"Container-of" in C++

[marcus hall]

Now that g++ is complaining about offset-of() applied to a class, it
raises the issue of how to implement a "container-of" concept in C++.

Specifically, I have a pointer to a class that is aggregated into an
outer class. I know the type of the outer class, and a pointer to the
member of that class. What is the preferred method of producing a pointer
to the outer class?

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

A _a;

I have &_a._b and I want to get &_a.

Is the only legal way now to store a "class A *" within class B?

Marcus Hall
(e-mail address removed)

 

[Victor Bazarov]

Now that g++ is complaining about offset-of() applied to a class, it
raises the issue of how to implement a "container-of" concept in C++.

Specifically, I have a pointer to a class that is aggregated into an
outer class. I know the type of the outer class, and a pointer to the
member of that class. What is the preferred method of producing a pointer
to the outer class?

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

A _a;

I have &_a._b and I want to get &_a.

Is the only legal way now to store a "class A *" within class B?

No. You may store A& there. Or any other value that would help you
identify the containing object. There is nothing automatic, of course.

V

 

[wkaras]

Now that g++ is complaining about offset-of() applied to a class, it
raises the issue of how to implement a "container-of" concept in C++.

Specifically, I have a pointer to a class that is aggregated into an
outer class. I know the type of the outer class, and a pointer to the
member of that class. What is the preferred method of producing a pointer
to the outer class?

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

A _a;

I have &_a._b and I want to get &_a.

Is the only legal way now to store a "class A *" within class B?

Marcus Hall
(e-mail address removed)

This:

(A *) (((char *) addrBinA) - (((char *) &(((A *) 0)->_b)) - ((char *)
0)))

Seems to be very portable, although I'm not sure the Standard guaratees
it
will always work, except perhaps with POD.

(BTW, I believe identifiers that start with _ are reserved for use by
the compiler and standard libraries.)

 

[Alipha]

This:

(A *) (((char *) addrBinA) - (((char *) &(((A *) 0)->_b)) - ((char *)
0)))

use reinterpret_cast so people know at a glance that the code is not
portable, however,

Seems to be very portable, although I'm not sure the Standard guaratees
it
will always work, except perhaps with POD.

is probably an accurate statement.

(BTW, I believe identifiers that start with _ are reserved for use by
the compiler and standard libraries.)

identifiers that begin with _ at global scope are not allowed.
therefore _b is valid and _a may be valid if that code is in a
namespace.

 

[verec]

This:

(A *) (((char *) addrBinA) - (((char *) &(((A *) 0)->_b)) - ((char *)
0)))

Just out of sheer curiosity, what platform/compiler does define

x - ((char *) 0)

as something other than x ?

 

[Victor Bazarov]

Just out of sheer curiosity, what platform/compiler does define

x - ((char *) 0)

as something other than x ?

If 'x' is a pointer, then it would seem that the result may actually
be the value of 'x' expressed as a ptrdiff_t, a different type.

V

 

[Guest]

Now that g++ is complaining about offset-of() applied to a class, it
raises the issue of how to implement a "container-of" concept in C++.

This:

(A *) (((char *) addrBinA) - (((char *) &(((A *) 0)->_b)) - ((char *)
0)))

Seems to be very portable, although I'm not sure the Standard guaratees
it
will always work, except perhaps with POD.

Yes, this is what I have been doing, but around gcc3.3 or so, it has started
complaining:

warning: invalid access to non-static data member ‘...' of NULL object
(perhaps the ‘offsetof’ macro was used incorrectly)

So, that raises my question on what is the "proper" way to get back to
the address of the containing object. I suppose that I could avoid
using "0" as a pointer, as in:

offset_of__b_within_A == (char *)&((A*)this)->_b - (char *)this

which avoids "dereferencing 0", but it seems even uglier.

Marcus Hall
(e-mail address removed)

 

[msalters]

marcus hall schreef:

Now that g++ is complaining about offset-of() applied to a class, it
raises the issue of how to implement a "container-of" concept in C++.

Specifically, I have a pointer to a class that is aggregated into an
outer class. I know the type of the outer class, and a pointer to the
member of that class.

In general, this may be impossible. Going from A to A.B may be
implemented
similar to dereferencing a B* in A. Obviously, that is a one-way
operation.
When you only have the B*, there is no way to get back the containing
A.
This may seem weird, but I could imagine something like this in the
presence of multiple/virtual inheritance, where the B* points to a base
class member.

Regards,
Michiel Salters