I have three objects; Action, Issue and Risk. These all contain a nunber of common variables/attributes (for example: Description, title, Due date, Raised by etc.) and some specific fields (risk has probability). The question is:
Should I create 3 separate
classes Action, Risk and Issue each
containing the repeat fields.
Create a parent class "Abstract_Item"
containing these fields and
operations on them and then have
Action, Risk and Issue subclass
Abstract_Item. This would adhere to
DRY principal.
My Perspective
Let's say, if you used inheritance. Over a period you have new attributes that are common to only Action and Issue but not Risk. How will you handle this? If you put them under parent then Risk is inheriting stuff that is irrelevant (Liskov Substituon Principle knocking?). If you put then in Action and Risk separately then you are breaking DRY, the initial reason why you started inheritance. Point is Inhertence for re-use is bad. If there is no "is-a" then better not use it and when you are in doubt then there is no real "is-a".
My Preference
There are other ways of achieving DRY as shown in below example code. With this addition of new properties my be another Common2, addition of new behavior is new CommonBehavior2 if they are not applicable to all 3 classes; if they are then just change existing Common and CommonBehavior
public class Common implements CommonBehavior
{
String Description;
String title;
public void f() {}
}
public interface CommonBehavior
{
void f();
}
public class Action implements CommonBehavior
{
private Common delegate;
public void f()
{
delegate.f();
}
}
Also look at my answer to a similar question with another practical example Design pattern to add a new class
Yes, adhering to DRY is usually a very good idea except if the classes have very, very different uses (i.e. both apples and cars may be red, still I wouldn't derive both of them from a base class called ProbablyRed). In your case, however, I'd definitely go for a base class since the implementations you describe (Action, Issue, Risk) all seem to be some kind of business rule with very similar semantics.
You seem to be answering this yourself. As you say, DRY.
The abstract parent class sounds like the way to go. It will also make it possible or easier (depending on your language) to implement and use functions which act on any of the three items. For example, you could have a "Get a list of all items raised by {user}" function.
Another facet may become visible if you look at the use cases, probably dealing with different subgroups of the properties.
If for example "label", "due date" and "raised" are used in a todo like application and other properties of "Action" and "Risk" will be prevalent when working on that task, you might consider an aggregation of lets say Task (label, due date,...) and Topic which is a polymorphic reference to things like Action, Issue or whatever will come up someday
I guess I have answered the same question here
When to create a class vs setting a boolean flag?
Don't subclass just because objects share some data or operations. Consider composition as the default way to follow DRY.
In your particular case, only create a parent class if your objects are actually related, i.e. there's a semantic "is a" relationship to the parent class. For example, if Action, Issue, and Risk are all Ticket objects.
Related
So I tend to favour composition over inheritance and I would like non-inheritance answers for this question.
There appears to be circumstances when using composition when there is some code in the superclass that requires a call to code in the subclass. This makes for unscaleable inheritance hierarchies which defeats the purpose of using composition in the first place. Here's a demonstration of the problem in C# (although this is a general oop question):
public interface IChemistry
{
void SeparateAtom(Atom atom);
void BreakBond(Bond bond);
}
public class BaseChemistry : IChemistry
{
public void SeparateAtom(Atom atom)
{
//possible extra logic here
for(int i=0;i < atom.BondCount;i++)
{
//maybe extra logic here etc.
BreakBond(atom.Bonds[i]);
}
}
public void BreakBond(Bond bond)
{
//do some bond breaking logic here
}
}
public class RealisticChemistry : IChemistry
{
private BaseChemistry base;
public RealisticChemistry(BaseChemistry base)
{
this.base = base;
}
public void SeparateAtom(Atom atom)
{
//subclass specific logic here perhaps
base.SeparateAtom(atom);
}
public void BreakBond(Bond bond)
{
//more subclass specific logic
base.BreakBond(bond);
}
}
As you can see with this design there is a glaring problem. When the subclass' SeparateAtom() method is called it executes some of it's own logic and then delegates the rest to the base class which will then call the BreakBond() method on the base class, not on the subclass.
