VTK/Tutorials/SmartPointers

From KitwarePublic
< VTK‎ | Tutorials
Revision as of 14:37, 18 December 2012 by Burlen (talk | contribs) (fix bug in example, Take is static and thus cannot modify instances.)
Jump to navigationJump to search

Idea

The idea behind smart pointers is reference counting. If the object goes out of scope and it is not being used anywhere else, it will be deleted automatically. This is an important part of writing exception-safe code. With the proper use of smart pointers, you should never see an explicit Delete invocation in your code. Pretty 'smart', eh?

Note: vtkSmartPointer in combination with the internal reference counting of vtkObject reproduces much of the functionality of the std::tr1::shared_ptr template class introduced in the C++ Standard's Library Extensions Technical Report 1 (known as TR1). (This class first appeared as part of the Boost library.) If you're familiar with shared_ptr, then you already know most of how to use vtkSmartPointer. The principle difference is that, in VTK, the reference count is stored in the object, not the smart pointer. A vtkObject "owns itself", and the special vtkSmartPointer<>::New() method creates an object and smart pointer in one step, with the pointer holding the initial ownership of the object.

Usage

Creating an Object with a Smart Pointer

One way to create a VTK object is

vtkObject* MyObject = vtkObject::New();

This method, however, can (and likely will) lead to memory management issues at some point or another. You must manually delete the object

MyObject->Delete();

or you will have a memory leak. VTK's solution to this ever-annoying problem is the smart pointer. To use it, you must

#include <vtkSmartPointer.h>

Then you can create an object as follows:

vtkSmartPointer<vtkObject> MyObject = vtkSmartPointer<vtkObject>::New();

Note the special syntax on the right of the assignment. One almost never wants to assign a raw object pointer to a smart pointer, as is done here:

vtkSmartPointer<vtkObject> MyObject = vtkObject::New();

This is almost always a coding error, because the reference count is incremented, requiring an explicit Delete later.

Putting an Existing Object into a Smart Pointer

When putting a VTK object allocated elsewhere into a Smart pointer use the TakeReference method. This passes the object's self-ownership to the smart pointer:

vtkPolyData *NewMesh()
{
  vtkPolyData *pd = vtkPolyData::New();
  // build the mesh
  // ...
  return pd;
}

vtkSmartPointer<vtkPolyData> MyObject;
MyObject.TakeReference(NewMesh());

Because MyObject now owns the dataset, there is no need to invoke the Delete method.

The Take method should be used on the right hand side of smart pointer assignments. For example:

vtkSmartPointer<vtkPolyData> MyObject
  = vtkSmartPointer<vtkPolyData>::Take(NewMesh());

A word of caution, Take is a static member function and as such won't modify the smart pointer instance. The following illustrates this incorrect usage.

vtkSmartPointer<vtkPolyData> MyObject;
MyObject.Take(NewMesh());     // WRONG!! MyObject remains empty!
MyObject->GetNumberOfCells(); // SEGV!!

Getting an Object with a Smart Pointer

When not allocating memory for an object, you can still use smart pointers. Take this simple example:

vtkSmartPointer<vtkXMLPolyDataReader> Reader = vtkSmartPointer<vtkXMLPolyDataReader>::New();
vtkPolyData* pd = Reader->GetOutput();

vs

vtkSmartPointer<vtkPolyData> pd = Reader->GetOutput();

In the first case, when the reader object goes out of scope, the data is deleted. In the second case, by using a smart pointer we have incremented the data's reference count by 1, so the data will not be deleted until the reader AND the polydata object go out of scope.

Passing a Smart Pointer to a function

If you have a smart pointer, you can pass it to a function that accepts either a smart pointer or a normal pointer.

