Callback functions in c++

Question

In c++, when and how do you use a callback function?

EDIT:
I would like to see a simple example to write a callback function.

Answer 1

There is also the C way of doing callbacks: function pointers

//Define a type for the callback signature,
//it is not necessary, but makes life easier

//Function pointer called CallbackType that takes a float
//and returns an int
typedef int (*CallbackType)(float);  


void DoWork(CallbackType callback)
{
  float variable = 0.0f;

  //Do calculations

  //Call the callback with the variable, and retrieve the
  //result
  int result = callback(variable);

  //Do something with the result
}

int SomeCallback(float variable)
{
  int result;

  //Interpret variable

  return result;
}

int main(int argc, char ** argv)
{
  //Pass in SomeCallback to the DoWork
  DoWork(&SomeCallback);
}

Now if you want to pass in class methods as callbacks, the declarations to those function pointers have more complex declarations, example:

//Declaration:
typedef int (ClassName::*CallbackType)(float);

//This method performs work using an object instance
void DoWorkObject(CallbackType callback)
{
  //Class instance to invoke it through
  ClassName objectInstance;

  //Invocation
  int result = (objectInstance.*callback)(1.0f);
}

//This method performs work using an object pointer
void DoWorkPointer(CallbackType callback)
{
  //Class pointer to invoke it through
  ClassName * pointerInstance;

  //Invocation
  int result = (pointerInstance->*callback)(1.0f);
}

int main(int argc, char ** argv)
{
  //Pass in SomeCallback to the DoWork
  DoWorkObject(&ClassName::Method);
  DoWorkPointer(&ClassName::Method);
}

Answer 2

The accepted answer is very useful and quite comprehensive. However, the OP states

I would like to see a simple example to write a callback function.

So here you go, from C++11 you have std::function so there is no need for function pointers and similar stuff:

#include <functional>
#include <string>
#include <iostream>

void print_hashes(std::function<int (const std::string&)> hash_calculator) {
    std::string strings_to_hash[] = {"you", "saved", "my", "day"};
    for(auto s : strings_to_hash)
        std::cout << s << ":" << hash_calculator(s) << std::endl;    
}

int main() {
    print_hashes( [](const std::string& str) {   /** lambda expression */
        int result = 0;
        for (int i = 0; i < str.length(); i++)
            result += pow(31, i) * str.at(i);
        return result;
    });
    return 0;
}

This example is by the way somehow real, because you wish to call function print_hashes with different implementations of hash functions, for this purpose I provided a simple one. It receives a string, returns an int (a hash value of the provided string), and all that you need to remember from the syntax part is std::function<int (const std::string&)> which describes such function as an input argument of the function that will invoke it.

Answer 3

There isn't an explicit concept of a callback function in C++. Callback mechanisms are often implemented via function pointers, functor objects, or callback objects. The programmers have to explicitly design and implement callback functionality.

Edit based on feedback:

In spite of the negative feedback this answer has received, it is not wrong. I'll try to do a better job of explaining where I'm coming from.

C and C++ have everything you need to implement callback functions. The most common and trivial way to implement a callback function is to pass a function pointer as a function argument.

However, callback functions and function pointers are not synonymous. A function pointer is a language mechanism, while a callback function is a semantic concept. Function pointers are not the only way to implement a callback function - you can also use functors and even garden variety virtual functions. What makes a function call a callback is not the mechanism used to identify and call the function, but the context and semantics of the call. Saying something is a callback function implies a greater than normal separation between the calling function and the specific function being called, a looser conceptual coupling between the caller and the callee, with the caller having explicit control over what gets called. It is that fuzzy notion of looser conceptual coupling and caller-driven function selection that makes something a callback function, not the use of a function pointer.

For example, the .NET documentation for IFormatProvider says that "GetFormat is a callback method", even though it is just a run-of-the-mill interface method. I don't think anyone would argue that all virtual method calls are callback functions. What makes GetFormat a callback method is not the mechanics of how it is passed or invoked, but the semantics of the caller picking which object's GetFormat method will be called.

