Velocity Reviews (http://www.velocityreviews.com/forums/index.php)
-   C++ (http://www.velocityreviews.com/forums/f39-c.html)
-   -   Template function specialization (http://www.velocityreviews.com/forums/t948819-template-function-specialization.html)

 davidkevin@o2.pl 07-31-2012 12:38 PM

Template function specialization

Hi,
Stroustrup has shown a following sample of the template function specialization in his book "The C++ Programming Language":

template<class T> class Vector {
/*...*/
void swap(Vector&);
/*...*/
}

template<class T> void swap(T&x, T& y)
{
T t = x;
x = y;
y = t;
}

Here he has stated that:
code which is quoted above will work but will be ineffective if executed for a vector of vectors. A reason is that function swap() replaces one vectorwith other by copying elements. An object whose type is Vector has enough data to give an indirect access to elements. So it is possible to make a new design of definition such that representations will be swapped. To make it possible to deal with the representation, in class Vector there is a method swap:

template<class T> void Vector<T>::swap(Vector& a)
{
swap(v, a.v);
swap(size, a.size);
}

Then he has said that:
It is possible to use a method swap to define a specialization of general swap() function.

template<class T> void swap(Vector<T>& a, Vector<T>& b)
{
a.swap(b);
}

I am sure I am missing something but I can't see how above code make it possible to achieve a better efficiency when swap() will be called for a vector of vectors.

The last piece of code makes that if swap will be executed for parameters which are both Vectors then a special version of that method will be called (a.swap(b) - that is method which is defined in Vector). But when we look how Vector<T>::Swap is defined, we can see that it simply calls that versionof swap which was defined in the general template.

Where is a gap in the above reasoning?

Greetings.

 Victor Bazarov 07-31-2012 01:25 PM

Re: Template function specialization

On 7/31/2012 8:38 AM, davidkevin@o2.pl wrote:
> Stroustrup has shown a following sample of the template function
> specialization in his book "The C++ Programming Language":
>
> template<class T> class Vector {
> /*...*/
> void swap(Vector&);
> /*...*/
> }
>
> template<class T> void swap(T&x, T& y)
> {
> T t = x;
> x = y;
> y = t;
> }
>
> Here he has stated that:
> code which is quoted above will work but will be ineffective if
> executed for a vector of vectors. A reason is that function swap()
> replaces one vector with other by copying elements. An object whose type
> is Vector has enough data to give an indirect access to elements. So it
> is possible to make a new design of definition such that representations
> will be swapped. To make it possible to deal with the representation, in
> class Vector there is a method swap:
>
> template<class T> void Vector<T>::swap(Vector& a)
> {
> swap(v, a.v);
> swap(size, a.size);
> }
>
> Then he has said that:
> It is possible to use a method swap to define a specialization of general swap() function.
>
> template<class T> void swap(Vector<T>& a, Vector<T>& b)
> {
> a.swap(b);
> }
>
> I am sure I am missing something but I can't see how above code make
> it possible to achieve a better efficiency when swap() will be called
> for a vector of vectors.

It doesn't matter what the vector *of*. If v1 and v2 are both Vector<U>
for some given U type, then

swap(v1, v2)

will invoke the specialized version of the 'swap' instead of the general
one that uses a temporary variable.

> The last piece of code makes that if swap will be executed for
> parameters which are both Vectors then a special version of that
> method will be called (a.swap(b) - that is method which is defined in
> Vector). But when we look how Vector<T>::Swap is defined, we can see
> that it simply calls that version of swap which was defined in the
> general template.

Yes, but it doesn't create any temporary Vector objects, and doesn't use
assignment of those. It uses the swap for dealing with the built-in
types for the members 'v' and 'size' (probably a pointer and an integer)

> Where is a gap in the above reasoning?

Well, try without the specialization, and see how your program runs, how
many vectors it creates, how many copies it makes. Then provide the
specialization and see if there's any difference. You may see the
light, or you may not.

V
--

 Nobody 07-31-2012 02:07 PM

Re: Template function specialization

On Tue, 31 Jul 2012 05:38:58 -0700, davidkevin wrote:

> I am sure I am missing something but I can't see how above code make it
> possible to achieve a better efficiency when swap() will be called for a
> vector of vectors.
>
> The last piece of code makes that if swap will be executed for parameters
> which are both Vectors then a special version of that method will be
> called (a.swap(b) - that is method which is defined in Vector). But when
> we look how Vector<T>::Swap is defined, we can see that it simply calls
> that version of swap which was defined in the general template.

The generic swap() calls Vector<T>::operator=, which copies the vector's
data (deep copy), while the specialised version only copies the pointer
(shallow copy) and the size.

 davidkevin@o2.pl 07-31-2012 05:49 PM

Re: Template function specialization

Yes... it is so simple :). Thanks!

 Luca Risolia 07-31-2012 07:28 PM

Re: Template function specialization

On 31/07/2012 19:49, davidkevin@o2.pl wrote:
> Yes... it is so simple :). Thanks!

In C++11 it is efficient and simpler at the same time, since you no
longer need to implement template specializations to "move" the values
of your objects. In facts, given a type T supporting move semantics, you
can swap two l-values this way:

template <class T> void swap(T& a, T& b) {
T tmp = std::move(a);
a = std::move(b);
b = std::move(tmp);
}

 All times are GMT. The time now is 11:55 PM.