There are various solutions I can think of for this and almost all of them have pretty substantial setbacks:
Copy and paste. The worst option in this case would be to simply copy the loop (and additional logic) within the base class' SeparateAtom() method, to the subclass' one. I don't feel that it is necessary to explain why copy and paste is not the best practice. Another option could be to package some of the extra logic around the loop into extra methods so that it's just the loop that is copied. But the calls to the additional methods are still copied, and breaking things up into multiple methods could break encapsulation. For example what if some of that logic is dependent on the specific context of SeparateAtom()and could lead to faulty data if called out-of-context by someone who does not know the code very well?
Listen to or observe bond breaking events in base class. This solution seems problematic to me because the way in which base class functionality should be extended becomes unclear. For example, without prior knowledge if one were to try to extend the class they might intuitively implement the design above and interpret the listener as optional, when it is in fact required if one wants to extend bond breaking behaviour.
Make the base class require a delegate. For example, the base class could require a reference to a IBondBreakDelegate which is called inside of BondBreak(). This has a similar problem to the listener approach in that the mixture of composition and other approaches makes the intended usage of the base class unclear. Also, even though now there is a delegate which is actually required, thus making the intended usage a little more clear, the base class can now no longer function on its own. Also if one needs to extend the hierarchy with an additional subclass (for example public class MoreRealistiChemistry etc.), how would one go about extending the delegated behaviour through composition?
Delegate everything instead of composition. I would prefer not to go down this route because when classes need extra functionality the amount of delegates needed increases (or the amount of methods in the delegates does). Also what if some of the delegated behaviour is optional? Then either there needs to be separate optional delegates for each behaviour that the subclass implements, or you end up with lots of empty method bodies in the subclass.
In general when I commit to a type of design, I would like to do so wholeheartedly. Of course in the real-world there are a ton of caveats. But I feel like this one must be so common that someone might know a good work-around. Any ideas?
(I cannot add a comment because of insufficient reputation, but I want to point out two things.)
First, your code does not compile because the classes do not implement IChemistry.
Second, 'favour composition over inheritance' is only a guideline and is not meant to be applied mindlessly. If the model that is under consideration for the solution requires either inheritance or composition, you should choose composition.
For this particular question, inheritance (or rather, specialisation) is the more sensible approach.
I have a question with regard to encapsulation. As I know, encapsulation enables to hide the implementation details using private/protected data members and provides public methods and properties to operate on the data. The idea here is to prevent the direct modification of the data members by the class consumers.
But I have a concern with the property getters or other public methods which return private/protected data members. For ex: if I have class like this
public class Inventory
{
private List<Guitar> guitars = new List<Guitar>();
public void AddGuitar(string serialnumber, string price)
{
Guitar guitar = new Guitar(serialnumber, price);
guitars.Add(guitar);
}
public List<Guitar> GetGuitars()
{
return guitars;
}
}
Now if the Inventory class consumer calls GetGuitars, he is going to get the list of guitars being maintained in the Inventory class. Now the consumer can modify the list, like delete/add/modify the items. For me it looks like we are not encapsulating. I think that I should be returning a copy of the Guitar list items in the GetGuitars(). What do you think?.
Is my understanding of the encapsulation right?.
Thanks
Encapsulating lists of objects can be achieved quite nicely by restricting access to them using a suitable interface.
I think you're right to control additions to your list via your AddGuitar method as you can exert control over what goes in. You can reinforce this design, IMHO, by altering GetGuitars to return IEnumerable instead of List.
This reduces the control the caller has on your list, whilst also being non-committal in returning an abstract type. This way your internal data structure can change without the public interface needing to also.
You are right. With a setter like that clients are able to modify the list. If adding a guitar requires some special handling, this is not desired. In this case you have two choices:
Return a copy of the list (as you already suggested).
Wrap it with ReadOnlyCollection within the getter.