Some languages include features with explicit callback semantics, typically related to events and event handling. For example, C# has the event type with syntax and semantics explicitly designed around the concept of callbacks. Visual Basic has its Handles clause, which explicitly declares a method to be a callback function while abstracting away the concept of delegates or function pointers. In these cases, the semantic concept of a callback is integrated into the language itself.

C and C++, on the other hand, does not embed the semantic concept of callback functions nearly as explicitly. The mechanisms are there, the integrated semantics are not. You can implement callback functions just fine, but to get something more sophisticated which includes explicit callback semantics you have to build it on top of what C++ provides, such as what Qt did with their Signals and Slots.

In a nutshell, C++ has what you need to implement callbacks, often quite easily and trivially using function pointers. What it does not have is keywords and features whose semantics are specific to callbacks, such as raise, emit, Handles, event +=, etc. If you're coming from a language with those types of elements, the native callback support in C++ will feel neutered.

Answer 4

Boost's singals2 allows you to subscribe generic member functions (without templates!) and in a threadsafe way.

Example: Document-View Signals can be used to implement flexible Document-View architectures. The document will contain a signal to which each of the views can connect. The following Document class defines a simple text document that supports mulitple views. Note that it stores a single signal to which all of the views will be connected.

class Document
{
public:
    typedef boost::signals2::signal<void ()>  signal_t;

public:
    Document()
    {}

    /* Connect a slot to the signal which will be emitted whenever
      text is appended to the document. */
    boost::signals2::connection connect(const signal_t::slot_type &subscriber)
    {
        return m_sig.connect(subscriber);
    }

    void append(const char* s)
    {
        m_text += s;
        m_sig();
    }

    const std::string& getText() const
    {
        return m_text;
    }

private:
    signal_t    m_sig;
    std::string m_text;
};

Next, we can begin to define views. The following TextView class provides a simple view of the document text.

class TextView
{
public:
    TextView(Document& doc): m_document(doc)
    {
        m_connection = m_document.connect(boost::bind(&TextView::refresh, this));
    }

    ~TextView()
    {
        m_connection.disconnect();
    }

    void refresh() const
    {
        std::cout << "TextView: " << m_document.getText() << std::endl;
    }
private:
    Document&               m_document;
    boost::signals2::connection  m_connection;
};

Answer 5

Note: Most of the answers cover function pointers which is one possibility to achieve "callback" logic in C++, but as of today not the most favourable one I think.

What are callbacks(?) and why to use them(!)

A callback is a callable (see further down) accepted by a class or function, used to customize the current logic depending on that callback.

One reason to use callbacks is to write generic code which is independant from the logic in the called function and can be reused with different callbacks.

Many functions of the standard algorithms library <algorithm> use callbacks. For example the for_each algorithm applies an unary callback to every item in a range of iterators:

template<class InputIt, class UnaryFunction>
UnaryFunction for_each(InputIt first, InputIt last, UnaryFunction f)
{
  for (; first != last; ++first) {
    f(*first);
  }
  return f;
}

which can be used to first increment and then print a vector by passing appropriate callables for example:

std::vector<double> v{ 1.0, 2.2, 4.0, 5.5, 7.2 };
double r = 4.0;
std::for_each(v.begin(), v.end(), [&](double & v) { v += r; });
std::for_each(v.begin(), v.end(), [](double v) { std::cout << v << " "; });

which prints

5 6.2 8 9.5 11.2

Another application of callbacks is the notification of callers of certain events which enables a certain amount of static / compile time flexibility.

Personally, I use a local optimization library that uses two different callbacks:

• The first callback is called if a function value and the gradient based on a vector of input values is required (logic callback: function value determination / gradient derivation).
• The second callback is called once for each algorithm step and receives certain information about the convergence of the algorithm (notification callback).

Thus, the library designer is not in charge of deciding what happens with the information that is given to the programmer via the notification callback and he needn't worry about how to actually determine function values because they're provided by the logic callback. Getting those things right is a task due to the library user and keeps the library slim and more generic.

Furthermore, callbacks can enable dynamic runtime behaviour.

