keeping private parts outside c++ headers: pure virtual base class vs pimpl - c++

I recently switched back from Java and Ruby to C++, and much to my surprise I have to recompile files that use the public interface when I change the method signature of a private method, because also the private parts are in the .h file.
I quickly came up with a solution that is, I guess, typical for a Java programmer: interfaces (= pure virtual base classes). For example:
BananaTree.h:
class Banana;
class BananaTree
{
public:
virtual Banana* getBanana(std::string const& name) = 0;
static BananaTree* create(std::string const& name);
};
BananaTree.cpp:
class BananaTreeImpl : public BananaTree
{
private:
string name;
Banana* findBanana(string const& name)
{
return //obtain banana, somehow;
}
public:
BananaTreeImpl(string name)
: name(name)
{}
virtual Banana* getBanana(string const& name)
{
return findBanana(name);
}
};
BananaTree* BananaTree::create(string const& name)
{
return new BananaTreeImpl(name);
}
The only hassle here, is that I can't use new, and must instead call BananaTree::create(). I do not think that that is really an problem, especially since I expect to be using factories a lot anyway.
Now, the wise men of C++ fame, however, came up with another solution, the pImpl idiom. With that, if I understand it correctly, my code would look like:
BananaTree.h:
class BananaTree
{
public:
Banana* addStep(std::string const& name);
private:
struct Impl;
shared_ptr<Impl> pimpl_;
};
BananaTree.cpp:
struct BananaTree::Impl
{
string name;
Banana* findBanana(string const& name)
{
return //obtain banana, somehow;
}
Banana* getBanana(string const& name)
{
return findBanana(name);
}
Impl(string const& name) : name(name) {}
}
BananaTree::BananaTree(string const& name)
: pimpl_(shared_ptr<Impl>(new Impl(name)))
{}
Banana* BananaTree::getBanana(string const& name)
{
return pimpl_->getBanana(name);
}
This would mean I have to implement a decorator-style forwarding method for every public method of BananaTree, in this case getBanana. This sounds like an added level of complexity and maintenance effort that I prefer not to require.
So, now for the question: What is wrong with the pure virtual class approach? Why is the pImpl approach so much better documented? Did I miss anything?

I can think of a few differences:
With the virtual base class you break some of the semantics people expect from well-behaved C++ classes:
I would expect (or require, even) the class to be instantiated on the stack, like this:
BananaTree myTree("somename");
otherwise, I lose RAII, and I have to manually start tracking allocations, which leads to a lot of headaches and memory leaks.
I also expect that to copy the class, I can simply do this
BananaTree tree2 = mytree;
unless of course, copying is disallowed by marking the copy constructor private, in which case that line won't even compile.
In the above cases, we obviously have the problem that your interface class doesn't really have meaningful constructors. But if I tried to use code such as the above examples, I'd also run afoul of a lot of slicing issues.
With polymorphic objects, you're generally required to hold pointers or references to the objects, to prevent slicing. As in my first point, this is generally not desirable, and makes memory management much harder.
Will a reader of your code understand that a BananaTree basically doesn't work, that he has to use BananaTree* or BananaTree& instead?
Basically, your interface just doesn't play that well with modern C++, where we prefer to
avoid pointers as much as possible, and
stack-allocate all objects to benefit form automatic lifetime management.
By the way, your virtual base class forgot the virtual destructor. That's a clear bug.
Finally, a simpler variant of pimpl that I sometimes use to cut down on the amount of boilerplate code is to give the "outer" object access to the data members of the inner object, so you avoid duplicating the interface. Either a function on the outer object just accesses the data it needs from the inner object directly, or it calls a helper function on the inner object, which has no equivalent on the outer object.
In your example, you could remove the function and Impl::getBanana, and instead implement BananaTree::getBanana like this:
Banana* BananaTree::getBanana(string const& name)
{
return pimpl_->findBanana(name);
}
then you only have to implement one getBanana function (in the BananaTree class), and one findBanana function (in the Impl class).

Actually, this is just a design decision to make. And even if you make the "wrong" decision, it's not that hard to switch.
pimpl is also used to provide ligthweight objects on stack or to present "copies" by referencing to the same implementation object.
The delegation-functions can be a hassle, but it's a minor issue (simple, so no real added complexity), especially with limited classes.
interfaces in C++ are typically more used in strategy-like ways where you expect to be able to choose implementations, although that is not required.

Related

Polymorphism and private data members

First of all I want to make it clear that this question relates to Scott Meyers' book Effective C++ (3rd edition) and specifically item 22: declare data-members private
I do understand it mostly and am trying to apply some of the things to my code to start practicing it. However I have a case where I am not sure how to solve it. Basically I have some abstract interfaces and some inheritance looking like this.
class abstractSystem {
...
protected:
AbstractThing *_thing;
}
class DerivedSystem : public AbstractSystem {
// inherits _thing so we can use it here.
