views:

232

answers:

1

This question is related to my last one. I am trying to solve the problem using traits<T> and traits<T*>. Please consider the following code.

template<typename T>
struct traits
{
    typedef const T& const_reference;
};

template<typename T>
struct traits<T*>
{
    typedef const T const_reference;
};

template<typename T>
class test
{
public:   
    typedef typename traits<T>::const_reference const_reference;
    test() {}   
    const_reference value() const {
        return f;
    }
private:
    T f;
};

int main()
{
    const test<foo*> t;
    const foo* f = t.value(); // error here. cannot convert ‘const foo’ to ‘const foo*’ in initialization
    return 0;
}

So it looks like compiler is not considering the traits specialization for pointers and taking return type of value() as const foo rather than const foo*. What am I doing wrong here?

Any help would be great!

+2  A: 

The specialization is being used. traits<foo*>::const_reference is const foo. If you want it to be a pointer, use:

template<typename T>
struct traits<T*>
{
    typedef const T* const_reference;
};

With this, traits<foo*>::const_reference will be const foo*.

Note that the use of T in the traits<T*> specialization is completely separate from the T in the traits template. You could rename it:

template<typename U>
struct traits<U*>
{
    typedef const U* const_reference;
};

and you'll have the same specialization. It makes more sense if you've experience in functional programming.

To start, think of the template <typename ...> as introducing an abstraction, rather like a function abstracts out a value. It's like turning

sum = 0
for item in [1,2,3]:
    sum += item

into:

function sum(l):
    sum = 0
    for item in l:
        sum += item
    return sum

where l takes the place of [1,2,3]. We can call sums from another function that itself has a formal parameter named l:

function sumsq(l):
    return sum(map(lambda x: x*x, l))

sumsq's "l" has nothing to do with sum's "l".

With templates, we abstract type names rather than values. That is, we turn:

struct traits {
    typedef const double& const_reference;
};

into:

template <typename T>
struct traits {
    typedef const T& const_reference;
};

Now consider a non-template specialization:

template <>
struct traits<double*> {
    typedef const double* const_reference;
};

Here there are no template parameters for the specialization, but you can think of traits<double*> as applying a traits template to a double*. Abstract out the double and you have:

template <typename T>
struct traits<T*> {
    typedef const T* const_reference;
};

Here the T is a parameter for the specialization, not the base template.

outis
I feel dump :( But, T itself is a pointer (`foo*`). So specifying `T*` will lead into `T**`?
Appu
No. In the `traits<T*>` specialization, `T` isn't a pointer, `T*` is a pointer. That you use the same typename parameter is inconsequential.
outis
OK. Thanks. It makes sense now..
Appu