Imagine some kind of game engine class which has a function that is fired, each time the users presses a button on his keyboard and a set of functions that control your game behaviour. With callbacks you can (re)decide at runtime which action will be taken.

void player_jump();
void player_crouch();

class game_core
{
    std::array<void(*)(), total_num_keys> actions;
    // 
    void key_pressed(unsigned key_id)
    {
        if(actions[key_id]) actions[key_id]();
    }
    // update keybind from menu
    void update_keybind(unsigned key_id, void(*new_action)())
    {
        actions[key_id] = new_action;
    }
};

Here the function key_pressed uses the callbacks stored in actions to obtain the desired behaviour when a certain key is pressed. If the player chooses to change the button for jumping, the engine can call

game_core_instance.update_keybind(newly_selected_key, &player_jump);

and thus change the behaviour of a call to key_pressed (which the calls player_jump) once this button is pressed the next time ingame.

What are callables in C++(11)?

See C++ concepts: Callable on cppreference for a more formal description.

Callback functionality can be realized in several ways in C++(11) since several different things turn out to be callable*:

• Function pointers (including pointers to member functions)
• std::function objects
• Lambda expressions
• Bind expressions
• Function objects (classes with overloaded function call operator operator())

_{* Note: Pointer to data members are callable as well but no function is called at all.}

Several important ways to write callbacks in detail

• X.1 "Writing" a callback in this post means the syntax to declare and name the callback type.
• X.2 "Calling" a callback refers to the syntax to call those objects.
• X.3 "Using" a callback means the syntax when passing arguments to a function using a callback.

Note: As of C++17, a call like f(...) can be written as std::invoke(f, ...) which also handles the pointer to member case.

1. Function pointers

A function pointer is the 'simplest' (in terms of generality; in terms of readability arguably the worst) type a callback can have.

Let's have a simple function foo:

int foo (int x) { return 2+x; }

1.1 Writing a function pointer / type notation

A function pointer type has the notation

return_type (*)(parameter_type_1, parameter_type_2, parameter_type_3)
// i.e. a pointer to foo has the type:
int (*)(int)

where a named function pointer type will look like

return_type (* name) (parameter_type_1, parameter_type_2, parameter_type_3)

// i.e. f_int_t is a type: function pointer taking one int argument, returning int
typedef int (*f_int_t) (int); 

// foo_p is a pointer to function taking int returning int
// initialized by pointer to function foo taking int returning int
int (* foo_p)(int) = &foo; 
// can alternatively be written as 
f_int_t foo_p = &foo;

The using declaration gives us the option to make things a little bit more readable, since the typedef for f_int_t can also be written as:

using f_int_t = int(*)(int);

Where (at least for me) it is clearer that f_int_t is the new type alias and recognition of the function pointer type is also easier

And a declaration of a function using a callback of function pointer type will be:

// foobar having a callback argument named moo of type 
// pointer to function returning int taking int as its argument
int foobar (int x, int (*moo)(int));
// if f_int is the function pointer typedef from above we can also write foobar as:
int foobar (int x, f_int_t moo);

1.2 Callback call notation

The call notation follows the simple function call syntax:

int foobar (int x, int (*moo)(int))
{
    return x + moo(x); // function pointer moo called using argument x
}
// analog
int foobar (int x, f_int_t moo)
{
    return x + moo(x); // function pointer moo called using argument x
}

1.3 Callback use notation and compatible types

A callback function taking a function pointer can be called using function pointers.

Using a function that takes a function pointer callback is rather simple:

 int a = 5;
 int b = foobar(a, foo); // call foobar with pointer to foo as callback
 // can also be
 int b = foobar(a, &foo); // call foobar with pointer to foo as callback

1.4 Example

A function ca be written that doesn't rely on how the callback works:

void tranform_every_int(int * v, unsigned n, int (*fp)(int))
{
  for (unsigned i = 0; i < n; ++i)
  {
    v[i] = fp(v[i]);
  }
}

where possible callbacks could be

int double_int(int x) { return 2*x; }
int square_int(int x) { return x*x; }

used like

int a[5] = {1, 2, 3, 4, 5};
tranform_every_int(&a[0], 5, double_int);
// now a == {2, 4, 6, 8, 10};
tranform_every_int(&a[0], 5, square_int);
// now a == {4, 16, 36, 64, 100};