}
However this is not coherent to item 22. I figured that it is better to have an interface to the base class for the derived classes and this works somewhat nice in many cases, but in this case since polymorphism is used to decide the _thing we would make a copy of it in a getter so that in the derived system any time we need to access it we need to copy it.
So I am guessing that's not great, and to be coherent to item 28: avoid returning "handles" object internals I can't seem to figure out how to do it without copying _thing:
class AbstractSystem {
protected:
AbstractThing thing() { return *_thing; }
private:
AbstractThing *_thing;
}
class DerivedSystem {
// now we need to use thing() to access _thing implying copy
}
Is this the way it must be done, and isn't it a bit hard on performance to copy (if done fairly often)?
I guess it might be my design that's wrong.
You could return a reference to the 'thing':
protected:
AbstractThing const& thing() { return *_thing; }
This would avoid copying the entire object.

C++ Vastly Different Derived Classes - Virtual methods? A cast?

I have a class hierarchy with lots of shared member functions in a base class, and also a large number of unique member functions in two derived classes:
class Scene {
public:
void Display();
void CreateDefaultShapes();
void AddShape(Shape myShape);
// lots more shared member functions
}
Class AnimatedScene : public Scene {
public:
void SetupAnimation();
void ChangeAnimationSpeed(float x, float y);
// lots of member functions unique to animation
}
Class ControllableScene : public Scene {
public:
void ConfigureControls();
void MoveCamera(float x, float y, float z);
void Rotate(float x, float y, float z);
// lots of member functions unique to a controllable scene
}
Not surprisingly, this doesn't work:
Scene myScene;
myScene.SetupAnimation();
What is the correct solution to this problem? Should I make all of the derived member functions virtual and add them to the base? Should I use a cast when calling SetupAnimation()? Is there a more clever design that solves this problem?
Note: I receive myScene from elsewhere and can't simply declare it as AnimatedScene when I instantiate it.
Edit: I've added a couple more member functions to illustrate the point. A small handful of initialization functions definitely lend themselves to simply using virtual.
You can cast it, preferably using static_cast. The least preferable option. If you are casting things, it usually means your design needs more thought
If you have a particular function/class that needs one or the other, declare the input as the type you need, which more accurately communicates the requirements of the function or class
If the function needs to be generic, and those methods don't require any input, then you could define a virtual method in the parent class, say init, which in the derived classes call the correct methods to set up the instance.
I have a similar problem in my compiler project, where the AST (Abstract Syntax Tree) is constructed from the statements, so while(x != 0) { if (a == b) x = 0; } would construct a whileAST with a binaryExpr inside it, then a blockAST with the ifAST, and so on. Each of these have some common properties, and a lot of things that only apply when you actually do something specific to that part. Most of the time, that is solved by calling a generic (virtual) member function.
However, SOMETIMES you need to do very specific things. There are two ways to do that:
use dynamic_cast (or typeid + reinterpret_cast or static cast).
Set up dozens of virtual member functions, which mostly are completely useless (doesn't do anything or return an "can't do that" indication of some sort)
In my case, I choose the first one. It shouldn't be the common case, but sometimes it is indeed the right thing to do.
So in this case, you'd do something like:
AnimatedScene *animScene = dynamic_cast<AnimatedScene*>(&scene);
if (!animScene)
{
... do something else, since it's not an AnimatedScene ...
}
animScene->SetupAnimation();
I am not yet able to comment, which is what I really wanted to do, but I am also interested in figuring this out as well.
A few months ago I had a similar problem. What I can tell you is that you can use typeid operator to figure out what type the object is, like so:
int main()
{
scene* ptr = new AnimatedScene();
if (typeid(*ptr) == typeid(AnimatedScene))
{
cout<<"ptr is indeed a animatedScene"<<endl;
AnimatedScene* ptr2 = (AnimatedScene*)(ptr);
ptr2->SetupAnimation();
}
else
{
cout<<"not a animatedscene!!!"<<endl;
}
}
This works, you'll then be able to use ptr2 to access the animatedScene's unique members.
Notice the use of pointers, you can't use the objects directly, due to something called "object slicing" when playing with polymorphism: https://en.wikipedia.org/wiki/Object_slicing
Like you I have heard something about the use of typeid and thus, casting being a bad idea, but as to why, I cannot tell you. I am hoping to have a more experienced programmer explain it.
What I can tell you is that this works without problems in this simple example, you've avoided the problem of declaring meaningless virtual functions in the basetype.
Edit: It's amazing how often I forget to use google: Why is usage of the typeid keyword bad design?