Both cases should be documented in method description so that clients are not "surprised" when they attempt to modify the list externally.
if u want your List array cannot be modified, why u dont use AsReadOnly method: http://msdn.microsoft.com/en-us/library/e78dcd75.aspx
about encapsulation inside members are only writable and readable through the methods where members are not available from outside.
In terms of risk, it is indeed better if you return a copy of your list of make it unmodifiable (create a whole new unmodifiable list when you add a guitar, functional programming-style).
In terms of encapsulation, it would be better to get rid of the getGuitars() method and then Inventory class should offer the functionality associated with it ( for example, printInventoryReport() or whatever). This way, no client class needs to know at all how you store your guitars and you keep the related code into the Inventory class. The tradeoff is that this class gets bigger and every time you need something new from the guitar list you need to modify the Inventory.
I recommend a good article : http://www.javaworld.com/javaworld/jw-09-2003/jw-0905-toolbox.html
It was quite incendiary back in the day, but i think there's a lot of truth in there.
And if you stay with the getter, a small tip would be to choose if you need it to be a List or a Collection can do. Maybe even an Iterable! This way you tell as less as possible about your implementation, which results in better encapsulation.
I would agree that returning the list leaves something to be desired in terms on encapsulation. You may want to consider writing a getter for individual items, or possibly an iterator. The list seems like an implementation detail, so other classes really have no business accessing it directly.
There are (at least) two issues here.
The first is about hiding the implementation. You could change the "guitars" field to an array or a database but you could leave the signature of the methods AddGuitar and getGuitars unchanged so client code wouldn't break.
The second is about whether or not you want to return a defensive copy of the guitar list or not. Once you have the list of guitars do you want to add and delete elements? Since you have a method to add guitars I would assume not.
I saw this in someone's code and thought wow, that's an elegant way to solve this particular problem, but it probably violates good OO principles in an epic way.
In the constructor for a set of classes that are all derived from a common base class, he requires a reference to the instancing class to be passed. For example,
Foo Foo_i = new(this);
Then later on Foo would call methods in the instancing class to get information about itself and the other objects contained by the instancing class.
On the one hand, this simplifies a TON of code that models a 5-layer tree structure in hardware (agents plugged into ports on multiple switches, etc). On the other hand, these objects are pretty tightly coupled to each other in a way that seems pretty wrong, but I don't know enough about OOA&D to put my finger on it.
So, is this okay? Or is this the OO equivalent to a goto statement?
You shoud try to avoid mutual references (especially when implemeting containment) but oftentimes they are impossible to avoid. I.e. parent child relationship - children often need to know the parent and notify it if some events happen. If you really need to do that - opt for interfaces (or abstract classes in case of C++).
So you instancing class should implement some interface, and the instanciated class should know it only as interface - this will sigificantly reduce coupling. In some respect this approach is similar to nested listener class as it exposes only part of the class, but it is easier to maintain. Here is little C# example:
interface IParent
{
//some methods here
}
class Child
{
// child will know parent (instancing class) as interface only
private readonly IParent parent_;
public Child(IParent parent)
{
parent_ = parent;
}
}
class Parent : IParent
{
// IParent implementation and other methods here
}
This sounds like it could be violating the Law of Demeter, depending on how much Foo needs to know to fish around in the instancing class. Objects should preferably be loosely coupled. You'd rather not have one class need to know a lot about the structure of another class. One example I've heard a few times is that you wouldn't hand your wallet over to a store clerk and let him fish around inside. Your wallet is your business, and you'll find what you need to give the clerk and hand it over yourself. You can reorganize your wallet and nobody will be the wiser. Looser coupling makes testing easier. Foo should ideally be testable without needing to maintain a complex context.
I try and avoid this if I can just from an information hiding point of view. The less information a class has or needs the easier it is to test and verify. That being said, it does lead to more elegant solutions in some cases so if not doing it is horribly convoluted involving an awful lot of parameter passing then by all means go for it.
Java for example uses this a lot with inner classes:
public class Outer {
private class Inner {
public Inner() {
// has access to the members of Outer for the instance that instantiated it
}
}
}
In Java, I remember avoiding this once by subclassing certain Listeners and Adapters in my controller and adding those listeners and adapters to my subclasses.
