class hierarchy problem (do I need MI?)

[Bonj]

Hello
Can anyone assist with the following class hierarcy problem?
I have a series of window classes, the object model currently being as such:

Window
/ | \
/ | \
MDIClientWindow | TreeViewWindow
WndProcWindow
/ \
/ \
DualPaneWindow SplitterWindow

i.e., the base class is Window. It has three direct subclasses,
MDIClientWindow, WndProcWindow, and TreeViewWindow. WndProcWindow also has
two further classes deriving from it, DualPaneWindow and SplitterWindow.
(MDIClient- and TreeView- don't have their own WndProcs)
Now, I am having the following problem: I want to create a few pure virtual
methods on the base Window class, in order that I can create a function that
will take a Window* and call a certain method on a [Something]Window,
knowing that it will have it - e.g. Create()=0. This will make Window a pure
abstract class, this is fine. However, I have been scratching my head as to
the best way to implement this, because WndProcWindow is already an abstract
class - it has a pure virtual method, WndProc.
I'd like to make Window an abstract class, with some methods e.g. Create(),
as pure virtual methods. But this would mean that WndProcWindow would have
to implement them, and it wouldn't be appropriate for it to. Is there any
neat way of getting around this, like some way of telling the compiler that
WndProcWindow won't actually implement the Size method, but any
instantiatable class deriving from it must, because it also derives from
Window?
e.g.
// (Window.h)
void Create(LPCTSTR class, DWORD style);
{
....(generic CreateWindow code here)
}
void Create() = 0; // ???

//WndProcWindow.h
LRESULT WndProc(...) = 0;
void Create(LPCTSTR, DWORD);
void Create(); // ???

//SplitterWindow.cpp
SplitterWindow::Create()
{
WndProcWindow::Create(RegisterClass(... "splitterclass", ...), WS_CHILD |
WS_VISIBLE);
}

//TreeViewWindow.cpp
TreeViewWindow::Create()
{
WndProcWindow::Create(WC_TREEVIEW, WS_CHILD | WS_VISIBLE);
}

//WndProcWindow.cpp
WndProcWindow::Create(LPCTSTR class, DWORD style)
{
Window::Create(class, style);
}

WndProcWindow::Create() //??? help! what goes here?....
{
}


I thought of one solution - multiple inheritance. This would solve it - as I
could have an abstract class, say CreatableWindow, that had the pure virtual
methods e.g. Size(...) = 0, and then DualPane-, MDIClient-, TreeView- and
SplitterWindow could all multiply inherit from WndProcWindow or Window, and
then CreatableWindow aswell. But I've heard horror stories about multiple
inheritance - and that MS says it's one of those "if you really must"
things. A hunch tells me a multiply-inheriting class is like an inbred. Is
my hunch right, will it slow down my code or lead to spaghetti code without
any real benefit, or is it a little-discovered, rarely used technology
that's perfect for this kind of scenario?

On this particular project the porting to other compilers isn't a problem as
I've used loads of VS specific features anyway, but I would like to know if
it is to be avoided or not? Is it a performance hit?

Thanks for any guidance on how to accomplish the above or suggestions on how
I can better structure this without ripping the design apart completely! (It
seems good apart from this...)

(I hope this doesn't seem irrelevant to c.l.c++ as it's really about classes
not windows...)

 

[Malte Starostik]

Hello
Can anyone assist with the following class hierarcy problem?
I have a series of window classes, the object model currently being as such:

Window
/ | \
/ | \
MDIClientWindow | TreeViewWindow
WndProcWindow
/ \
/ \
DualPaneWindow SplitterWindow

i.e., the base class is Window. It has three direct subclasses,
MDIClientWindow, WndProcWindow, and TreeViewWindow. WndProcWindow also has
two further classes deriving from it, DualPaneWindow and SplitterWindow.
(MDIClient- and TreeView- don't have their own WndProcs)
Now, I am having the following problem: I want to create a few pure virtual
methods on the base Window class, in order that I can create a function that
will take a Window* and call a certain method on a [Something]Window,
knowing that it will have it - e.g. Create()=0. This will make Window a pure
abstract class, this is fine. However, I have been scratching my head as to
the best way to implement this, because WndProcWindow is already an abstract
class - it has a pure virtual method, WndProc.
I'd like to make Window an abstract class, with some methods e.g. Create(),
as pure virtual methods. But this would mean that WndProcWindow would have
to implement them, and it wouldn't be appropriate for it to. Is there any
neat way of getting around this, like some way of telling the compiler that
WndProcWindow won't actually implement the Size method, but any
instantiatable class deriving from it must, because it also derives from
Window?