If I understand mr Bolas correctly, typeid incentivizes bad coding practices. However, in your example you want to access a subtypes non-virtual function. As far as I know, there is no way of doing that without checking type at compiletime, ie typeid.
If such problem arises with your hierarchy that proves that hierarchy was too generalized. You might want to implement interfaces pattern, if class have certain functionality, it would inherit an interface that defines that functionality.
Proven that dogs are animals, do all animal but dogs fail to bark, or do only dogs bark?
The first approach lead to a class animal failing all the verses of the entire zoo, implemented one-by one in each animal. And in particular class dog will override just bark().
In this approach animal becomes a sort of "god object" (knows everything), requiring to be constantly updated every time something new is introduced, and requiring It's entire "universe" to be re-created (recompile everything) after it.
The second approach requires first to check the animal is a dog (via dynamic cast) and then ask it to bark. (or check for cat before asking a mieow)
The trade-off will probably consist in understanding how frequent is the possibility you have to check a bark out of its context (not knowing which animal are you deal with), how to report a fail, and what to do in case of such fail.
In other words, the key driver is not the bark, but the context around it inside your program.
//Are you trying to do something like this?
class Scene{
public:
virtual void Display()=0; //Pure virtual func
};
class AnimatedScene : public Scene{
public:
void Display(){
std::cout<<"function Display() inside class AnimatedScene" <<std::endl;
}
};
class ControllableScene : public Scene{
public:
void Display(){
std::cout<<"function Display() inside class ControllableScene" <<std::endl;
}
};
int main(){
AnimatedScene as;
ControllableScene cs;
Scene *ex1 = &as;
Scene *ex2 = &cs;
ex1->Display();
ex2->Display();
return 0;
}

Is it ever a good idea to put virtual methods on a copyable type?

Have seen some related questions, but not this exact one...
I've treated classes as fitting into a few major categories, let's say these four for simplicity:
Value Classes which have some data and a bunch of operations. They can be copied and meaningfully compared for equality (with copies expected to be equal via ==). These pretty much always lack virtual methods.
Unique Classes whose instances have identity that you disable assignment and copying on. There's usually not an operator== on these because you compare them as pointers, not as objects. These quite often have a lot of virtual methods, as there isn't risk of object-slicing since you're being forced to pass them by pointer or reference.
Unique-but-Clonable Classes which disable copying, but are pre-designed to support cloning if that's what you really want. These have virtual methods, most importantly those following the virtual construction / cloning idiom
Container Classes which inherit the properties of whatever they're holding. These tend not to have virtual methods...see for instance "Why don't STL containers have virtual destructors?".
Regardless of holding this informal belief system, a couple times I've tried adding a virtual method to something copyable. While I may have thought it would "be really cool if that worked", inevitably it breaks.
This led me to wonder if anyone has an actual good example of a type which has virtual methods and doesn't disable copying?
The only counter-example that I have are classes that are meant to be stack-allocated and not heap-allocated. One scheme I use it for is Dependency Injection:
class LoggerInterface { public: virtual void log() = 0; };
class FileLogger final: public LoggerInterface { ... };
int main() {
FileLogger logger("log.txt");
callMethod(logger, ...);
}
The key point here is the final keyword though, it means that copying a FileLogger cannot lead to object-slicing.
However, it might just be that being final turned FileLogger into a Value class.
Note: I know, copying a logger seems weird...
There's nothing inherently wrong in being able to copy a polymorphic class. The problem is being able to copy a non-leaf class. Object slicing will get you.
A good rule of thumb to follow is never derive from a concrete class. This way, non-leaf classes are automatically non-instantiable and thus non-copyable. It won't hurt to disable assignment in them though, just to be on the safe side.
Of course nothing is wrong with copying an object via a virtual function. This kind of copying is safe.
Polymorphic classes are normally not "value-classes" but it does happen. std::stringstream comes to mind. It'not copyable, but it is movable (in C++11) and moving is no different from copying with regard to slicing.
Virtual dispatch happens a runtime. The only reason one should want it is when the actual, dynamic type of an object cannot be known until runtime. If you already knew the desired dynamic type when writing the program, you could use different, non-virtual techniques (such as templates, or non-polymorphic inheritance) to structure your code.
A good example for the need for runtime typing is parsing I/O messages, or handling events – any kind of situation where one way or another you'll either have some sort of big switch table to pick the correct concrete type, or you write your own registration-and-dispatch system, which basically reinvents polymorphism, or you just use virtual dispatch.
(Let me interject a warning: Many people misuse virtual functions to solve problems that don't need them, because they're not dynamic. Beware, and be critical of what you see.)
With this said, it's now clear that your code will be dealing mostly with the polymorphic base classes, e.g. in function interfaces or in containers. So let's rephrase the question: Should such a base class be copyable? Well, since you never have actual, most-derived base objects (i.e. the base class is essentially abstract), this isn't really an issue, and there's no need for this. You've already mentioned the "clone" idiom, which is the appropriate analogue of copying in a polymorphic.