2. Pointer to member function

A pointer to member function (of some class C) is a special type of (and even more complex) function pointer which requires an object of type C to operate on.

struct C
{
    int y;
    int foo(int x) const { return x+y; }
};

2.1 Writing pointer to member function / type notation

A pointer to member function type for some class T has the notation

// can have more or less parameters
return_type (T::*)(parameter_type_1, parameter_type_2, parameter_type_3)
// i.e. a pointer to C::foo has the type
int (C::*) (int)

where a named pointer to member function will -in analogy to the function pointer- look like this:

return_type (T::* name) (parameter_type_1, parameter_type_2, parameter_type_3)

// i.e. a type `f_C_int` representing a pointer to member function of `C`
// taking int returning int is:
typedef int (C::* f_C_int_t) (int x); 

// The type of C_foo_p is a pointer to member function of C taking int returning int
// Its value is initialized by a pointer to foo of C
int (C::* C_foo_p)(int) = &C::foo;
// which can also be written using the typedef:
f_C_int_t C_foo_p = &C::foo;

Example: Declaring a function taking a pointer to member function callback as one of its arguments:

// C_foobar having an argument named moo of type pointer to member function of C
// where the callback returns int taking int as its argument
// also needs an object of type c
int C_foobar (int x, C const &c, int (C::*moo)(int));
// can equivalently declared using the typedef above:
int C_foobar (int x, C const &c, f_C_int_t moo);

2.2 Callback call notation

The pointer to member function of C can be invoked, with respect to an object of type C by using member access operations on the dereferenced pointer. Note: Parenthesis required!

int C_foobar (int x, C const &c, int (C::*moo)(int))
{
    return x + (c.*moo)(x); // function pointer moo called for object c using argument x
}
// analog
int C_foobar (int x, C const &c, f_C_int_t moo)
{
    return x + (c.*moo)(x); // function pointer moo called for object c using argument x
}

Note: If a pointer to C is available the syntax is equivalent (where the pointer to C must be dereferenced as well):

int C_foobar_2 (int x, C const * c, int (C::*meow)(int))
{
    if (!c) return x;
    // function pointer meow called for object *c using argument x
    return x + ((*c).*meow)(x); 
}
// or equivalent:
int C_foobar_2 (int x, C const * c, int (C::*meow)(int))
{
    if (!c) return x;
    // function pointer meow called for object *c using argument x
    return x + (c->*meow)(x); 
}

2.3 Callback use notation and compatible types

A callback function taking a member function pointer of class T can be called using a member function pointer of class T.

Using a function that takes a pointer to member function callback is -in analogy to function pointers- quite simple as well:

 C my_c{2}; // aggregate initialization
 int a = 5;
 int b = C_foobar(a, my_c, &C::foo); // call C_foobar with pointer to foo as its callback

3. std::function objects (header <functional>)

The std::function class is a polymorphic function wrapper to store, copy or invoke callables.

3.1 Writing a std::function object / type notation

The type of a std::function object storing a callable looks like:

std::function<return_type(parameter_type_1, parameter_type_2, parameter_type_3)>

// i.e. using the above function declaration of foo:
std::function<int(int)> stdf_foo = &foo;
// or C::foo:
std::function<int(const C&, int)> stdf_C_foo = &C::foo;

3.2 Callback call notation

The class std::function has operator() defined which can be used to invoke its target.

int stdf_foobar (int x, std::function<int(int)> moo)
{
    return x + moo(x); // std::function moo called
}
// or 
int stdf_C_foobar (int x, C const &c, std::function<int(C const &, int)> moo)
{
    return x + moo(c, x); // std::function moo called using c and x
}

3.3 Callback use notation and compatible types

The std::function callback is more generic than function pointers or pointer to member function since different types can be passed and implicitly converted into a std::function object.

