OOP55-CPP. Do not use pointer-to-member operators to access nonexistent members

The pointer-to-member operators .* and ->* are used to obtain an object or a function as though it were a member of an underlying object. For instance, the following are functionally equivalent ways to call the member function f() on the object o:

struct S {
  void f() {}
};

void func() {
  S o;
  void (S::*pm)() = &S::f;
  
  o.f();
  (o.*pm)();
}

The call of the form o.f() uses class member access at compile time to look up the address of the function S::f() on the object o. The call of the form (o.*pm)() uses the pointer-to-member operator .* to call the function at the address specified by pm. In both cases, the object o is the implicit this object within the member function S::f().

The C++ Standard, [expr.mptr.oper], paragraph 4 [ISO/IEC 14882-2014], states :

Abbreviating pm-expression.*cast-expression as E1.*E2, E1 is called the object expression. If the dynamic type of E1 does not contain the member to which E2 refers, the behavior is undefined.

(A pointer-to-member expression of the form E1->*E2 is converted to its equivalent form, (*(E1)).*E2, so use of pointer-to-member expressions of either form behave equivalently in terms of undefined behavior.)

Further, the C++ Standard, [expr.mptr.oper], paragraph 6, states, in part:

If the second operand is the null pointer to member value, the behavior is undefined.

Do not use a pointer-to-member expression where the dynamic type of the first operand does not contain the member to which the second operand refers, including the use of a null pointer-to-member value as the second operand.

Noncompliant Code Example

In this noncompliant code example, a pointer-to-member object is obtained from D::g but upcast to be a B::*. When called on an object whose dynamic type is D, the pointer-to-member call is well defined. However, in this noncompliant code example, the dynamic type of the underlying object is B, resulting in undefined behavior:

struct B {
  virtual ~B() = default;
};

struct D : B {
  virtual ~D() = default;
  virtual void g() { /* ... */ }
};

void f() {
  B *b = new B;
 
  // ...
 
  void (B::*gptr)() = static_cast<void(B::*)()>(&D::g);
  (b->*gptr)();
  delete b;
}

Compliant Solution

In this compliant solution, the upcast is removed, rendering the initial code ill-formed and emphasizing the underlying problem: that B::g() does not exist. This compliant solution assumes that the programmer's intention was to use the correct dynamic type for the underlying object:

struct B {
  virtual ~B() = default;
};

struct D : B {
  virtual ~D() = default;
  virtual void g() { /* ... */ }
};

void f() {
  B *b = new D; // Corrected dynamic object type
 
  // ...
  void (D::*gptr)() = &D::g; // Moved static_cast to next line
  (static_cast<D *>(b)->*gptr)();
  delete b;
}

Noncompliant Code Example

In this noncompliant code example, a null pointer-to-member value is passed as the second operand to a pointer-to-member expression, resulting in undefined behavior:

struct B {
  virtual ~B() = default;
};

struct D : B {
  virtual ~D() = default;
  virtual void g() { /* ... */ }
};
 
static void (D::*gptr)();
void call_memptr(D *ptr) {
  (ptr->*gptr)();
}
 
void f() {
  D *d = new D;
  call_memptr(d);
  delete d;
}

Compliant Solution

In this compliant solution, gptr is properly initialized to a valid pointer-to-member value instead of to the default value of nullptr:

struct B {
  virtual ~B() = default;
};
 
struct D : B {
  virtual ~D() = default;
  virtual void g() { /* ... */ }
};
 
static void (D::*gptr)() = &D::g;
void call_memptr(D *ptr) {
  (ptr->*gptr)();
}
 
void f() {
  D *d = new D;
  call_memptr(d);
  delete d;
}

Risk Assessment

Automated Detection

Related Vulnerabilities

Search for other vulnerabilities resulting from the violation of this rule on the CERT website.

Bibliography