Multiple inheritance constructor/destructor design question

[Joseph Turian]

I have a global method "update_var(dou ble v);", which updates the value
of global variable 'var'.
I have a class Foo, which would like to know when 'var' is updated.

Let's say I make an ABC:
class Notify {
virtual void notify_that_var _has_updated() = 0;
};

and then I use MI to make Foo inherit from Notify.
Finally, I make a global method:
notify_object(N otify* n);
that will keep track of all Notify objects that want to be told when
update_var is called.

Can I make it that constructing Notify will automatically call
'notify_object' , and destructing Notify will automatically call
'denotify_objec t'? This will prevent dangling pointers. If so, how do I
make sure that any deriving class (e.g. Foo) will also call Notify's
constructors and destructors?

[Joseph Turian]

One followup question:

If I write:
Notify::Notify( ) {
notify_object(t his);
}
and this constructor is invoked from a derived class (Foo object), then
will the 'this' pointer be valid for the derived object? i.e. will
this->notify_that_va r_has_updated() do The Right Thing and invoke
Foo::notify_tha t_var_has_updat ed()?

Thanks,
Joseph

[=?ISO-8859-15?Q?Juli=E1n?= Albo]

Joseph Turian wrote:

Can I make it that constructing Notify will automatically call
'notify_object' , and destructing Notify will automatically call
'denotify_objec t'?

If you want to do something in the construction and some other thing in the
destruction, you just need to do it in the constructor and in the
destructor, respectively.

--
Salu2

[Ondra Holub]

Joseph Turian napsal:

One followup question:

If I write:
Notify::Notify( ) {
notify_object(t his);
}
and this constructor is invoked from a derived class (Foo object), then
will the 'this' pointer be valid for the derived object? i.e. will
this->notify_that_va r_has_updated() do The Right Thing and invoke
Foo::notify_tha t_var_has_updat ed()?

Thanks,
Joseph

Yes, it will be valid. This has type Notify* and that is all, what
notify_object needs to know, because notify_that_var _has_updated is
virtual.

[Sylvester Hesp]

"Joseph Turian" <tu****@gmail.c omwrote in message
news:11******** **************@ a75g2000cwd.goo glegroups.com.. .

One followup question:

If I write:
Notify::Notify( ) {
notify_object(t his);
}
and this constructor is invoked from a derived class (Foo object), then
will the 'this' pointer be valid for the derived object? i.e. will
this->notify_that_va r_has_updated() do The Right Thing and invoke
Foo::notify_tha t_var_has_updat ed()?

Only until _after_ Foo::Foo has ran. If your notify_object() calls
notify_that_var _has_updated() through the passed pointer, this will result
in undefined behaviour because the type of the object is still merely a
Notify, not a Foo (and therefore hasn't a valid
notify_that_var _has_updated() implementation) .

Regarding your earlier question:

Can I make it that constructing Notify will automatically call
'notify_object' , and destructing Notify will automatically call
'denotify_objec t'? This will prevent dangling pointers. If so, how do I
make sure that any deriving class (e.g. Foo) will also call Notify's
constructors and destructors?

A derived class _always_ calls a ctor and dtor of the base class if they
exist (whether they're auto-generated or not). If you don't specify a base
ctor in the initializer list of the derived ctor, the default ctor of the
base is invoked. So yes, you can safely do what you want to do (but keep in
mind my earlier statement about calling virtual functions during
construction) :).

- Sylvester

[Grizlyk]

Joseph Turian wrote:

Notify::Notify( ) {
notify_object(t his);
}
and this constructor is invoked from a derived class (Foo object), then
will the 'this' pointer be valid for the derived object? i.e. will
this->notify_that_va r_has_updated() do The Right Thing and invoke
Foo::notify_tha t_var_has_updat ed()?

If you are mix Notify() as MI in one place in base class, all derived
can overload the method as in ordinary inheritance and if the method is
virtual, correct "this" adjustment will be done (if needed)
automaticaly, so your overloaded Foo::notify_tha t_var_has_updat ed()
will get correct "this" of "Foo" class.

The trouble in MI only if you want to do reinterprete_ca st<from base
to derived, it can be failed without dynamic_cast<>.

Take example :)

#include <stdio.h>

class Base;
static Base *p=0;

class Base
{
int a;
public:
virtual void method(){ printf("%p: Base::method\n" ,this); }
Base(){ printf("%p: Base::\n",this) ; p=this;}
};

class Base2
{
int a;
public:
virtual void method2(){ printf("%p: Base::method2\n ",this); }
Base2(){ printf("%p: Base2::\n",this ); }
};

class Derived: public Base2, public Base
{
public:
void method(){ printf("%p: Derived::method \n",this); }
Derived(){ printf("%p: Derived::\n",th is); }
};

int main()
{
Derived d;

p->method();
p->Base::method() ;
}