Now, the "clone" function is necessarily implemented in every leaf class, and it necessarily requires copying of the leaf classes. So yes, every leaf class in a clonable hierarchy is a class with virtual functions and a copy constructor. And since the copy constructor of a derived class needs to copy its base subobjects, all the bases need to be copyable, too.
So, now I believe we've distilled the problem down to two possible cases: Either everything in your class hierarchy is completely uncopyable, or your hierarchy supports cloning, and thus by necessity every class in it is copyable.
So should a class with virtual functions have a copy constructor? Absolutely. (This answers your original question: when you integrate your class into a clonable, polymorphic hierarchy, you add virtual functions to it.)
Should you try to make a copy from a base reference? Probably not.
Not with a single, but with two classes:
#include <iostream>
#include <vector>
#include <stdexcept>
class Polymorph
{
protected:
class Implementation {
public:
virtual ~Implementation() {};
// Postcondition: The result is allocated with new.
// This base class throws std::logic error.
virtual Implementation* duplicate() {
throw std::logic_error("Duplication not supported.");
}
public:
virtual const char* name() = 0;
};
// Precondition: self is allocated with new.
Polymorph(Implementation* self)
: m_self(self)
{}
public:
Polymorph(const Polymorph& other)
: m_self(other.m_self->duplicate())
{}
~Polymorph() {
delete m_self;
}
Polymorph& operator = (Polymorph other) {
swap(other);
return *this;
}
void swap(Polymorph& other) {
std::swap(m_self, other.m_self);
}
const char* name() { return m_self->name(); }
private:
Implementation* m_self;
};
class A : public Polymorph
{
protected:
class Implementation : public Polymorph::Implementation
{
protected:
Implementation* duplicate() {
return new Implementation(*this);
}
public:
const char* name() { return "A"; }
};
public:
A()
: Polymorph(new Implementation())
{}
};
class B : public Polymorph {
protected:
class Implementation : public Polymorph::Implementation {
protected:
Implementation* duplicate() {
return new Implementation(*this);
}
public:
const char* name() { return "B"; }
};
public:
B()
: Polymorph(new Implementation())
{}
};
int main() {
std::vector<Polymorph> data;
data.push_back(A());
data.push_back(B());
for(auto x: data)
std::cout << x.name() << std::endl;
return 0;
}
Note: In this example the objects are copied, always (you may implement shared semantics, though)

C++ Singleton class - inheritance good practice

In an existing project, I am to inherit a Controller class (MVC) declared as Singleton so as to define my own treatment. How to appropriately derive this Singleton class?
First, I expand on context and need for this inheritance.
The application that I am added to the existing software wants to use a MVC module that performs almost same task as the one I am willing to perform. It is using the same methods up to signature and slight modifications. Rewriting my own MVC module would definitively be duplication of code. The existing module is intrinsically oriented towards its application to another part of the software, and I cannot simply use the same module. But is written as a Model-View-Controller pattern where Controller is Singleton. I derived View already.
Second, I have doubt that I can classicaly derive Singleton class.
Calling constructor from inherited class would simply call getinstance() for parent class and fail to return an object from derived class (?).
Third, it's how I see some way to deal with. Please comment/help me improve!
I copy the whole Singleton class in a class I could call AbstractController. I derive this class twice. The first child is singleton and adopts the whole treatment of parent class. The second child is the Controller for my part of the application, with own redefined treatment.
Thanks!
Truth is, singletons and inheritance do not play well together.
Yeah, yeah, the Singleton lovers and GoF cult will be all over me for this, saying "well, if you make your constructor protected..." and "you don't have to have a getInstance method on the class, you can put it...", but they're just proving my point. Singletons have to jump through a number of hoops in order to be both a singleton and a base class.
But just to answer the question, say we have a singleton base class. It can even to some degree enforce its singleness through inheritance. (The constructor does one of the few things that can work when it can no longer be private: it throws an exception if another Base already exists.) Say we also have a class Derived that inherits from Base. Since we're allowing inheritance, let's also say there can be any number of other subclasses of Base, that may or may not inherit from Derived.
But there's a problem -- the very one you're either already running into, or will soon. If we call Base::getInstance without having constructed an object already, we'll get a null pointer. We'd like to get back whatever singleton object exists (it may be a Base, and/or a Derived, and/or an Other). But it's hard to do so and still follow all the rules, cause there are only a couple of ways to do so -- and all of them have some drawbacks.
We could just create a Base and return it. Screw Derived and Other. End result: Base::getInstance() always returns exactly a Base. The child classes never get to play. Kinda defeats the purpose, IMO.