3.3.1 Function pointers and pointers to member functions

A function pointer

int a = 2;
int b = stdf_foobar(a, &foo);
// b == 6 ( 2 + (2+2) )

or a pointer to member function

int a = 2;
C my_c{7}; // aggregate initialization
int b = stdf_C_foobar(a, c, &C::foo);
// b == 11 == ( 2 + (7+2) )

can be used.

3.3.2 Lambda expressions

An unnamed closure from a lambda expression can be stored in a std::function object:

int a = 2;
int c = 3;
int b = stdf_foobar(a, [c](int x) -> int { return 7+c*x; });
// b == 15 ==  a + (7*c*a) == 2 + (7+3*2)

3.3.3 std::bind expressions

The result of a std::bind expression can be passed. For example by binding parameters to a function pointer call:

int foo_2 (int x, int y) { return 9*x + y; }
using std::placeholders::_1;

int a = 2;
int b = stdf_foobar(a, std::bind(foo_2, _1, 3));
// b == 23 == 2 + ( 9*2 + 3 )
int c = stdf_foobar(a, std::bind(foo_2, 5, _1));
// c == 49 == 2 + ( 9*5 + 2 )

Where also objects can be bound as the object for the invocation of pointer to member functions:

int a = 2;
C const my_c{7}; // aggregate initialization
int b = stdf_foobar(a, std::bind(&C::foo, my_c, _1));
// b == 1 == 2 + ( 2 + 7 )

3.3.4 Function objects

Objects of classes having a proper operator() overload can be stored inside a std::function object, as well.

struct Meow
{
  int y = 0;
  Meow(int y_) : y(y_) {}
  int operator()(int x) { return y * x; }
};
int a = 11;
int b = stdf_foobar(a, Meow{8});
// b == 99 == 11 + ( 8 * 11 )

3.4 Example

Changing the function pointer example to use std::function

void stdf_tranform_every_int(int * v, unsigned n, std::function<int(int)> fp)
{
  for (unsigned i = 0; i < n; ++i)
  {
    v[i] = fp(v[i]);
  }
}

gives a whole lot more utility to that function because (see 3.3) we have more possibilities to use it:

// using function pointer still possible
int a[5] = {1, 2, 3, 4, 5};
stdf_tranform_every_int(&a[0], 5, double_int);
// now a == {2, 4, 6, 8, 10};

// use it without having to write another function by using a lambda
stdf_tranform_every_int(&a[0], 5, [](int x) -> int { return x/2; });
// now a == {1, 2, 3, 4, 5}; again

// use std::bind :
int nine_x_and_y (int x, int y) { return 9*x + y; }
using std::placeholders::_1;
// calls nine_x_and_y for every int in a with y being 4 every time
stdf_tranform_every_int(&a[0], 5, std::bind(nine_x_and_y, _1, 4));
// now a == {13, 22, 31, 40, 49};

4. Templated callback type

Using templates, the code calling the callback can be even more general than using std::function objects.

Note that templates are a compile-time feature and are a design tool for compile-time polymorphism. If runtime dynamic behaviour is to be achieved through callbacks, templates will help but they won't induce runtime dynamics.

4.1 Writing (type notations) and calling templated callbacks

Generalizing i.e. the std_ftransform_every_int code from above even further can be achieved by using templates:

template<class R, class T>
void stdf_transform_every_int_templ(int * v,
  unsigned const n, std::function<R(T)> fp)
{
  for (unsigned i = 0; i < n; ++i)
  {
    v[i] = fp(v[i]);
  }
}

with an even more general (as well as easiest) syntax for a callback type being a plain, to-be-deduced templated argument:

template<class F>
void transform_every_int_templ(int * v, 
  unsigned const n, F f)
{
  std::cout << "transform_every_int_templ<" 
    << type_name<F>() << ">\n";
  for (unsigned i = 0; i < n; ++i)
  {
    v[i] = f(v[i]);
  }
}

Note: The included output prints the type name deduced for templated type F. The implementation of type_name is given at the end of this post.

