Casting pointer to members

[Paolo Capriotti]

Compiling the following code (gcc 3.3.6), I get an invalid conversion
error.
Is there a valid reason why this conversion can't be performed?

---

class A {};
class B : public A {};

class T {
public:
B b;
};

int main() {
A T::*a = &T::b;
}

---

Paolo Capriotti

[Alf P. Steinbach]

* Paolo Capriotti:

Compiling the following code (gcc 3.3.6), I get an invalid conversion
error.
Is there a valid reason why this conversion can't be performed?

class A {};
class B : public A {};

class T {
public:
B b;
};

int main() {
A T::*a = &T::b;
}

For the purposes of this group, the explanation is that §3.9.2/3 tells us that
statements regarding pointers do not apply to pointers to members (except for
pointers to static members), so that the general B* -> A* conversion doesn't
apply, and then §4.11 does not list the conversion above for member pointers.

As to why, I think that's a defect in the standard.

I cannot think of any reason why the standard should forbid this conversion,
but then I don't understand e.g. how the dynamic type of an object that a
pointer to member is applied to, can not contain that member, as mentioned in
§5.5/4 (except for perhaps pure virtual functions, but that's covered
elsewhere and I think if it was that then the standard would just say so).

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[Greg]

Alf P. Steinbach wrote:

* Paolo Capriotti:

Compiling the following code (gcc 3.3.6), I get an invalid conversion
error.
Is there a valid reason why this conversion can't be performed?

class A {};
class B : public A {};

class T {
public:
B b;
};

int main() {
A T::*a = &T::b;
}

For the purposes of this group, the explanation is that §3.9.2/3 tells us that
statements regarding pointers do not apply to pointers to members (exceptfor
pointers to static members), so that the general B* -> A* conversion doesn't
apply, and then §4.11 does not list the conversion above for member pointers.

As to why, I think that's a defect in the standard.

I cannot think of any reason why the standard should forbid this conversion,
but then I don't understand e.g. how the dynamic type of an object that a
pointer to member is applied to, can not contain that member, as mentioned in
§5.5/4 (except for perhaps pure virtual functions, but that's covered
elsewhere and I think if it was that then the standard would just say so).

An object of class B may have stricter alignment requirements than an
object of class A. As a consequence, it is not safe to assume that an A
T::* member pointer can be safely converted to a B T::* member pointer.
They may in point to different locations in A and B respectively.

But as long as B is a subclass of A - and its alignment requirements
are no stricter than A's - then it is safe to use reinterpret_cast to
convert from A T::* to an B T::* (see 5.2/9).

Greg

[Alf P. Steinbach]

* Greg:

An object of class B may have stricter alignment requirements than an
object of class A. As a consequence, it is not safe to assume that an A
T::* member pointer can be safely converted to a B T::* member pointer.
They may in point to different locations in A and B respectively.

But as long as B is a subclass of A - and its alignment requirements
are no stricter than A's - then it is safe to use reinterpret_cast to
convert from A T::* to an B T::* (see 5.2/9).

I don't understand what you're saying, in particular the conversion you're
discussing seems to be opposite of the original example, and using the result
of a one-way reinterpret_cast is simply undefined (do you perhaps mean
in-practice safe?), but I see that alignment requirements are mentioned for
round-trip (there and back again) reinterpret_cast discussed in §5.2.10/9.

Could you give an example or elaboration?

Anyway, thanks for the reference to §5.2/9, I assume you meant §5.2.9/9?

--
A: Because it messes up the order in which people normally read text.
Q: Why is it such a bad thing?
A: Top-posting.
Q: What is the most annoying thing on usenet and in e-mail?

[hamburger]

class A {};
class B : public A {};

class T {
public:
B b;
};

int main() {
T t;
A*a = &t.b;
}