from devx.com site
Implementing the Singleton Design Pattern
By Danny Kalev
Singleton is probably the most widely used design pattern. Its intent is to ensure that a class has only one instance, and to provide a global point of access to it. There are many situations in which a singleton object is necessary: a GUI application must have a single mouse, an active modem needs one and only one telephone line, an operating system can only have one window manager, and a PC is connected to a single keyboard. I will show how to implement the singleton pattern in C++ and explain how you can optimize its design for single-threaded applications.
Using a global object ensures that the instance is easily accessible but it doesn't keep you from instantiating multiple objects—you can still create a local instance of the same class in addition to the global one. The Singleton pattern provides an elegant solution to this problem by making the class itself responsible for managing its sole instance. The sole instance is an ordinary object of its class, but that class is written so that only one instance can ever be created. This way, you guarantee that no other instance can be created. Furthermore, you provide a global point of access to that instance. The Singleton class hides the operation that creates the instance behind a static member function. This member function, traditionally called Instance(), returns a pointer to the sole instance. Here's a declaration of such a class:
static Singleton* Instance();
Singleton& operator= (const Singleton&);
static Singleton* pinstance;
Instead of Singleton, you can name your class Mouse, FileManager, Scheduler, etc., and declare additional members accordingly. To ensure that users can't create local instances of the class, Singleton's constructor, assignment operator, and copy constructor are declared protected. The class also declares a private static pointer to its instance, pinstance. When the static function Instance() is called for the first time, it creates the sole instance, assigns its address to pinstance, and returns that address. In every subsequent invocation, Instance() will merely return that address.
The class's implementation looks like this:
Singleton* Singleton::pinstance = 0;// initialize pointer
Singleton* Singleton::Instance ()
if (pinstance == 0) // is it the first call?
pinstance = new Singleton; // create sole instance
return pinstance; // address of sole instance
//... perform necessary instance initializations
Users access the sole instance through the Instance() member function exclusively. Any attempt to create an instance not through this function will fail because the class's constructor is protected. Instance() uses lazy initialization. This means the value it returns is created when the function is accessed for the first time. Note that this design is bullet-proof—all the following Instance() calls return a pointer to the same instance:
Singleton *p1 = Singleton::Instance();
Singleton *p2 = p1->Instance();
Singleton & ref = * Singleton::Instance();
Although our example uses a single instance, with minor modifications to the function Instance(), this design pattern permits a variable number of instances. For example, you can design a class that allows up to five instances.
Optimizing Singleton for Single-Threaded Applications
Singleton allocates its sole instance on the free-store using operator new. Because operator new is thread-safe, you can use this design pattern in multi-threaded applications. However, there's a fly in the ointment: you must destroy the instance manually by calling delete before the application terminates. Otherwise, not only are you causing a memory leak, but you also bring about undefined behavior because Singleton's destructor will never get called. Single-threaded applications can easily avoid this hassle by using a local static instance instead of a dynamically allocated one. Here's a slightly different implementation of Instance() that's suitable for single-threaded applications:
Singleton* Singleton::Instance ()
static Singleton inst;
The local static object inst is constructed when Instance() is called for the first time and remains alive until the application terminates. Note that the pointer pinstance is now redundant and can be removed from the class's declaration. Unlike dynamically allocated objects, static objects are destroyed automatically when the application terminates, so you shouldn't destroy the instance manually.