The most general implementation for the unary transformation of a range is part of the standard library, namely std::transform, which is also templated with respect to the iterated types.

template<class InputIt, class OutputIt, class UnaryOperation>
OutputIt transform(InputIt first1, InputIt last1, OutputIt d_first,
  UnaryOperation unary_op)
{
  while (first1 != last1) {
    *d_first++ = unary_op(*first1++);
  }
  return d_first;
}

4.2 Examples using templated callbacks and compatible types

The compatible types for the templated std::function callback method stdf_transform_every_int_templ are identical to the above mentioned types (see 3.4).

Using the templated version however, the signature of the used callback may change a little:

// Let
int foo (int x) { return 2+x; }
int muh (int const &x) { return 3+x; }
int & woof (int &x) { x *= 4; return x; }

int a[5] = {1, 2, 3, 4, 5};
stdf_transform_every_int_templ<int,int>(&a[0], 5, &foo);
// a == {3, 4, 5, 6, 7}
stdf_transform_every_int_templ<int, int const &>(&a[0], 5, &muh);
// a == {6, 7, 8, 9, 10}
stdf_transform_every_int_templ<int, int &>(&a[0], 5, &woof);

Note: std_ftransform_every_int (non templated version; see above) does work with foo but not using muh.

// Let
void print_int(int * p, unsigned const n)
{
  bool f{ true };
  for (unsigned i = 0; i < n; ++i)
  {
    std::cout << (f ? "" : " ") << p[i]; 
    f = false;
  }
  std::cout << "\n";
}

The plain templated parameter of transform_every_int_templ can be every possible callable type.

int a[5] = { 1, 2, 3, 4, 5 };
print_int(a, 5);
transform_every_int_templ(&a[0], 5, foo);
print_int(a, 5);
transform_every_int_templ(&a[0], 5, muh);
print_int(a, 5);
transform_every_int_templ(&a[0], 5, woof);
print_int(a, 5);
transform_every_int_templ(&a[0], 5, [](int x) -> int { return x + x + x; });
print_int(a, 5);
transform_every_int_templ(&a[0], 5, Meow{ 4 });
print_int(a, 5);
using std::placeholders::_1;
transform_every_int_templ(&a[0], 5, std::bind(foo_2, _1, 3));
print_int(a, 5);
transform_every_int_templ(&a[0], 5, std::function<int(int)>{&foo});
print_int(a, 5);

The above code prints:

1 2 3 4 5
transform_every_int_templ <int(*)(int)>
3 4 5 6 7
transform_every_int_templ <int(*)(int&)>
6 8 10 12 14
transform_every_int_templ <int& (*)(int&)>
9 11 13 15 17
transform_every_int_templ <main::{lambda(int)#1} >
27 33 39 45 51
transform_every_int_templ <Meow>
108 132 156 180 204
transform_every_int_templ <std::_Bind<int(*(std::_Placeholder<1>, int))(int, int)>>
975 1191 1407 1623 1839
transform_every_int_templ <std::function<int(int)>>
977 1193 1409 1625 1841

type_name implementation used above

#include <type_traits>
#include <typeinfo>
#include <string>
#include <memory>
#include <cxxabi.h>

template <class T>
std::string type_name()
{
  typedef typename std::remove_reference<T>::type TR;
  std::unique_ptr<char, void(*)(void*)> own
    (abi::__cxa_demangle(typeid(TR).name(), nullptr,
    nullptr, nullptr), std::free);
  std::string r = own != nullptr?own.get():typeid(TR).name();
  if (std::is_const<TR>::value)
    r += " const";
  if (std::is_volatile<TR>::value)
    r += " volatile";
  if (std::is_lvalue_reference<T>::value)
    r += " &";
  else if (std::is_rvalue_reference<T>::value)
    r += " &&";
  return r;
}

Answer 6

Scott Meyers gives a nice example:

class GameCharacter;
int defaultHealthCalc(const GameCharacter& gc);

class GameCharacter
{
public:
  typedef std::function<int (const GameCharacter&)> HealthCalcFunc;

  explicit GameCharacter(HealthCalcFunc hcf = defaultHealthCalc)
  : healthFunc(hcf)
  { }

  int healthValue() const { return healthFunc(*this); }

private:
  HealthCalcFunc healthFunc;
};

I think the example says it all.

std::function<> is the "modern" way of writing C++ callbacks.