Template problem: Why does this work?

[Thomas Lorenz]

Hello,

im working on a problem related to smart pointers. If the following is
legal C++

A* a(new A());
B* b;
b = a;

I want to be able to do the following:

Ptr<A> a(new A());
Ptr<B> b;
b = a;

with equivalent semantics. Eventually, I came up with the following
"solution", involving a strange friend declaration:

template<class T>
class Ptr {
private:
T* m_pObject;
public:

// Constructors etc. removed for simplicity

template<class U>
inline friend void copy(Ptr<T>& to, Ptr<U>& from)
{
to.m_pObject = from.m_pObject;
};

// Curious trick
template<class U>
friend void copy(Ptr<U>&, Ptr<T>&);

template<class U>
inline Ptr(Ptr<U>& rhs)
: m_pObject(0)
{
copy(*this, rhs);
}

template<class U>
inline Ptr<T> operator=(Ptr<U>& rhs)
{
copy(*this,rhs);
return *this;
}
}

This works quite well using the MS 7.0 compiler. But I really don't know
why this "trick" enables copy() to access private members of Ptr<A> and
Ptr<B>. I've got a strange feeling this isn't supposed to work.

Oh, and the compiler produces "internal errors" if I try to repeat this
trick in Ptr with other functions...

Any ideas?
Thomas

[Victor Bazarov]

Thomas Lorenz wrote:

im working on a problem related to smart pointers. If the following is
legal C++

A* a(new A());
B* b;
b = a;

I want to be able to do the following:

Ptr<A> a(new A());
Ptr<B> b;
b = a;

with equivalent semantics. Eventually, I came up with the following
"solution", involving a strange friend declaration:

template<class T>
class Ptr {
private:
T* m_pObject;
public:

// Constructors etc. removed for simplicity

template<class U>
inline friend void copy(Ptr<T>& to, Ptr<U>& from)
{
to.m_pObject = from.m_pObject;
};

// Curious trick
template<class U>
friend void copy(Ptr<U>&, Ptr<T>&);

template<class U>
inline Ptr(Ptr<U>& rhs)
: m_pObject(0)
{
copy(*this, rhs);
}
I wonder what the need for this templatised c-tor is...

template<class U>
inline Ptr<T> operator=(Ptr<U>& rhs)
{
copy(*this,rhs);
return *this;
}
} ;
This works quite well using the MS 7.0 compiler. But I really don't know
why this "trick" enables copy() to access private members of Ptr<A> and
Ptr<B>. I've got a strange feeling this isn't supposed to work.
Looks OK (at a glance).
Oh, and the compiler produces "internal errors" if I try to repeat this
trick in Ptr with other functions...

Any ideas?

Upgrade to 7.1. Post the "other functions" on which the compiler chokes.

Victor

[Thomas Lorenz]

Victor Bazarov <v.********@comAcast.net> wrote in
news:Ln***************@newsread1.dllstx09.us.to.ve rio.net:

This works quite well using the MS 7.0 compiler. But I really don't
know why this "trick" enables copy() to access private members of
Ptr<A> and Ptr<B>. I've got a strange feeling this isn't supposed to
work.

Looks OK (at a glance).

Sure, and it works. But why?

Oh, and the compiler produces "internal errors" if I try to repeat
this trick in Ptr with other functions...

Any ideas?

Upgrade to 7.1. Post the "other functions" on which the compiler
chokes.

I tried adding equality operations with something like

template<class U>
inline bool equal(RefPtr<T>& lhs, RefPtr<U>& rhs) {
return lhs.m_pObject == rhs.m_pObject;
}

template<class U>
inline bool equal(RefPtr<U>& lhs, RefPtr<T>& rhs);

Thomas