Simple reference counting: smart pointers
I would like to implement a simple reference counting using smart pointers. The variable pointer represents pointer to stored object, reference_count represents total count of copies of the object.
- if we initialize an object using NULL: reference_count = -1 else reference_count = 1
- copy ctor and operator = increment reference_count
- destructor decrement reference_count and if there is no other reference to pointed object performs its deletion.
Here is my code:
#ifndef smart_pointer_H
#define smart_pointer_H
template < typename T > class smart_pointer
{
private:
T* pointer;
int reference_count;
public:
smart_pointer() : pointer(0), reference_count(-1) {}
smart_pointer(T* p) : pointer(p)
{
if (p != NULL)
{
this->reference_count = 1;
}
else
{
this->reference_count = -1;
}
}
smart_pointer(const smart_pointer <T> & p) : pointer(p.pointer), reference_count(p.reference_count + 1) {}
bool operator == (const smart_pointer <T>& p) { return pointer == p.pointer; }
bool operator != (const smart_pointer <T>& p) { return pointer != p.pointer; }
~ smart_pointer()
{
if(-- reference_count == 0)
{
std::cout << "Destructing: " << '\n';
delete pointer;
}
}
T& operator * () { return *pointer; }
T* operator -> () { return pointer; }
smart_pointer <T> & operator = (const smart_pointer <T> & p)
{
if (this != &p)
{
if( -- reference_count == 0)
{
delete pointer;
}
pointer = p.pointer;
reference_count = p.reference_count + 1;
}
return *this;
}
};
Here is my testing code, class sample stores 2D point and two pointers to any other 2D points.
template < typename T >
class smart_pointer;
class Point
{
private:
double x, y;
smart_pointer <Point> p1;
smart_pointer <Point> p2;
public:
Point(double xx, double yy): x(xx), y(yy) {this-> p1 = NULL; this->p2 = NULL;}
Point(double xx, double yy, smart_pointer <Point> p1, smart_pointer <Point> p2): x(xx), y(yy) {this-> p1 = p1, this->p2 = p2; }
double getX(){ return x;}
double getY(){ return y;}
void setX(double xx) {this->x = xx;}
void setY(double yy) {this->y = yy;}
void setP1(smart_pointer <Point> p1) {this->p1 = p1;}
void setP2(smart_pointer <Point> p2) {this->p2 = p2;}
void print()
{
std::cout << "x= " << x << " y= " << y << '\n';
std::cout << "p1" << '\n';
if (p1 != NULL)
{
p1->print();
}
std::cout << "p2" << '\n';
if (p2 != NULL)
{
p2->print();
}
std::cout << '\n';
}
};
List of 2D points:
#include "Point.h"
class PointsList
{
private:
std::vector <smart_pointer <Point> > points;
public:
smart_pointer <Point> & operator [] ( int index ) {return points[index];}
public:
void push_back(smart_pointer <Point> p) {points.push_back(p);}
void erase(unsigned int index) {points.erase(points.begin() += index );}
void printPoints()
{
std::cout << "List of points" << '\n';
for (unsigned int i = 0; i < points.size(); i++)
{
points[i]->print();
}
}
};
Test code:
#include "Point.h"
#include "PointsList.h"
int main()
{
smart_pointer <Point> pb = NULL;
pb = (new Point(0,0));
smart_pointer <Point> p0(new Point(0,0));
p0->print();
smart_pointer <Point> p1(new Point(10,10));
p1->print();
smart_pointer <Point> p2(new Point(20,20));
p2->print();
smart_pointer <Point> p3(new Point(30,30));
p3->print();
smart_pointer <Point> pa(p3);
p0->setP1(p2);
p0->setP2(p3);
p0->print();
p0 = p1;
p0->print();
p0->print();
PointsList pl1;
pl1.push_back(p0);
pl1.push_back(p1);
PointsList pl2;
pl2.push_back(p2);
pl2.push_back(p3);
pl1.erase(0);
pl1.printPoints();
pl2.printPoints();
return 0;
}
Where are advanteges or disadvanteges of such solution? What about running speed for huge amount of data, casting, possible problems with inheritance, etc. Thanx for your help.
