {
  int x, y; is >> x >> y;
  if (!is) { // Pay attention to input failures!
    x = 0;
    y = 1;
  }
  if (y < 0) { y = -y; x = -x; }
  r = Rational(x, y);
  return is;
}

Example: Dynarray

operator[]

class Dynarray {
public:
  int &operator[](std::size_t n) {
    return m_storage[n];
  }
  const int &operator[](std::size_t n) const {
    return m_storage[n];
  }
};

The use of a[i] is interpreted as a.operator[](i).

(C++23 allows a[i, j, k]!)

Homework: Define operator[] and relational operators for Dynarray.

Example: WindowPtr

WindowPtr: indirection (dereference) operator

Recall the WindowPtr class we defined in the previous lecture.

struct WindowWithCounter {
  Window theWindow;
  int refCount = 1;
};
class WindowPtr {
  WindowWithCounter *m_ptr;
public:
  Window &operator*() const { // Why should it be const?
    return m_ptr->theWindow;
  }
};

We want *sp to return reference to the managed object.

WindowPtr: indirection (derefernce) operator

Why should operator* be const?

class WindowPtr {
  WindowWithCounter *m_ptr;
public:
  Window &operator*() const {
    return m_ptr->theWindow;
  }
};

On a const WindowPtr ("top-level" const), obtaining a non-const reference to the managed object may still be allowed.

• The (smart) pointer is const, but the managed object is not.
• this is const WindowPtr *, so m_ptr is WindowWithCounter *const.

WindowPtr: member access through pointer

To make operator-> consistent with operator* (make ptr->mem equivalent to (*ptr).mem), operator-> is almost always defined like this:

class WindowPtr {
public:
  Window *operator->() const {
    return std::addressof(operator*());
  }
};

std::addressof(x) is almost always equivalent to &x, but the latter may not return the address of x if operator& for x has been overloaded!

User-defined type conversions

Type conversions

A type conversion is a function for two different types and .

Type conversions can happen either implicitly or explicitly. A conversion is explicit if and only if the target type U is written explicitly in the conversion expression.

Explicit conversions can happen in one of the following forms:

Type conversions

A type conversion is a function for two different types and .

Type conversions can happen either implicitly or explicitly. A conversion is explicit if and only if the target type U is written explicitly in the conversion expression.

• Arithmetic conversions are often allowed to happen implicitly:

  int sum = /* ... */, n = /* ... */;
  auto average = 1.0 * sum / n; // `sum` and `n` are converted to `double`,
                                // so `average` has type `double`.
  

• The dangerous conversions for built-in types must be explicit:

  const int *cip = something();
  auto ip = const_cast<int *>(cip);       // int *
  auto cp = reinterpret_cast<char *>(ip); // char *
  

Type conversions

A type conversion is a function for two different types and .

Type conversions can happen either implicitly or explicitly. A conversion is explicit if and only if the target type U is written explicitly in the conversion expression.

• This is also a type conversion, isn't it?

  std::string s = "hello"; // from `const char [6]` to `std::string`
  

• This is also a type conversion, isn't it?

  std::size_t n = 1000;
  std::vector<int> v(n); // from `std::size_t` to `std::vector<int>`
  

How do these type conversions happen? Are they implicit or explicit?

Type conversions

We can define a type conversion for our class X in one of the following ways:

1. A constructor with exactly one parameter of type T is a conversion from T to X.

   • Example: std::string has a constructor accepting a const char *. std::vector has a constructor accepting a std::size_t.

2. A type conversion operator: a conversion from X to some other type.

   class Rational {
   public:
     // conversion from `Rational` to `double`.
     operator double() const { return 1.0 * m_num / m_denom; }
   };
   Rational r(3, 4);
   double dval = r;  // 0.75
   

Type conversion operator

A type conversion operator is a member function of class X, which defines the type conversion from X to some other type T.

class Rational {
public:
  // conversion from `Rational` to `double`.
  operator double() const { return 1.0 * m_num / m_denom; }
};
Rational r(3, 4);
double dval = r;  // 0.75

• The name of the function is operator T.
• The return type is T, which is not written before the name.
• A type conversion is usually a read-only operation, so it is usually const.

Explicit type conversion

Some conversions should be allowed to happen implicitly:

void foo(const std::string &str) { /* ... */ }
foo("hello"); // implicit conversion from `const char [6]` to `const char *`,
              // and then to `std::string`.

Some should never happen implicitly!

void bar(const std::vector<int> &vec) { /* ...*/ }
bar(1000);                  // ??? Too weird!
bar(std::vector<int>(1000)) // OK.
std::vector<int> v1(1000);  // OK.
std::vector<int> v2 = 1000; // No! This should never happen. Too weird!

Explicit type conversion

To disallow the implicit use of a constructor as a type conversion, write explicit before the return type:

class string { // Suppose this is the `std::string` class.
public:
  string(const char *cstr); // Not marked `explicit`. Implicit use is allowed.
};

template <typename T> class vector { // Suppose this is the `std::vector` class.
public:
  explicit vector(std::size_t n); // Implicit use is not allowed.
};

class Dynarray {
public:
  explicit Dynarray(std::size_t n) : m_length{n}, m_storage{new int[n]{}} {}
};

Explicit type conversion

To disallow the implicit use of a type conversion operator, also write explicit:

class Rational {
public:
  explicit operator double() const { return 1.0 * m_num / m_denom; }
};
Rational r(3, 4);
double d = r;                     // Error.
void foo(double x) { /* ... */ }
foo(r);                           // Error.
foo(double(r));                   // OK.
foo(static_cast<double>(r));      // OK.

[Best practice] Avoid the abuse of type conversion operators.

Type conversion operators can lead to unexpected results!

class Rational {
public:
  operator double() const { return 1.0 * m_num / m_denom; }
  operator std::string() const {
    return std::to_string(m_num) + " / " + std::to_string(m_denom);
  }
};
int main() {
  Rational r(3, 4);
  std::cout << r << '\n'; // Ooops! Is it `0.75` or `3 / 4`?
}

In the code above, either mark the type conversions as explicit, or remove them and define named functions like to_double() and to_string() instead.

Contextual conversion to bool

A special rule for conversion to bool.

Suppose expr is an expression of a class type X, and suppose X has an explicit type conversion operator to bool. In the following contexts, that conversion is applicable even if it is not written as bool(expr) or static_cast<bool>(expr):

• if (expr), while (expr), for (...; expr; ...), do ... while (expr)
• as the operand of !, &&, ||
• as the first operand of ?:: expr ? something : something_else

Contextual conversion to bool

Exercise: We often test whether a pointer is non-null like this:

if (ptr) {
  // ...
}
auto val = ptr ? ptr->some_value : 0;

Define a conversion from WindowPtr to bool, so that we can test whether a WindowPtr is non-null in the same way.

• Should this conversion be allowed to happen implicitly? If not, mark it explicit.

Summary

Operator overloading

• As a non-member function: @a operator@(a), a @ b operator@(a, b)
• As a member function: @a a.operator@(), a @ b a.operator@(b)
  • The postfix ++ and -- are special: They have a special int parameter to make them different from the prefix ones.
  • The arrow operator -> is special: Although it looks like a binary operator in ptr->mem, it is unary and involves special rules.
    • You don't need to understand the exact rules for ->.
• Avoid repetition.
• Avoid abuse of operator overloading.

Summary

Type conversions

• Implicit vs explicit
• User-defined type conversions: either through a constructor or through a type conversion operator.
• To disable the implicit use of the user-defined type conversion: explicit
• Avoid abuse of type conversion operators.
• Conversion to bool has some special rules (contextual conversion).