I've got a interface that's implemented as an abstract base class with a number of pure virtual public methods. These pure virtual functions can be implemented using a template since the differences between subclasses aren't large - so my idea was to use multiple inheritance to mix-in the appropriately templated helper-class that provides the implementation. However, the compiler complains that the base class is abstract; it isn't considering the helper mix-in's implementation so thinks there's no implementation of a required method.
For example:
class TrivialList {
int count;
public:
TrivialList(int count) : count(count){}
virtual double Average() const=0;
int Count() const {return count;}
virtual ~TrivialList(){}
};
template<typename TIndexable> class AverageHelper {
public:
double Average() const {
TIndexable const & self = static_cast<TIndexable const &>(*this);
double sum=0.0;
for(int i=0;i<self.Count();++) sum += self.Get(i);
return sum / self.Count();
}
};
class IndexableList : public TrivialList, public AverageHelper<IndexableList> {
std::vector<double> backend;
public:
IndexableList(int count) : TrivialList(count), backend(count) { }
double & Get(int i) { return backend[i];}
double const & Get(int i) const { return backend[i];}
};
IndexableList * MakeList() {return new IndexableList(5);} //error!
// cannot instantiate abstract class
I'm using MSC 10.0 (Visual Studio 2010); the code fails with a similar error using g++ 4.5.
Get or the real-world equivalents in my project cannot be virtual because they're extremely minor operations that need to be inlined for adequate performance (think put-pixel/get-pixel) - so I need the generic algorithms to be templated rather than generic via virtual function calls.
For implementing mix-ins via templates, you need the template implementing the abstract function to derive from the abstract base class.
So you may fix your code by changing it the following way:
// ...
template<typename TIndexable> class AverageHelper : public TriviaList{
// ...
class IndexableList : public AverageHelper<IndexableList> {
In general, if you want to provide more than one mix-in, you may either use a virtual inheritance in order not multiplying the instances of the base classes, or to use chain inheritance as in the following sample:
class Abstract {
public:
virtual void foo() = 0;
virtual void bar() = 0;
};
template<class Base>
class FooImpl : Base {
public:
void foo() { /* default foo implementation */ }
};
template<class Base>
class BarImpl : Base {
public:
void bar() { /* default bar implementation */ }
};
class Derived : public BarImpl<FooImpl<Abstract> > {
// You have both foo() and bar() implementations available
};
It doesn't work because AverageHelper<>::Average() doesn't override TrivialList::Average(). In order to override a virtual function, the overriding class must inherit from the class containing the function to be overridden.
You could change your template thus:
template<typename TIndexable, typename Base >
class AverageHelper : public Base {
public:
template< typename T >
AverageHelper(T arg) : Base(arg) {}
// ...
};
class IndexableList : public AverageHelper<IndexableList,TrivialList> {
public:
IndexableList(int count) : AverageHelper<IndexableList,TrivialList>(count) {}
// ...
};
You might want to virtually derive from TrivialList:
template<typename TIndexable, typename Base >
class AverageHelper : virtual public Base {
// ...
};
Related
I've got a class Base from which I have two classes, DerivedA and DerivedB as defined below.
template <typename Derived>
class Base{
public:
double interface(){
static_cast<Derived*>(this)->implementation();
}
};
class DerivedA : public Base<DerivedA>{
public:
double implementation(){ return 2.0;}
};
class DerivedB : public Base<DerivedB>{
public:
double implementation(){ return 1.0;}
};
In short, I'm trying to do the following to maintain a collection of objects, some of which are DerivedA and some of which are DerivedB:
std::vector<std::shared_ptr<Derived>>
Which is obviously impossible beacuse I've now made the class Derived a templated class.
Is there any way I can create / maintain a polymorphic collection of objects?
EDIT: Unfortunately, a simple templated structure does not work as the function implementation is templated in my actual program -- so then implementation would have to be a templated pure virtual function, which cannot be. Pardon my lack of explanation.
This answer pertains to the question as it was at the time of this answer.
Don't use CRTP, which is not dynamic polymorphism, to create dynamic polymorphism.