In other words my controller was
class p {
private member x
private methods
private class q {
// methods referencing p's private members and methods
}
x.setListener(new q());
}
I think this is more loosely coupled, but I would also like some confirmation.
This design pattern can make a lot of sense in some situations. For example, iterators are always associated with a specific collection, so it makes sense for the iterator's constructor to require a collection.
You didn't provide a concrete example, but if the class reminds you of goto, it probably is a bad idea.
You said the new object must interrogate the instantiating object for information. Perhaps the class makes too many assumptions about its environment? If those assumptions complicate unit testing, debugging, or (non-hypothetical) code reuse, then you should consider refactoring.
But if the design saves developer time overall and you don't expect an unmaintainable beast in two years' time, the practice is completely acceptable from a practical standpoint.
I asked a similar question yesterday that was specific to a technology, but now I find myself wondering about the topic in the broad sense.
For simplicity's sake, we have two classes, A and B, where B is derived from A. B truly "is a" A, and all of the routines defined in A have the same meaning in B.
Let's say we want to display a list of As, some of which are actually Bs. As we traverse our list of As, if the current object is actually a B, we want to display some of Bs additional properties....or maybe we just want to color the Bs differently, but neither A nor B have any notion of "color" or "display stuff".
Solutions:
Make the A class semi-aware of B by basically including a method called isB() in A that returns false. B will override the method and return true. Display code would have a check like: if (currentA.isB()) B b = currentA;
Provide a display() method in A that B can override.... but then we start merging the UI and the model. I won't consider this unless there is some cool trick I'm not seeing.
Use instanceof to check if the current A object to be displayed is really a B.
Just add all the junk from B to A, even though it doesn't apply to A. Basically just contain a B (that does not inherit from A) in A and set it to null until it applies. This is somewhat attractive. This is similar to #1 I guess w/ composition over inheritance.
It seems like this particular problem should come up from time to time and have an obvious solution.
So I guess the question maybe really boils down to:
If I have a subclass that extends a base class by adding additional functionality (not just changing the existing behavior of the base class), am I doing something tragically wrong? It all seems to instantly fall apart as soon as we try to act on a collection of objects that may be A or B.
A variant of option 2 (or hybrid of 1 and 2) may make sense: after all, polymorphism is the standard solution to "Bs are As but need to behave differently in situation X." Agreed, a display() method would probably tie the model to the UI too closely, but presumably the different renderings you want at the UI level reflect semantic or behavioural differences at the model level. Could those be captured in a method? For example, instead of an outright getDisplayColour() method, could it be a getPriority() (for example) method, to which A and B return different values but it is still up to the UI to decide how to translate that into a colour?
Given your more general question, however, of "how can we handle additional behaviour that we can't or won't allow to be accessed polymorphically via the base class," for example if the base class isn't under our control, your options are probably option 3, the Visitor pattern or a helper class. In both cases you are effectively farming out the polymorphism to an external entity -- in option 3, the UI (e.g. the presenter or controller), which performs an instanceOf check and does different things depending on whether it's a B or not; in Visitor or the helper case, the new class. Given your example, Visitor is probably overkill (also, if you were not able/willing to change the base class to accommodate it, it wouldn't be possible to implement it I think), so I'd suggest a simple class called something like "renderer":
public abstract class Renderer {
public static Renderer Create(A obj) {
if (obj instanceOf B)
return new BRenderer();
else
return new ARenderer();
}
public abstract Color getColor();
}
// implementations of ARenderer and BRenderer per your UI logic
This encapsulates the run-time type checking and bundles the code up into reasonably well-defined classes with clear responsibilities, without the conceptual overhead of Visitor. (Per GrizzlyNyo's answer, though, if your hierarchy or function set is more complex than what you've shown here, Visitor could well be more appropriate, but many people find Visitor hard to get their heads around and I would tend to avoid it for simple situations -- but your mileage may vary.)