What makes you think that this is a problem?

class Window
{
public:
virtual ~Window() {}
virtual void Create() = 0;
};

class WndProcWindow : public Window
{
public:
virtual void WndProc() = 0;
};

class SplitterWindow : public WndProcWindow
{
public:
virtual void Create() { /* stuff */ }
virtual void WndProc() { /* stuff */ }
};

is perfectly valid. No need for WndProcWindow to implement any of
Window's pure virtuals as long as you don't want to instantiate
WndProcWindow - which you can't do anyway because of the pure virtual
WndProc().

I thought of one solution - multiple inheritance. This would solve it - as I
could have an abstract class, say CreatableWindow, that had the pure virtual
methods e.g. Size(...) = 0, and then DualPane-, MDIClient-, TreeView- and
SplitterWindow could all multiply inherit from WndProcWindow or Window, and
then CreatableWindow aswell. But I've heard horror stories about multiple
inheritance - and that MS says it's one of those "if you really must"
things. A hunch tells me a multiply-inheriting class is like an inbred. Is
my hunch right, will it slow down my code or lead to spaghetti code without
any real benefit, or is it a little-discovered, rarely used technology
that's perfect for this kind of scenario?

*Never* trust any generalisations
MI is a powerful tool for some situations and should be avoided for
others, I personally don't subscribe to the "if you really must" theory
nor a general ban on it. I would say you shouldn't use it unless you
really understand it though - but that goes for just about any
programming technique ;-)
Either way, there isn't much of a reason for you to use MI unless I
totally misunderstood your problem - it would be the perfect option to
go for if there e.g. might be a subclass of TreeViewWindow that also has
a WndProc.

Cheers,
Malte

 

[Bonj]

Hello
Can anyone assist with the following class hierarcy problem?
I have a series of window classes, the object model currently being as
such:

Window
/ | \
/ | \
MDIClientWindow | TreeViewWindow
WndProcWindow
/ \
/ \
DualPaneWindow SplitterWindow

i.e., the base class is Window. It has three direct subclasses,
MDIClientWindow, WndProcWindow, and TreeViewWindow. WndProcWindow also
has
two further classes deriving from it, DualPaneWindow and SplitterWindow.
(MDIClient- and TreeView- don't have their own WndProcs)
Now, I am having the following problem: I want to create a few pure
virtual
methods on the base Window class, in order that I can create a function
that
will take a Window* and call a certain method on a [Something]Window,
knowing that it will have it - e.g. Create()=0. This will make Window a
pure
abstract class, this is fine. However, I have been scratching my head as
to
the best way to implement this, because WndProcWindow is already an
abstract
class - it has a pure virtual method, WndProc.
I'd like to make Window an abstract class, with some methods e.g.
Create(),
as pure virtual methods. But this would mean that WndProcWindow would
have
to implement them, and it wouldn't be appropriate for it to. Is there any
neat way of getting around this, like some way of telling the compiler
that
WndProcWindow won't actually implement the Size method, but any
instantiatable class deriving from it must, because it also derives from
Window?

What makes you think that this is a problem?

class Window
{
public:
virtual ~Window() {}
virtual void Create() = 0;
};

class WndProcWindow : public Window
{
public:
virtual void WndProc() = 0;
};

class SplitterWindow : public WndProcWindow
{
public:
virtual void Create() { /* stuff */ }
virtual void WndProc() { /* stuff */ }
};

is perfectly valid. No need for WndProcWindow to implement any of
Window's pure virtuals as long as you don't want to instantiate
WndProcWindow - which you can't do anyway because of the pure virtual
WndProc().

ah, yes - cheers, i didn't realise that you could do that, so that's pretty
much exactly what I've gone for.

*Never* trust any generalisations
MI is a powerful tool for some situations and should be avoided for
others, I personally don't subscribe to the "if you really must" theory
nor a general ban on it. I would say you shouldn't use it unless you
really understand it though - but that goes for just about any
programming technique ;-)

I've gone for a class 'ChildWindow' but this doesn't derive from the main
class, it is essentially just an interface, so it's like a 'V' shape but not
a diamond shape.
The SplitterWindow derives from this and Window, but I can see clearly that
only one 'copy' of each is being pulled in.

thanks

Either way, there isn't much of a reason for you to use MI unless I
totally misunderstood your problem - it would be the perfect option to
go for if there e.g. might be a subclass of TreeViewWindow that also has

I didn't know whether a TreeView could even have its own WndProc, but even
if it can I'm not going for that, I'm instead going to put something like
'HandleNotifier(code)=0' in the ChildWindow abstract class. However there
will be another class derived from TreeViewWindow - the treeview that's
specific to this particular application.