Use a virtual function.
That's what they're for.
class Base
{
private:
virtual
auto implementation() -> double = 0;
public:
auto interface() -> double { return implementation(); }
};
class DerivedA
: public Base
{
private:
auto implementation() -> double override { return 2.0; }
};
class DerivedB
: public Base
{
private:
auto implementation() -> double override { return 1.0; }
};
Alf's suggestion is on target. It is easy to adapt it to your additional requirement. Define an interface with a pure virtual method:
struct BaseInterface {
virtual ~BaseInterface() {}
virtual double interface() = 0;
};
Now, your template base class can derive from the interface:
template <typename Derived>
class Base : BaseInterface {
public:
double interface(){
static_cast<Derived*>(this)->implementation();
}
};
Now, you can create a vector of pointers to the interface:
std::vector<std::shared_ptr<BaseInterface>>
Because Base<DerivedA> is a completely different type compared to Base<DerivedB>, you are right that you can't just do something like std::vector<std::shared_ptr<Base>>, as it would be syntactically invalid and has no meaningful semantics with regards to C++.
One way to achieve what you want and preserve your current CRTP hierarchy is to create a type erasing interface (or is it what it should be called? I'm not sure...). It is basically a wrapper that defines a certain interface in which you could wrap objects that obey that interface.
#include <vector>
#include <memory>
#include <iostream>
class VirtualBase { // Really am not sure what it should be called, sorry
class Interface {
public:
virtual ~Interface() = default;
virtual double get() = 0;
};
template<typename T>
class Impl : public Interface {
T m_impl_obj;
public:
Impl(T impl_obj) : m_impl_obj(std::move(impl_obj)) {}
double get() override {
return m_impl_obj.get();
}
};
std::shared_ptr<Interface> m_obj;
public:
template<typename T>
VirtualBase(T obj) : m_obj(new Impl<T>(std::move(obj))) {}
double get() {
return m_obj->get();
}
};
template <typename Derived>
class Base{
public:
double get(){
return static_cast<Derived*>(this)->implementation();
}
};
class DerivedA : public Base<DerivedA>{
public:
double get(){ return 2.0;}
};
class DerivedB : public Base<DerivedB>{
public:
double get(){ return 1.0;}
};
int main() {
std::vector<VirtualBase> v;
v.emplace_back(DerivedA{});
v.emplace_back(DerivedB{});
for(auto b : v) {
std::cout << b.get() << std::endl;
}
}
Live example
This is quite incomplete, but it should work, at least in my case if I need such a design. An excellent introduction, with explanations and rationales of how and why, is Sean Parent's talk at GoingNative 2013: http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil . Really, you should see it, including all the other great presentations in GoingNative.
Edit: this is based on an assumption that OP actually needs virtual function templates. This is apparently not the case.
It is not possible to have virtual function templates.
One way to partially simulate them is the Visitor pattern. It is often used to add dynamically dispatched functionality to existing classes. The fact that this added functionality can be templatized is overlooked all too often.
A pseudocode example
class Base
{
virtual void accept (Visitor*) = 0;
}
class Derived1 : public Base
{
void accept(Visitor* v) { v->visit(this); }
}
class Derived2 : public Base
{
void accept(Visitor* v) { v->visit(this); }
}
class Visitor
{
virtual void visit (Derived1*) = 0;
virtual void visit (Derived2*) = 0;
}
template<class X, class Y>
class FooVisitor
{
X x; Y y;
FooVisitor(X x, Y y): x(x), y(y) {}
void visit (Derived1*) { x.func(y); }
void visit (Derived2*) { y.otherfunc(x); }
}
Of course all the downsides of Visitor are here as well.
How do I ensure my derived class implements at least one of two chosen methods in the base class?
class base {
public:
virtual int sample()=0;
virtual Eigen::VectorXf sample()=0;
};
class Derived : Base {
int sample() override {return 1;}
}
This code returns an error, as the sample method is not implemented with the VectorXf return type. However, my intention is that only one of these need to be implemented. The only reason they are seperate in the base class is that they have different return type. How can I do this in C++?
