Multiple Inheritance size problem

[Nicolas Hognon]

Hi

When coding classes with multiple inheritance, we faced a problem :

struct a
{
};

struct b
{
};

struct d : public a, public b
{
int m;
}

void foo()
{
int s = sizeof(d); // s == 8 !!!!!
}

We bumped into this problem with Visual .NET 2003, so we tried on
another compiler and platform (namely Metrowerks Codewarrior on MacOSX)
and the code showed just the same... So it must be normal, but why ??

(for the record, the extra dword is not initialized when we look the
raw memory in the debugger (0xcccccccc))

AGoTH

[Jeff Schwab]

Nicolas Hognon wrote:

Hi

When coding classes with multiple inheritance, we faced a problem :

struct a
{
};

struct b
{
};

struct d : public a, public b
{
int m;
}

void foo()
{
int s = sizeof(d); // s == 8 !!!!!
}

We bumped into this problem with Visual .NET 2003, so we tried on
another compiler and platform (namely Metrowerks Codewarrior on MacOSX)
and the code showed just the same... So it must be normal, but why ??

(for the record, the extra dword is not initialized when we look the
raw memory in the debugger (0xcccccccc))

AGoTH

Each of d's subobjects, the a, the b, and m, must have a unique address.
So, an instance of d must span at least enough address space to make
this possible, and to guarantee proper alignment of m. If this makes
you unhappy, Google for the Empty Base Optimization.

[Rob Williscroft]

Jeff Schwab wrote in news:8Z********************@comcast.com:

Each of d's subobjects, the a, the b, and m, must have a unique
address.
So, an instance of d must span at least enough address space to make
this possible, and to guarantee proper alignment of m. If this makes
you unhappy, Google for the Empty Base Optimization.

This is a matter of debate, Each subobject of the same type *must* have
a unique address.

i.e:

struct empty {};
struct derived : empty
{
empty value;
};

EBO can't work above as that would leave derived::value having the
same address as derived::empty.

VC7.1, and Borland 5.x (with EBO turned on) agrees with you.

g++ 3.2.3 and 3.4 (prerelese) and CBuilder X (preview) (EDG) agrees
with the above.

My test programme:

#include <iostream>
#include <ostream>

struct a
{
};

struct b
{
};

struct d : public a, public b
{
int m;
};
struct empty {};
struct derived : empty
{
empty value;
};

int main()
{
using namespace std;
d dd;
cerr << &dd.m << '\n';
cerr << static_cast< a *>( &dd ) << '\n';
cerr << static_cast< b *>( &dd ) << '\n';

cerr << sizeof(d) << '\n';
derived der;
cerr << &der.value << '\n';
cerr << static_cast< empty *>( &der ) << '\n';

cerr << sizeof(derived) << '\n';
}

Interetingly VC7.1 and Borland 5.x give derived::value the same address
as derived::empty (VC7.1 also gives sizeof( derived ) as 1). All the
others got it write.

Rob.
--
http://www.victim-prime.dsl.pipex.com/

[Rob Williscroft]

Rob Williscroft wrote in news:Xns9492A5067AB98ukcoREMOVEfreenetrtw@
195.129.110.130:

Interetingly VC7.1 and Borland 5.x give derived::value the same address
as derived::empty (VC7.1 also gives sizeof( derived ) as 1). All the
others got it write.

others got it right, (whoops No.2) apparently I need a grammer checker
as well as a compiler before posting.
Rob.
--
http://www.victim-prime.dsl.pipex.com/