I had one more question to this example: Which type of the smart pointer (shared, scoped) would be for such a data structures the most suitable:
//Class with cross-references to points p1, p2
class PointTopo
{
private:
double x, y;
PointTopo * p1;
Point * p2;
public:
PointTopo(double xx, double yy): x(xx), y(yy) {this-> p1 = NULL; this->p2 = NULL;}
...
};
//Class with cross references: topological model for Delaunay triangulation
class Edge
{
private:
Point2D * start;
Edge *next;
Edge *previous;
Edge *twin;
...
};
Thanks for your help...
Answer
Your reference counting does not work.
If you copy or assign two smart pointers together they need to use the same location to perform the counting.
Currently each object keeps its own count and thus they may become out of sync.
smart_pointer<int> x(new x); // x.pointer: <good> x.reference_count: 1
{
smart_pointer<int> y; // y.pointer: NULL y.reference_count: -1
y = x; // x.pointer: <good> x.reference_count: 1
// y.pointer: <good> y.reference_count: 2
smart_pointer<int> z;
x = z; // x.pointer: NULL x.reference_count: 0 (BAD)
// z.pointer: NULL z.reference_count: -1
// y.pointer: <bad> (it was deleted by x) y.reference_count: 2
}
Edit:
Illustrate problem as requested in comments.
At the point. Where we have just created z. But not yet done x = z;
x { pointer: 0xabcd1234 reference_count: 1 }
y { pointer: 0xabcd1234 reference_count: 2 }
z { pointer: NULL reference_count: -1 }
// So here is your assignment operator.
// Doing the `x = z` we will walk through the following code.
//
smart_pointer <T> & operator = (const smart_pointer <T> & p)
{
if (this != &p)
{
// We get here.
// Left hand side is 'x' so this condition will be true.
if( -- reference_count == 0)
{
// Now we are deleting a pointer.
// That is held by 'x'
delete pointer;
// But 'y' is holding a pointer with the same value.
// Now y is holding a pointer to a deleted variable.
}
// Here we copy 'z' into 'x'
// Copy the pointer. That happens to be NULL.
pointer = p.pointer;
// Now we copy and increment the reference count.
// So 'x' has a value of 0 while 'z' has a value of -1.
// This also breaks the invariant on 'x' that NULL values should
// have a reference count of -1 (as X is NULL and ref-count is 0)
reference_count = p.reference_count + 1;
}
return *this;
}
If anybody tries to use 'y' now we have undefined behavior as it contains a pointer to memory that has been de-allocated.
Edit Classic (but overly simplistic smart pointer:
#include <vector>
template<typename T>
class SP
{
T* object;
size_t* count;
public:
SP(T* data)
try
// Use weird try around initializer list to catch new throwing.
// If it does we delete data to stop it from leaking.
:object(data)
,count(data ? new int(1) : NULL)
{ /* This is the constructor */}
catch(...)
{delete data;}
SP(): object(NULL), count(NULL) {}
//SP(T* data): object(data), count(new int(1)) {} // Lined up here so it look neat but implemented above to use weird try catch
SP(SP<T> const& rhs): object(rhs.object), count(rhs.count) {if (count) {++(*count);}}
SP<T>& operator=(SP<T> rhs) // Note implicit copy construction in rhs
{
// Using copy swap idiom for assignment.
// The copy is hidden because the parameter is pass by value.
this->swap(rhs);
return *this;
}
void swap(SP<T>& rhs) throw()
{
std::swap(object, rhs.object);
std::swap(count, rhs.count);
}
~SP()
{
if ((count) && (--(*count) == 0))
{
delete count;
delete object;
}
}
};