Overloading by return type is not possible. You may use std::variant instead:
#include <variant>
class Base {
public:
virtual std::variant<int, Eigen::VectorXf> sample()=0;
};
class Derived : public Base {
std::variant<int, Eigen::VectorXf> sample() override {return 1;}
};
If one is restricted to C++11, then there are many alternatives.
Implement and use something like variant: a class that has a enumerator selecting between two active types, and a union to contain these types.
Use Boost variant.
std::pair
Implement a hierarchy of classes (a simplification of std::any), and return on the right pointer to object:
class AbstractBase {
public:
virtual ~AbstractBase() = 0;
template <class T>
const T* get() const;
};
template <class T>
class ValueWrapper : public AbstractBase {
public:
ValueWrapper(const T& value) : m_value(value) {}
const T & getValue() const { return m_value; }
private:
T m_value;
};
template <class T>
inline const T * AbstractBase::get() const {
auto child = dynamic_cast<ValueWrapper<T> const*>(this);
return child ? &child->getValue() : nullptr;
}
class Base {
public:
virtual std::unique_ptr<AbstractBase> sample()=0;
};
The question is, why would you need this?
I've got a class Base from which I have two classes, DerivedA and DerivedB as defined below.
template <typename Derived>
class Base{
public:
double interface(){
static_cast<Derived*>(this)->implementation();
}
};
class DerivedA : public Base<DerivedA>{
public:
double implementation(){ return 2.0;}
};
class DerivedB : public Base<DerivedB>{
public:
double implementation(){ return 1.0;}
};
In short, I'm trying to do the following to maintain a collection of objects, some of which are DerivedA and some of which are DerivedB:
std::vector<std::shared_ptr<Derived>>
Which is obviously impossible beacuse I've now made the class Derived a templated class.
Is there any way I can create / maintain a polymorphic collection of objects?
EDIT: Unfortunately, a simple templated structure does not work as the function implementation is templated in my actual program -- so then implementation would have to be a templated pure virtual function, which cannot be. Pardon my lack of explanation.
This answer pertains to the question as it was at the time of this answer.
Don't use CRTP, which is not dynamic polymorphism, to create dynamic polymorphism.
Use a virtual function.
That's what they're for.
class Base
{
private:
virtual
auto implementation() -> double = 0;
public:
auto interface() -> double { return implementation(); }
};
class DerivedA
: public Base
{
private:
auto implementation() -> double override { return 2.0; }
};
class DerivedB
: public Base
{
private:
auto implementation() -> double override { return 1.0; }
};
Alf's suggestion is on target. It is easy to adapt it to your additional requirement. Define an interface with a pure virtual method:
struct BaseInterface {
virtual ~BaseInterface() {}
virtual double interface() = 0;
};
Now, your template base class can derive from the interface:
template <typename Derived>
class Base : BaseInterface {
public:
double interface(){
static_cast<Derived*>(this)->implementation();
}
};
Now, you can create a vector of pointers to the interface:
std::vector<std::shared_ptr<BaseInterface>>
Because Base<DerivedA> is a completely different type compared to Base<DerivedB>, you are right that you can't just do something like std::vector<std::shared_ptr<Base>>, as it would be syntactically invalid and has no meaningful semantics with regards to C++.
One way to achieve what you want and preserve your current CRTP hierarchy is to create a type erasing interface (or is it what it should be called? I'm not sure...). It is basically a wrapper that defines a certain interface in which you could wrap objects that obey that interface.
