Calling a copy constructor with instance being created

Question

I am not asking about the logic of such a call, rather I'm interested in a difference of support b/w Visual C++ and GCC / Clang. Visual C++ will not allow a new instance of an object to be used as the parameter for its own copy constructor. GCC and Clang allow this. Considering that 'int i = i;' is allowed, I'm wondering whether Visual C++ has a bug.

class test {
private:
    test() {}
public:
    test(const test& t) {}
};

int main(void) {
    int i = i;
    test t(t); -- this line gives an error in Visual C++
    return 0;
}

`int i = i` and `test t(t)` are rather different forms. The latter is very clearly invalid (really, the former should be as well). — ssube, Sep 22 '11 at 16:45
int i = i is an assignment, no copy construction inoked there.. — stijn, Sep 22 '11 at 16:46
I'm curious (and I don't have MSVC handy). What happens with `int i2(i2);` and `test t2 = t2;`. — Robᵩ, Sep 22 '11 at 17:00
@Stargazer712: in principle it could make sense, for example consider `struct node { node *pnext; explicit node(node &nextnode) : pnext(&nextnode) {} };`. Then `node loop(loop)` would create an infinite list (which might mean "empty" if your other list functions interpret it that way. Not so much for singly-linked, but for doubly-linked lists you get a couple of nice features from storing them as a ring, with the same "head" node at both ends). — Steve Jessop, Sep 22 '11 at 17:12
@SteveJessop, I don't disagree with that, but there are probably clearer ways to do it (like in the default constructor). — riwalk, Sep 22 '11 at 17:17
@Stargazer712: yeah, and to be honest it's probably not a good idea even for my constructor to take a reference. It should be a pointer, to make it clearer what the parameter is for. — Steve Jessop, Sep 22 '11 at 17:50
@stijn: It's called initialization; more precisly, *dynamic* initialization. See this : [What is dynamic intialization of object in c++?](http://stackoverflow.com/questions/5945897/what-is-dynamic-intialization-of-object-in-c) — Nawaz, Sep 22 '11 at 18:23

score 2 · Accepted Answer · answered Sep 22 '11 at 17:13

To quote the C++ standard (3.3.2):

The point of declaration for a name is immediately after its complete declarator and before its initializer

In your first statement, the declarator ends after int i, and so the name i is available where it's used in the initializer (= i), so the statement is well-formed, but its behaviour is not defined.

In your second statement, the declarator ends after test t(t), and there is no initializer; the name t is not available where you use it, so the statement is ill-formed.

So the compiler is behaving correctly. I would hope it could be configured to give a warning about the first statement, but it's not required to reject it; it is required to reject the second, as you say it does.

score 0 · Answer 2 · answered Sep 22 '11 at 16:46

0

 test t(t); -- this line gives an error in Visual C++

Where have you defined t? The compiler is not physic!

answered Sep 22 '11 at 16:46

Ed Heal

59,252
17
87
127

Only perl can ignore physics. – ssube Sep 22 '11 at 16:57

Calling a copy constructor with instance being created

2 Answers2