We could put a getInstance of our own in our derived class, and have the caller say Derived::getInstance() if they specifically want a Derived. This significantly increases coupling (because a caller now has to know to specifically request a Derived, and ends up tying itself to that implementation).
We could do a variant of that last one -- but instead of getting the instance, the function just creates one. (While we're at it, let's rename the function to initInstance, since we don't particularly care what it gets -- we're just calling it so that it creates a new Derived and sets that as the One True Instance.)
So (barring any oddness unaccounted for yet), it works out kinda like this...
class Base {
static Base * theOneTrueInstance;
public:
static Base & getInstance() {
if (!theOneTrueInstance) initInstance();
return *theOneTrueInstance;
}
static void initInstance() { new Base; }
protected:
Base() {
if (theOneTrueInstance) throw std::logic_error("Instance already exists");
theOneTrueInstance = this;
}
virtual ~Base() { } // so random strangers can't delete me
};
Base* Base::theOneTrueInstance = 0;
class Derived : public Base {
public:
static void initInstance() {
new Derived; // Derived() calls Base(), which sets this as "the instance"
}
protected:
Derived() { } // so we can't be instantiated by outsiders
~Derived() { } // so random strangers can't delete me
};
And in your init code, you say Base::initInstance(); or Derived::initInstance();, depending on which type you want the singleton to be. You'll have to cast the return value from Base::getInstance() in order to use any Derived-specific functions, of course, but without casting you can use any functions defined by Base, including virtual functions overridden by Derived.
Note that this way of doing it also has a number of drawbacks of its own, though:
It puts most of the burden of enforcing singleness on the base class. If the base doesn't have this or similar functionality, and you can't change it, you're kinda screwed.
The base class can't take all of the responsibility, though -- each class needs to declare a protected destructor, or someone could come along and delete the one instance after casting it (in)appropriately, and the whole thing goes to hell. What's worse, this can't be enforced by the compiler.
Because we're using protected destructors to prevent some random schmuck from deleting our instance, unless the compiler's smarter than i fear it is, even the runtime won't be able to properly delete your instance when the program ends. Bye bye, RAII...hello "memory leak detected" warnings. (Of course the memory will eventually be reclaimed by any decent OS. But if the destructor doesn't run, you can't depend on it to do cleanup for you. You'll need to call a cleanup function of some sort before you exit, and that won't give you anywhere near the same assurances that RAII can give you.)
It exposes an initInstance method that, IMO, doesn't really belong in an API everyone can see. If you wanted, you could make initInstance private and let your init function be a friend, but then your class is making assumptions about code outside itself, and the coupling thing comes back into play.
Also note that the code above is not at all thread safe. If you need that, you're on your own.
Seriously, the less painful route is to forget trying to enforce singleness. The least complicated way to ensure that there's only one instance is to only create one. If you need to use it multiple places, consider dependency injection. (The non-framework version of that amounts to "pass the object to stuff that needs it". :P ) I went and designed the above stuff just to try and prove myself wrong about singletons and inheritance, and just reaffirmed to myself how evil the combination is. I wouldn't recommend ever actually doing it in real code.
I'm not sure I understand the situation you're dealing with fully, and whether or not it's possible or appropriate to derive from the singleton depends very much on how the singleton is implemented.
But since you mentioned "good practice" there's some general points that come to mind when reading the question:
Inheritance isn't usually the best tool to achieve code re-use. See: Prefer composition over inheritance?
Using singleton and "good practice" generally do not go together! See: What is so bad about singletons?
Hope that helps.
I recently had the similar need in my app... in any case here is my out of code implementation :
h.
class icProjectManagerHandler;
class icProjectManager : public bs::icBaseManager {
friend class icProjectManagerHandler;
protected:
icProjectManager();
public:
~icProjectManager();
template<typename t>
static t *PM() {
return dynamic_cast<t *>(icProjectManagerHandler::PMH()->mCurrentManager);
};
template<typename t>
static t *PMS() {
static icProjectManagerHandler pm;
return static_cast<t *>(icProjectManagerHandler::PMH()->mCurrentManager);
};
};
class icProjectManagerHandler {
friend class icProjectManager;
icProjectManager *mCurrentManager;
icProjectManagerHandler();
public:
~icProjectManagerHandler();
static icProjectManagerHandler *PMH();
inline void setProjectManager(icProjectManager *pm) {
if (mCurrentManager) { delete mCurrentManager; }
mCurrentManager = pm;
}
};
Cpp.
icProjectManagerHandler::icProjectManagerHandler() {
mCurrentManager = new icProjectManager();
}
icProjectManagerHandler::~icProjectManagerHandler() {
}
icProjectManagerHandler *icProjectManagerHandler::PMH() {
static icProjectManagerHandler pmh;
return &pmh;
}
icProjectManager::icProjectManager() {
}
icProjectManager::~icProjectManager() {
}
And example:
class icProjectX : public ic::project::icProjectManager {
public:
icProjectX() {};
~icProjectX() {};
};
int main(int argc, char *argv[]) {
auto pro = new icProjectX();
pro->setIcName("Hello");
ic::project::icProjectManagerHandler::PMH()->setProjectManager(pro);
qDebug() << "\n" << pro << "\n" << ic::project::icProjectManager::PMS<icProjectX>();
return 10;
}
The issue of this implementation is that you have to initialize your "singleton" class 1st or else you will get the default base class. But other than that... it should work?