The answer given by itowlson covers pretty well most part of the question. I will now deal with the very last paragraph as simply as I can.
Inheritance should be implemented for reuse, for your derived class to be reused in old code, not for your class reusing parts of the base class (you can use aggregation for that).
From that standpoint, if you have a class that is to be used on new code with some new functionality, but should be used transparently as a former class, then inheritance is your solution. New code can use the new functionality and old code will seamlessly use your new objects.
While this is the general intention, there are some common pitfals, the line here is subtle and your question is about precisely that line. If you have a collection of objects of type base, that should be because those objects are meant to be used only with base's methods. They are 'bases', behave like bases.
Using techniques as 'instanceof' or downcasts (dynamic_cast<>() in C++) to detect the real runtime type is something that I would flag in a code review and only accept after having the programmer explain to great detail why any other option is worse than that solution. I would accept it, for example, in itowlson's answer under the premises that the information is not available with the given operations in base. That is, the base type does not have any method that would offer enough information for the caller to determine the color. And if it does not make sense to include such operation: besides the prepresentation color, are you going to perform any operation on the objects based on that same information? If logic depends on the real type, then the operation should be in base class to be overriden in derived classes. If that is not possible (the operation is new and only for some given subtypes) there should at least be an operation in the base to allow the caller to determine that a downcast will not fail. And then again, I would really require a sound reason for the caller code to require knowledge of the real type. Why does the user want to see it in different colors? Will the user perform different operations on each one of the types?
If you endup requiring to use code to bypass the type system, your design has a strange smell to it. Of course, never say never, but you can surely say: avoid depending on instanceof or downcasts for logic.
This looks like text book case for the Visitor design pattern (also known as "Double Dispatch").
See this answer for link to a thorough explanation on the Visitor and Composite patterns.
Can anyone think of any situation to use multiple inheritance? Every case I can think of can be solved by the method operator
AnotherClass() { return this->something.anotherClass; }
Most uses of full scale Multiple inheritance are for mixins. As an example:
class DraggableWindow : Window, Draggable { }
class SkinnableWindow : Window, Skinnable { }
class DraggableSkinnableWindow : Window, Draggable, Skinnable { }
etc...
In most cases, it's best to use multiple inheritance to do strictly interface inheritance.
class DraggableWindow : Window, IDraggable { }
Then you implement the IDraggable interface in your DraggableWindow class. It's WAY too hard to write good mixin classes.
The benefit of the MI approach (even if you are only using Interface MI) is that you can then treat all kinds of different Windows as Window objects, but have the flexibility to create things that would not be possible (or more difficult) with single inheritance.
For example, in many class frameworks you see something like this:
class Control { }
class Window : Control { }
class Textbox : Control { }
Now, suppose you wanted a Textbox with Window characteristics? Like being dragable, having a titlebar, etc... You could do something like this:
class WindowedTextbox : Control, IWindow, ITexbox { }
In the single inheritance model, you can't easily inherit from both Window and Textbox without having some problems with duplicate Control objects and other kinds of problems. You can also treat a WindowedTextbox as a Window, a Textbox, or a Control.
Also, to address your .anotherClass() idiom, .anotherClass() returns a different object, while multiple inheritance allows the same object to be used for different purposes.
I find multiple inheritance particularly useful when using mixin classes.
As stated in Wikipedia:
In object-oriented programming
languages, a mixin is a class that
provides a certain functionality to be
inherited by a subclass, but is not
meant to stand alone.
An example of how our product uses mixin classes is for configuration save and restore purposes. There is an abstract mixin class which defines a set of pure virtual methods. Any class which is saveable inherits from the save/restore mixin class which automatically gives them the appropriate save/restore functionality.
But they may also inherit from other classes as part of their normal class structure, so it is quite common for these classes to use multiple inheritance in this respect.
