Re: [Dev-C++] Re: Design decision in C++

[Doug M. <dou...@al...>]

Per,

I understand now.  You all have been talking about taking a regular 
class and adding code to it to make it a singleton.  My expressed 
ambivalence is toward something a bit different - it's a Singleton 
utility, so to speak.  Below is what I am referring to.  The idea is 
that you implement the Singleton behavior once and for all, as a 
template, then any class can become a singleton by inheriting from it as

class Klass : public Singleton<Klass> {
...
}

It's in making it work in general, through inheritance, that the 
ugliness I spoke of is introduced.

--Doug

#ifndef __SINGLETON_H__INCLUDED__
#define __SINGLETON_H__INCLUDED__

#include <assert.h>

template <typename Type>
class Singleton
{
	protected:
		static Type * ms_pInstance;
		
	public:
	// Constructor
	// The Singleton construct takes care of assinging a
	// value to the
	// instance pointer.  Since this class' child may have
	// more than one
	// parent, we have to do some pointer math to figure out
	// the correct
	// address for the child class
	Singleton ()
	{
		assert (!ms_pInstance);
		int iOffset = (int)(Type *)1 -
			(int)(Singleton<Type> *)(Type *)1;
		ms_pInstance = (Type *)((int)this + iOffset);
	}
		
	// Destructor
	virtual ~Singleton ()
	{
		assert (ms_pInstance);
		ms_pInstance = 0;
	}
		
	// Instance
	// Returns a reference to the singleton
	static Type & Instance ()
	{
		assert (ms_pInstance);
		return (*ms_pInstance);
	}
		
	// InstancePtr
	// Returns a pointer to the singleton
	static Type * InstancePtr ()
	{
		assert (ms_pInstance);
		return ms_pInstance;
	}

         // HasBeenInstantiated
         // Returns true if the class has been instantiated
         static bool HasBeenInstantiated ()
         {
             return (NULL != ms_pInstance);
         }
};

template <typename Type> Type * Singleton<Type>::ms_pInstance = 0;

#endif // __SINGLETON_H__INCLUDED__

> I haven't looked at any Singleton<DataType> class doing any instance 
> pointer offset hack.
> 
> The sample code I sent for a singleton object was just a very fiew
> lines of manually added code to the class in question. 100% valid C++
> without any typecasts.
> 
> /Per W
> 
> On Thu, 11 Mar 2004, Doug MacDonald wrote:
> 
> 
>>> Per and Kevin,
>>> 
>>> Thanks for your advice.
>>> 
>>> My ambivalence wrt the Singleton<DataType> class is in part due
>>> to the instance pointer offset hack, which seems to introduce a
>>> risk of implementation dependency: are base and derived classes
>>> always included in a standard order with offsets that never vary
>>> from compiler to compiler?.  Further, the instance pointer
>>> adjustment involves pointer <-> int casts that are a bit ugly,
>>> despite the elegance of Singleton as seen from the outside.
>>> 
>>> But I take your points as to the limited desirability of fully
>>> static classes.
>>> 
>>> (I guess declaring a top-level static _instance_ of a class is a 
>>> different matter, and is useful, e.g., when there is no C++ main 
>>> routine, as was the case in a former project of mine at my last
>>> employer.)
>>> 
>>> --Doug