When should I use C++ private inheritance?

Unlike protected inheritance, C++ private inheritance found its way into mainstream C++ development. However, I still haven't found a good use for it.
When do you guys use it?
I use it all the time. A few examples off the top of my head:
When I want to expose some but not all of a base class's interface. Public inheritance would be a lie, as Liskov substitutability is broken, whereas composition would mean writing a bunch of forwarding functions.
When I want to derive from a concrete class without a virtual destructor. Public inheritance would invite clients to delete through a pointer-to-base, invoking undefined behaviour.
A typical example is deriving privately from an STL container:
class MyVector : private vector<int>
{
public:
// Using declarations expose the few functions my clients need
// without a load of forwarding functions.
using vector<int>::push_back;
// etc...
};
When implementing the Adapter Pattern, inheriting privately from the Adapted class saves having to forward to an enclosed instance.
To implement a private interface. This comes up often with the Observer Pattern. Typically my Observer class, MyClass say, subscribes itself with some Subject. Then, only MyClass needs to do the MyClass -> Observer conversion. The rest of the system doesn't need to know about it, so private inheritance is indicated.
Note after answer acceptance: This is NOT a complete answer. Read other answers like here (conceptually) and here (both theoretic and practic) if you are interested in the question. This is just a fancy trick that can be achieved with private inheritance. While it is fancy it is not the answer to the question.
Besides the basic usage of just private inheritance shown in the C++ FAQ (linked in other's comments) you can use a combination of private and virtual inheritance to seal a class (in .NET terminology) or to make a class final (in Java terminology). This is not a common use, but anyway I found it interesting:
class ClassSealer {
private:
friend class Sealed;
ClassSealer() {}
};
class Sealed : private virtual ClassSealer
{
// ...
};
class FailsToDerive : public Sealed
{
// Cannot be instantiated
};
Sealed can be instantiated. It derives from ClassSealer and can call the private constructor directly as it is a friend.
FailsToDerive won't compile as it must call the ClassSealer constructor directly (virtual inheritance requirement), but it cannot as it is private in the Sealed class and in this case FailsToDerive is not a friend of ClassSealer.
EDIT
It was mentioned in the comments that this could not be made generic at the time using CRTP. The C++11 standard removes that limitation by providing a different syntax to befriend template arguments:
template <typename T>
class Seal {
friend T; // not: friend class T!!!
Seal() {}
};
class Sealed : private virtual Seal<Sealed> // ...
Of course this is all moot, since C++11 provides a final contextual keyword for exactly this purpose:
class Sealed final // ...
The canonical usage of private inheritance is the "implemented in terms of" relationship (thanks to Scott Meyers' 'Effective C++' for this wording). In other words, the external interface of the inheriting class has no (visible) relationship to the inherited class, but it uses it internally to implement its functionality.
One useful use of private inheritence is when you have a class that implements an interface, that is then registered with some other object. You make that interface private so that the class itself has to register and only the specific object that its registered with can use those functions.
For example:
class FooInterface
{
public:
virtual void DoSomething() = 0;
};
class FooUser
{
public:
bool RegisterFooInterface(FooInterface* aInterface);
};
class FooImplementer : private FooInterface
{
public:
explicit FooImplementer(FooUser& aUser)
{
aUser.RegisterFooInterface(this);
}
private:
virtual void DoSomething() { ... }
};
Therefore the FooUser class can call the private methods of FooImplementer through the FooInterface interface, while other external classes cannot. This is a great pattern for handling specific callbacks that are defined as interfaces.
I think the critical section from the C++ FAQ Lite is:
A legitimate, long-term use for private inheritance is when you want to build a class Fred that uses code in a class Wilma, and the code from class Wilma needs to invoke member functions from your new class, Fred. In this case, Fred calls non-virtuals in Wilma, and Wilma calls (usually pure virtuals) in itself, which are overridden by Fred. This would be much harder to do with composition.
If in doubt, you should prefer composition over private inheritance.
I find it useful for interfaces (viz. abstract classes) that I'm inheriting where I don't want other code to touch the interface (only the inheriting class).