An example of multiple inheritance:
class Animal
{
virtual void KeepCool() const = 0;
}
class Vertebrate
{
virtual void BendSpine() { };
}
class Dog : public Animal, public Vertebrate
{
void KeepCool() { Pant(); }
}
What is most important when doing any form of public inheritance (single or multiple) is to respect the is a relationship. A class should only inherit from one or more classes if it "is" one of those objects. If it simply "contains" one of those objects, aggregation or composition should be used instead.
The example above is well structured because a dog is an animal, and also a vertebrate.
Most people use multiple-inheritance in the context of applying multiple interfaces to a class. This is the approach Java and C#, among others, enforce.
C++ allows you to apply multiple base classes fairly freely, in an is-a relationship between types. So, you can treat a derived object like any of its base classes.
Another use, as LeopardSkinPillBoxHat points out, is in mix-ins. An excellent example of this is the Loki library, from Andrei Alexandrescu's book Modern C++ Design. He uses what he terms policy classes that specify the behavior or the requirements of a given class through inheritance.
Yet another use is one that simplifies a modular approach that allows API-independence through the use of sister-class delegation in the oft-dreaded diamond hierarchy.
The uses for MI are many. The potential for abuse is even greater.
Java has interfaces. C++ has not.
Therefore, multiple inheritance can be used to emulate the interface feature.
If you're a C# and Java programmer, every time you use a class that extends a base class but also implements a few interfaces, you are sort of admitting multiple inheritance can be useful in some situations.
I think it would be most useful for boilerplate code. For example, the IDisposable pattern is exactly the same for all classes in .NET. So why re-type that code over and over again?
Another example is ICollection. The vast majority of the interface methods are implemented exactly the same. There are only a couple of methods that are actually unique to your class.
Unfortunately multiple-inheritance is very easy to abuse. People will quickly start doing silly things like LabelPrinter class inherit from their TcpIpConnector class instead of merely contain it.
One case I worked on recently involved network enabled label printers. We need to print labels, so we have a class LabelPrinter. This class has virtual calls for printing several different labels. I also have a generic class for TCP/IP connected things, which can connect, send and receive.
So, when I needed to implement a printer, it inherited from both the LabelPrinter class and the TcpIpConnector class.
I think fmsf example is a bad idea. A car is not a tire or an engine. You should be using composition for that.
MI (of implementation or interface) can be used to add functionality. These are often called mixin classes.. Imagine you have a GUI. There is view class that handles drawing and a Drag&Drop class that handles dragging. If you have an object that does both you would have a class like
class DropTarget{
public void Drop(DropItem & itemBeingDropped);
...
}
class View{
public void Draw();
...
}
/* View you can drop items on */
class DropView:View,DropTarget{
}
It is true that composition of an interface (Java or C# like) plus forwarding to a helper can emulate many of the common uses of multiple inheritance (notably mixins). However this is done at the cost of that forwarding code being repeated (and violating DRY).
MI does open a number of difficult areas, and more recently some language designers have taken decisions that the potential pitfalls of MI outweigh the benefits.
Similarly one can argue against generics (heterogeneous containers do work, loops can be replaced with (tail) recursion) and almost any other feature of programming languages. Just because it is possible to work without a feature does not mean that that feature is valueless or cannot help to effectively express solutions.
A rich diversity of languages, and language families makes it easier for us as developers to pick good tools that solve the business problem at hand. My toolbox contains many items I rarely use, but on those occasions I do not want to treat everything as a nail.
An example of how our product uses mixin classes is for configuration save and restore purposes. There is an abstract mixin class which defines a set of pure virtual methods. Any class which is saveable inherits from the save/restore mixin class which automatically gives them the appropriate save/restore functionality.
This example doesn't really illustrate the usefulness of multiple inheritance. What being defined here is an INTERFACE. Multiple inheritance allows you to inherit behavior as well. Which is the point of mixins.
An example; because of a need to preserve backwards compatibility I have to implement my own serialization methods.
So every object gets a Read and Store method like this.
Public Sub Store(ByVal File As IBinaryWriter)
Public Sub Read(ByVal File As IBinaryReader)
I also want to be able to assign and clone object as well. So I would like this on every object.