Smart pointer parameter:

void MyFunction(vtkSmartPointer<vtkPolyData> polydata)
{

}

...

vtkSmartPointer<vtkPolyData> polydata = vtkSmartPointer<vtkPolyData>::New();
MyFunction(polydata);

Normal pointer parameter:

void MyFunction(vtkPolyData* polydata)
{

}

...

vtkSmartPointer<vtkPolyData> polydata = vtkSmartPointer<vtkPolyData>::New();
MyFunction(polydata);

The reverse is also true. If you have a normal pointer, you can pass it to a function that accepts either a smart pointer or a normal pointer.

Smart pointer parameter:

void MyFunction(vtkSmartPointer<vtkPolyData> polydata)
{

}

...

vtkPolyData* polydata = vtkPolyData::New();
MyFunction(polydata);

Normal pointer parameter:

void MyFunction(vtkPolyData* polydata)
{

}

...

vtkPolyData* polydata = vtkPolyData::New();
MyFunction(polydata);

Returning a Smart Pointer

Correct

You should define a function like this:

vtkSmartPointer<vtkPolyData> MyFunction()
{
  vtkSmartPointer<vtkPolyData> myObject = vtkSmartPointer<vtkPolyData>::New();
  return myObject;
}

And call the function using:

vtkSmartPointer<vtkPolyData> MyPolydata = MyFunction();

The smart pointer in the function is copied to the smart pointer in the caller, so the reference count remains unchanged and the associated object is not deleted.

Incorrect

vtkPolyData* MyFunction()
{
  vtkSmartPointer<vtkPolyData> MyObject = vtkSmartPointer<vtkPolyData>::New();
  return MyObject;
}

vtkPolyData* MyPolydata = MyFunction();

In this case, the smart pointer is converted to a raw pointer before being returned to the caller. As the function exits, the smart pointer's reference count goes to zero and the actual object is deleted, leaving the raw pointer dangling, pointing at freed memory.

Using Smart Pointers as Class Member Variables

Smart pointers make class destructors simple, by automatically releasing ownership of shared objects.

Declare the pointer like this:

class MyClass
{
  vtkSmartPointer<vtkFloatArray> Distances;
};

You can initialize the smart pointer in your constructor using an initializer:

MyClass::MyClass()
: Distances(vtkSmartPointer<vtkFloatArray>::New())
{}

Or you can initialize it with an assignment statement:

MyClass::MyClass()
{
  Distances = vtkSmartPointer<vtkFloatArray>::New();
}

There is no need to explicitly Delete the object in your class destructor. By using smart pointers in your classes, your destructors become much simpler. You may find that you don't have to write a destructor at all, as the default destructor will Delete all your objects through the magic of the smart pointer.

Pitfalls

  • If you create an object and then change where it is pointing, the reference count will be incorrect. e.g.
vtkSmartPointer<vtkPolyData> Polydata = vtkSmartPointer<vtkPolyData>::New();
Polydata = Reader->GetOutput();

In this case, memory is allocated for Polydata, but then we change Polydata to point to the output of Reader rather than the memory we just allocated. Instead, we should have done simply:

vtkPolyData* Polydata = Reader->GetOutput();

It was not necessary to use a smart pointer because we did not actually create any new objects.

Example

Here is an example of equivalent operations with and without smart pointers:

SmartPointers.cpp

#include <vtkFloatArray.h>
#include <vtkSmartPointer.h>

void WithSmartPointers();
void WithoutSmartPointers();

int main(int argc, char *argv[])
{
  WithSmartPointers();
  WithoutSmartPointers();
	
  return 0;
}

void WithSmartPointers()
{
  vtkSmartPointer<vtkFloatArray> Distances = vtkSmartPointer<vtkFloatArray>::New();
}

void WithoutSmartPointers()
{
  vtkFloatArray* Distances = vtkFloatArray::New();
  Distances->Delete();
}

CMakeLists.txt

cmake_minimum_required(VERSION 2.6)

PROJECT(SmartPointers)

FIND_PACKAGE(VTK REQUIRED)
INCLUDE(${VTK_USE_FILE})

ADD_EXECUTABLE(SmartPointers SmartPointers.cpp)
TARGET_LINK_LIBRARIES(SmartPointers vtkHybrid)