[edited in an example]
Take the example linked to above. Saying that
[...] class Wilma needs to invoke member functions from your new class, Fred.
is to say that Wilma is requiring Fred to be able to invoke certain member functions, or, rather it is saying that Wilma is an interface. Hence, as mentioned in the example
private inheritance isn't evil; it's just more expensive to maintain, since it increases the probability that someone will change something that will break your code.
comments on the desired effect of programmers needing to meet our interface requirements, or breaking the code. And, since fredCallsWilma() is protected only friends and derived classes can touch it i.e. an inherited interface (abstract class) that only the inheriting class can touch (and friends).
[edited in another example]
This page briefly discusses private interfaces (from yet another angle).
Sometimes I find it useful to use private inheritance when I want to expose a smaller interface (e.g. a collection) in the interface of another, where the collection implementation requires access to the state of the exposing class, in a similar manner to inner classes in Java.
class BigClass;
struct SomeCollection
{
iterator begin();
iterator end();
};
class BigClass : private SomeCollection
{
friend struct SomeCollection;
SomeCollection &GetThings() { return *this; }
};
Then if SomeCollection needs to access BigClass, it can static_cast<BigClass *>(this). No need to have an extra data member taking up space.
Private Inheritance to be used when relation is not "is a", But New class can be "implemented in term of existing class" or new class "work like" existing class.
example from "C++ coding standards by Andrei Alexandrescu, Herb Sutter" :-
Consider that two classes Square and Rectangle each have virtual functions for setting their height and width. Then Square cannot correctly inherit from Rectangle, because code that uses a modifiable Rectangle will assume that SetWidth does not change the height (whether Rectangle explicitly documents that contract or not), whereas Square::SetWidth cannot preserve that contract and its own squareness invariant at the same time. But Rectangle cannot correctly inherit from Square either, if clients of Square assume for example that a Square's area is its width squared, or if they rely on some other property that doesn't hold for Rectangles.
A square "is-a" rectangle (mathematically) but a Square is not a Rectangle (behaviorally). Consequently, instead of "is-a," we prefer to say "works-like-a" (or, if you prefer, "usable-as-a") to make the description less prone to misunderstanding.
I found a nice application for private inheritance, although it has a limited usage.
Problem to solve
Suppose you are given the following C API:
#ifdef __cplusplus
extern "C" {
#endif
typedef struct
{
/* raw owning pointer, it's C after all */
char const * name;
/* more variables that need resources
* ...
*/
} Widget;
Widget const * loadWidget();
void freeWidget(Widget const * widget);
#ifdef __cplusplus
} // end of extern "C"
#endif
Now your job is to implement this API using C++.
C-ish approach
Of course we could choose a C-ish implementation style like so:
Widget const * loadWidget()
{
auto result = std::make_unique<Widget>();
result->name = strdup("The Widget name");
// More similar assignments here
return result.release();
}
void freeWidget(Widget const * const widget)
{
free(result->name);
// More similar manual freeing of resources
delete widget;
}
But there are several disadvantages:
Manual resource (e.g. memory) management
It is easy to set up the struct wrong
It is easy to forget freeing the resources when freeing the struct
It is C-ish
C++ Approach
We are allowed to use C++, so why not use its full powers?
Introducing automated resource management
The above problems are basically all tied to the manual resource management. The solution that comes to mind is to inherit from Widget and add a resource managing instance to the derived class WidgetImpl for each variable:
class WidgetImpl : public Widget
{
public:
// Added bonus, Widget's members get default initialized
WidgetImpl()
: Widget()
{}
void setName(std::string newName)
{
m_nameResource = std::move(newName);
name = m_nameResource.c_str();
}
// More similar setters to follow
private:
std::string m_nameResource;
};
This simplifies the implementation to the following:
Widget const * loadWidget()
{
auto result = std::make_unique<WidgetImpl>();
result->setName("The Widget name");
// More similar setters here
return result.release();
}
void freeWidget(Widget const * const widget)
{
// No virtual destructor in the base class, thus static_cast must be used
delete static_cast<WidgetImpl const *>(widget);
}
Like this we remedied all the above problems. But a client can still forget about the setters of WidgetImpl and assign to the Widget members directly.
Private inheritance enters the stage
To encapsulate the Widget members we use private inheritance. Sadly we now need two extra functions to cast between both classes:
class WidgetImpl : private Widget
{
public:
WidgetImpl()
: Widget()
{}
void setName(std::string newName)
{
m_nameResource = std::move(newName);
name = m_nameResource.c_str();
}
// More similar setters to follow
Widget const * toWidget() const
{
return static_cast<Widget const *>(this);
}
static void deleteWidget(Widget const * const widget)
{
delete static_cast<WidgetImpl const *>(widget);
}
private:
std::string m_nameResource;
};
This makes the following adaptions necessary:
Widget const * loadWidget()
{
auto widgetImpl = std::make_unique<WidgetImpl>();
widgetImpl->setName("The Widget name");
// More similar setters here
auto const result = widgetImpl->toWidget();
widgetImpl.release();
return result;
}
void freeWidget(Widget const * const widget)
{
WidgetImpl::deleteWidget(widget);
}
This solution solves all the problems. No manual memory management and Widget is nicely encapsulated so that WidgetImpl does not have any public data members anymore. It makes the implementation easy to use correctly and hard (impossible?) to use wrong.