Public Sub Assign(ByVal tObject As <Class_Name>)
Public Function Clone() As <Class_Name>
Now in VB6 I have this code repeated over and over again.
Public Assign(ByVal tObject As ObjectClass)
Me.State = tObject.State
End Sub
Public Function Clone() As ObjectClass
Dim O As ObjectClass
Set O = New ObjectClass
O.State = Me.State
Set Clone = 0
End Function
Public Property Get State() As Variant
StateManager.Clear
Me.Store StateManager
State = StateManager.Data
End Property
Public Property Let State(ByVal RHS As Variant)
StateManager.Data = RHS
Me.Read StateManager
End Property
Note that Statemanager is a stream that read and stores byte arrays.
This code is repeated dozens of times.
Now in .NET i am able to get around this by using a combination of generics and inheritance. My object under the .NET version get Assign, Clone, and State when they inherit from MyAppBaseObject. But I don't like the fact that every object inherits from MyAppBaseObject.
I rather just mix in the the Assign Clone interface AND BEHAVIOR. Better yet mix in separately the Read and Store interface then being able to mix in Assign and Clone. It would be cleaner code in my opinion.
But the times where I reuse behavior are DWARFED by the time I use Interface. This is because the goal of most object hierarchies are NOT about reusing behavior but precisely defining the relationship between different objects. Which interfaces are designed for. So while it would be nice that C# (or VB.NET) had some ability to do this it isn't a show stopper in my opinion.
The whole reason that this is even an issue that that C++ fumbled the ball at first when it came to the interface vs inheritance issue. When OOP debuted everybody thought that behavior reuse was the priority. But this proved to be a chimera and only useful for specific circumstances, like making a UI framework.
Later the idea of mixins (and other related concepts in aspect oriented programming) were developed. Multiple inheritance was found useful in creating mix-ins. But C# was developed just before this was widely recognized. Likely an alternative syntax will be developed to do this.
I suspect that in C++, MI is best use as part of a framework (the mix-in classes previously discussed). The only thing I know for sure is that every time I've tried to use it in my apps, I've ended up regretting the choice, and often tearing it out and replacing it with generated code.
MI is one more of those 'use it if you REALLY need it, but make sure you REALLY need it' tools.
The following example is mostly something I see often in C++: sometimes it may be necessary due to utility classes that you need but because of their design cannot be used through composition (at least not efficiently or without making the code even messier than falling back on mult. inheritance). A good example is you have an abstract base class A and a derived class B, and B also needs to be a kind of serializable class, so it has to derive from, let's say, another abstract class called Serializable. It's possible to avoid MI, but if Serializable only contains a few virtual methods and needs deep access to the private members of B, then it may be worth muddying the inheritance tree just to avoid making friend declarations and giving away access to B's internals to some helper composition class.
I had to use it today, actually...
Here was my situation - I had a domain model represented in memory where an A contained zero or more Bs(represented in an array), each B has zero or more Cs, and Cs to Ds. I couldn't change the fact that they were arrays (the source for these arrays were from automatically generated code from the build process). Each instance needed to keep track of which index in the parent array they belonged in. They also needed to keep track of the instance of their parent (too much detail as to why). I wrote something like this (there was more to it, and this is not syntactically correct, it's just an example):
class Parent
{
add(Child c)
{
children.add(c);
c.index = children.Count-1;
c.parent = this;
}
Collection<Child> children
}
class Child
{
Parent p;
int index;
}
Then, for the domain types, I did this:
class A : Parent
class B : Parent, Child
class C : Parent, Child
class D : Child
The actually implementation was in C# with interfaces and generics, and I couldn't do the multiple inheritance like I would have if the language supported it (some copy paste had to be done). So, I thought I'd search SO to see what people think of multiple inheritance, and I got your question ;)
I couldn't use your solution of the .anotherClass, because of the implementation of add for Parent (references this - and I wanted this to not be some other class).
It got worse because the generated code had A subclass something else that was neither a parent or a child...more copy paste.