#include <vector>
#include <memory>
#include <iostream>
class VirtualBase { // Really am not sure what it should be called, sorry
class Interface {
public:
virtual ~Interface() = default;
virtual double get() = 0;
};
template<typename T>
class Impl : public Interface {
T m_impl_obj;
public:
Impl(T impl_obj) : m_impl_obj(std::move(impl_obj)) {}
double get() override {
return m_impl_obj.get();
}
};
std::shared_ptr<Interface> m_obj;
public:
template<typename T>
VirtualBase(T obj) : m_obj(new Impl<T>(std::move(obj))) {}
double get() {
return m_obj->get();
}
};
template <typename Derived>
class Base{
public:
double get(){
return static_cast<Derived*>(this)->implementation();
}
};
class DerivedA : public Base<DerivedA>{
public:
double get(){ return 2.0;}
};
class DerivedB : public Base<DerivedB>{
public:
double get(){ return 1.0;}
};
int main() {
std::vector<VirtualBase> v;
v.emplace_back(DerivedA{});
v.emplace_back(DerivedB{});
for(auto b : v) {
std::cout << b.get() << std::endl;
}
}
Live example
This is quite incomplete, but it should work, at least in my case if I need such a design. An excellent introduction, with explanations and rationales of how and why, is Sean Parent's talk at GoingNative 2013: http://channel9.msdn.com/Events/GoingNative/2013/Inheritance-Is-The-Base-Class-of-Evil . Really, you should see it, including all the other great presentations in GoingNative.
Edit: this is based on an assumption that OP actually needs virtual function templates. This is apparently not the case.
It is not possible to have virtual function templates.
One way to partially simulate them is the Visitor pattern. It is often used to add dynamically dispatched functionality to existing classes. The fact that this added functionality can be templatized is overlooked all too often.
A pseudocode example
class Base
{
virtual void accept (Visitor*) = 0;
}
class Derived1 : public Base
{
void accept(Visitor* v) { v->visit(this); }
}
class Derived2 : public Base
{
void accept(Visitor* v) { v->visit(this); }
}
class Visitor
{
virtual void visit (Derived1*) = 0;
virtual void visit (Derived2*) = 0;
}
template<class X, class Y>
class FooVisitor
{
X x; Y y;
FooVisitor(X x, Y y): x(x), y(y) {}
void visit (Derived1*) { x.func(y); }
void visit (Derived2*) { y.otherfunc(x); }
}
Of course all the downsides of Visitor are here as well.
template <class CollectionItem>
class Collection
{
void A();
// Many other utility functions
}
class ICollection
{
virtual void B() = 0;
}
class Base : public Collection<BaseItem>, public IBase
{
virtual void B();
}
Is there any way of offering Collection functions via ICollection interface without wrapping all the functions in Base class? ICollection : public Collection<CollectionItem> is not an option.
Bounty Update:
OK, so the original idea was to have Interface to all Collection classes. Before we continue, every CollectionItem also has Interface, let's call it ICollectionItem and ICollection only knows about ICollectionItem.
So what I did was create another template class as Interface to Collection template class - ICollection (pure virtual) accepting ICollectionItem(s). Collection class inherits this interface.
Every Collection class (inheriting Collection<CollectionItem> class) would also inherit it's Interface Collection class. That Interface then virtual inherits ICollection<ICollectionItem>. I'll just post the code :)
Here is the code:
template <class ICollectionItem>
class ICollection
{
public:
virtual const ICollectionItem* At(const int idx) = 0;
};
template <class CollectionItem, class ICollectionItem>
class Collection
: public ICollection,
public virtual ICollection<ICollectionItem> // Weak point
{
private:
List<CollectionItem*> fContainer;
public:
Collection(void) {}
virtual ~Collection() {}
virtual const ICollectionItem* At(const int idx); // Casting GetAt result
virtual const TCollectionItem& GetAt(const int idx) const
virtual ListIterator<TCollectionItem> >* GetIterator(void) const;
virtual ListIterator<ICollectionItem> >* Iterator(void) const; // Weak point
}
Example usage:
class IBaseItem
{
public:
virtual int Number() = 0;
{
class BaseItem
: public IBaseItem
{
public:
virtual int Number();
void SetNumber(int value);
}
class IBase
: public virtual ICollection<IBaseItem>
{
public:
virtual IBaseItem* ItemByName(String name) = 0;
virtual ~IBase() {}
}
class Base
: public Collection<BaseItem, IBaseItem>,
public IBase
{
public:
BaseItem* GetItemByName(String name);
virtual IBaseItem* ItemByName(String name);
}
Weak points:
First is at using virtual inheritance ... lots written about it, not much to talk about, or is it?