The code snippets form a compiling example on Coliru.
If you need a std::ostream with some small changes (like in this question) you may need to
Create a class MyStreambuf which derives from std::streambuf and implement changes there
Create a class MyOStream which derives from std::ostream that also initializes and manages an instance of MyStreambuf and passes the pointer to that instance to the constructor of std::ostream
The first idea might be to add the MyStream instance as a data member to the MyOStream class:
class MyOStream : public std::ostream
{
public:
MyOStream()
: std::basic_ostream{ &m_buf }
, m_buf{}
{}
private:
MyStreambuf m_buf;
};
But base classes are constructed before any data members so you are passing a pointer to a not yet constructed std::streambuf instance to std::ostream which is undefined behavior.
The solution is proposed in Ben's answer to the aforementioned question, simply inherit from the stream buffer first, then from the stream and then initialize the stream with this:
class MyOStream : public MyStreamBuf, public std::ostream
{
public:
MyOStream()
: MyStreamBuf{}
, basic_ostream{ this }
{}
};
However the resulting class could also be used as a std::streambuf instance which is usually undesired. Switching to private inheritance solves this problem:
class MyOStream : private MyStreamBuf, public std::ostream
{
public:
MyOStream()
: MyStreamBuf{}
, basic_ostream{ this }
{}
};
If derived class
- needs to reuse code and
- you can't change base class and
- is protecting its methods using base's members under a lock.
then you should use private inheritance, otherwise you have danger of unlocked base methods exported via this derived class.
Sometimes it could be an alternative to aggregation, for example if you want aggregation but with changed behaviour of aggregable entity (overriding the virtual functions).
But you're right, it has not many examples from the real world.
A class holds an invariant. The invariant is established by the constructor. However, in many situations it's useful to have a view of the representation state of the object (which you can transmit over network or save to a file - DTO if you prefer). REST is best done in terms of an AggregateType. This is especially true if you're const correct. Consider:
struct QuadraticEquationState {
const double a;
const double b;
const double c;
// named ctors so aggregate construction is available,
// which is the default usage pattern
// add your favourite ctors - throwing, try, cps
static QuadraticEquationState read(std::istream& is);
static std::optional<QuadraticEquationState> try_read(std::istream& is);
template<typename Then, typename Else>
static std::common_type<
decltype(std::declval<Then>()(std::declval<QuadraticEquationState>()),
decltype(std::declval<Else>()())>::type // this is just then(qes) or els(qes)
if_read(std::istream& is, Then then, Else els);
};
// this works with QuadraticEquation as well by default
std::ostream& operator<<(std::ostream& os, const QuadraticEquationState& qes);
// no operator>> as we're const correct.
// we _might_ (not necessarily want) operator>> for optional<qes>
std::istream& operator>>(std::istream& is, std::optional<QuadraticEquationState>);
struct QuadraticEquationCache {
mutable std::optional<double> determinant_cache;
mutable std::optional<double> x1_cache;
mutable std::optional<double> x2_cache;
mutable std::optional<double> sum_of_x12_cache;
};
class QuadraticEquation : public QuadraticEquationState, // private if base is non-const
private QuadraticEquationCache {
public:
QuadraticEquation(QuadraticEquationState); // in general, might throw
QuadraticEquation(const double a, const double b, const double c);
QuadraticEquation(const std::string& str);
QuadraticEquation(const ExpressionTree& str); // might throw
}
At this point, you might just store collections of cache in containers and look it up on construction. Handy if there's some real processing. Note that cache is part of the QE: operations defined on the QE might mean the cache is partially reusable (e.g., c does not affect the sum); yet, when there's no cache, it's worth to look it up.
Private inheritance can almost always modelled by a member (storing reference to the base if needed). It's just not always worth it to model that way; sometimes inheritance is the most efficient representation.
Just because C++ has a feature, doesn't mean it's useful or that it should be used.
I'd say you shouldn't use it at all.
If you're using it anyway, well, you're basically violating encapsulation, and lowering cohesion. You're putting data in one class, and adding methods that manipulates the data in another one.
Like other C++ features, it can be used to achieve side effects such as sealing a class (as mentioned in dribeas' answer), but this doesn't make it a good feature.