C++ Pure virtual functions


We can also specify that a virtual function is pure virtual (abstract), by appending = 0 to the declaration. Classes with one or more pure virtual functions are considered to be abstract, and cannot be instantiated; only derived classes which define, or inherit definitions for, all pure virtual functions can be instantiated.

struct Abstract {
    virtual void f() = 0;

struct Concrete {
    void f() override {}

Abstract a; // Error.
Concrete c; // Good.

Even if a function is specified as pure virtual, it can be given a default implementation. Despite this, the function will still be considered abstract, and derived classes will have to define it before they can be instantiated. In this case, the derived class' version of the function is even allowed to call the base class' version.

struct DefaultAbstract {
    virtual void f() = 0;
void DefaultAbstract::f() {}

struct WhyWouldWeDoThis : DefaultAbstract {
    void f() override { DefaultAbstract::f(); }

There are a couple of reasons why we might want to do this:

  • If we want to create a class that can't itself be instantiated, but doesn't prevent its derived classes from being instantiated, we can declare the destructor as pure virtual. Being the destructor, it must be defined anyways, if we want to be able to deallocate the instance. And as the destructor is most likely already virtual to prevent memory leaks during polymorphic use, we won't incur an unnecessary performance hit from declaring another function virtual. This can be useful when making interfaces.

      struct Interface {
          virtual ~Interface() = 0;
      Interface::~Interface() = default;
      struct Implementation : Interface {};
      // ~Implementation() is automatically defined by the compiler if not explicitly
      //  specified, meeting the "must be defined before instantiation" requirement.
  • If most or all implementations of the pure virtual function will contain duplicate code, that code can instead be moved to the base class version, making the code easier to maintain.

      class SharedBase {
          State my_state;
          std::unique_ptr<Helper> my_helper;
          // ...
          virtual void config(const Context& cont) = 0;
          // ...
      /* virtual */ void SharedBase::config(const Context& cont) {
          my_helper = new Helper(my_state, cont.relevant_field);
      class OneImplementation : public SharedBase {
          int i;
          // ...
          void config(const Context& cont) override;
          // ...
      void OneImplementation::config(const Context& cont) /* override */ {
          my_state = { cont.some_field, cont.another_field, i };
      // And so on, for other classes derived from SharedBase.