Unable to access Iterator using ICollection interface. See ListIterator function, only first one can be implemented, the second one would require some kind of new List of IBaseItem. I decided to live with that and just use for loop.
Even tho I somehow managed to get what I wanted (With wrapping and casting), I would still like to hear an second opinion. I don't like using virtual inheritance, specially in such delicate situations - using Collections for application Base creation.
I can not see any other solution than calling some Collection method in Base implementation of IBase virtual methods.
class Base : public Collection<BaseItem>, public IBase
{
virtual void B()
{
A();
}
}
You say, and I quote:
I want to call Collection functions using IBase pointer
I really don't see what is to be done here besides dynamic_cast. It does exactly what you want it to do.
void fun(IBase * base) {
auto * coll = dynamic_cast<Collection<BaseItem>*>(base);
if (coll) {
coll->A();
}
}
Your Collection class must have a virtual destructor.
You can, of course, offer a templated version, if you'd need different baseitems in different, scenarios for some reasons. This has bad code smell and I think your architecture is bad at this point, but oh well.
template <typename T> void fun(IBase * base) {
auto * coll = dynamic_cast<Collection<T>*>(base);
if (coll) {
coll->A();
}
}
void test(IBase * p) {
fun<BaseItem5>(p);
}
If you have some other specific scenario in mind, please edit your question to say what you mean.
Hmm...So you wanna to reuse the Collection class's utility functions, and you want to design a class which will implement an interface defined by IBase. As you mentioned above,"wrapping all the functions in Base class" is a way to offer Collection functions.
(1) Via inheritance,derived class has a good knowledge of Collection
class Derived:public Collection<DerivedType>,public IBase{};
or
template <typename T>
class Derived:public Collection<T>,public IBase{};
(2) Via inheritance,derived class knows little about Collection,but through IBase
class IBase : public Collection<BaseItem>{};
class Derived:public IBase{};
By (1),If you want to call Collection functions using IBase pointer,you have to wrap the functions.
By (2), any Derived instance is " a kind of " IBase which is "a kind of " Collection. So you can use IBase pointer to call Collection functions.
So,the key point is that the objects pointed by the IBase pointer should have the method you want to call.Wrap it or inherit it. I can not see any other solution than these two ways.
Edit: the idea is refined based on your example:
Here is an idea:
//generic interface can be kept as it is
template <class ICollectionItem>
class ICollection
{
public:
virtual const ICollectionItem* At(const int idx) = 0;
};
class Empty
{
};
template <class CollectionItem , class BaseClass = Empty>
class GenericCollection
: public BaseClass
{
public:
const CollectionItem* At(const int idx);
// At and ItemByName are standard functions for a collection
CollectionItem* ItemByName(String name);
//note that here nothing has to be declared as virtual
};
//example usage:
class IBase
: public virtual ICollection<IBaseItem>
{
public:
virtual IBaseItem* ItemByName(String name) = 0;
virtual ~IBase() {}
};
class Base
: public GenericCollection<BaseItem, IBase >
{
public:
//nothing to be implemented here, all functions are implemented in GenericCollection and defined as virtual in IBase
//The definition of the functions has to be the same:
};
In collection you can implement whatever and in the interface you can define what ever you want to be virtual from your collection. The only thing is that you need to have some standard in naming convention for functions.
Hope this helps,
Raxvan.
From your comments in another answer, it seems you want a collection of interfaces, and an implementation of this interface. The simplest I can advise you is the following:
template<typename T>
class ICollection
{
public:
virtual iterator<T>* begin() const = 0;
};
template<typename T, typename TBase>
class Collection : public ICollection<TBase>
{
public:
iterator_impl<T>* begin() const { return whatever; }
};
Example:
class IItem {};
class Item : public IItem {};
class Base : public Collection<Item, IItem> {};
old answer:
Is there any way of offering Collection functions via IBase interface without wrapping all the functions in Base class ?
If I understood your problem, you want to use it like this:
void myfunc()
{
// ...
IBase* obj = ...;
obj->A();
obj->B();
}
I think here is a misunderstanding here: if you want A() to be callable from an IBase, then you have to add it to Ibase declaration.
If you want to use the Collection functions on an object, then you should cast this object to a Collection, via dynamic_cast for example.
Furthermore, if you have such a funcion:
void fun(IBase* base) { /* ... */ }
you cannot cast to a Collection*, since there are no relationship between these two classes, unless you have another way to be sure base is a Collection:
void fun(IBase* base)
{
if(base && base->isABaseItemCollection())
{
// Valid, since the real type was checked before
Collection* collection = (Collection*)base;
// ...
}
}
On a side note: you can generate bases almost automatically:
template
class Base : public Collection, public U {};
typedef Base BaseCollection;
According to comment/chat:
You have something like:
class IAnimal { /*...*/ };
class Cat : public IAnimal { /*...*/ };
class Dog : public IAnimal { /*...*/ };
class Cats
{
std::vector<Cat*> cats;
public:
Cat* at(size_t index) { return cats[index]; }
/*...*/
};
class Dogs
{
std::vector<Dog*> dogs;
public:
Dog* at(size_t index) { return dogs[index]; }
/*...*/
};
And you want to factorize some code using something like
class IAnimals
{
public:
std::vector<IAnimals*> animals; // or getter/setter which works with IAnimals.
/* some common factorized code */
};
// And so
class Cats : public IAnimals { /**/ };
class Dogs : public IAnimals { /**/ };
I propose, instead of creating class IAnimals, to use template functions as:
template <typename TAnimals>
void foo(TAnimals& animals)
{
Ianimals* animal = animals.at(42);
// ...
animal->eat(food);
// ...
}
You have to give compatible "interface" (names) to the type used in template.
Maybe you could have an operator() in IBase that would be delegated to Base?
class CollectionBase {};
template <class Item> class Collection: public CollectionBase {};
class IBase
{
public:
virtual CollectionBase* operator()() = 0;
};
class Base : public Collection<BaseItem>, public IBase
{
public:
virtual Collection<BaseItem>* operator()() { return this; }
};
This is somewhat hypothetical as I'm not too worried about performance - just wondering which option is actually the fastest/most efficient in general, or if there is no difference whatsoever.
Suppose I have the following code for a visitor template that supports overloading:
#define IMPLEMENT_VISITOR_WITH_SUPERCLASS(superclass) \
typedef superclass visitor_super_t; \
virtual void visit(Visitor& v) { v.visit(*this); }
//-----------------------------------------------------------------------------
// Implementation detail:
// Selective dispatcher for the visitor - required to handle overloading.
//
template <typename T>
struct VisitorDispatch {
static void dispatch(Visitor* v, T* t) { v->visit(*t); }
};
// Specalization for cases where dispatch is not defined
template <> struct VisitorDispatch<void> {
static void dispatch(Visitor* v, void* t) { throw std::bad_cast(""); }
};
//-----------------------------------------------------------------------------
// Derive visitors from this and 'Visitor'.
template <typename T>
class VTarget
{
public:
// Don't really need a virtual dtor.
virtual void dispatch(T& t) = 0;
};
//-----------------------------------------------------------------------------
class Visitor
{
public:
virtual ~Visitor() = 0;
template <typename T>
void visit(T& t) {
typedef VTarget<T> target_t;
target_t* tgt = dynamic_cast<target_t*>(this);
if (tgt) {
tgt->dispatch(t);
}
else {
// Navigate up inhertiance hierarchy.
// requires 'super' to be defined in all classes in hierarchy
// applicable to this visitor.
typedef typename T::visitor_super_t super;
super* s = static_cast<super*>(&t);
VisitorDispatch<super>::dispatch(this, s);
}
}
};
//-----------------------------------------------------------------------------
inline Visitor::~Visitor() {}
This is then used to create generic visitors:
class CommonBase {
IMPLEMENT_VISITOR_WITH_SUPERCLASS(void)
virtual ~CommonBase() = 0;
};
class A : public CommonBase {
IMPLEMENT_VISITOR_WITH_SUPERCLASS(CommonBase)
};
class B : public CommonBase {
IMPLEMENT_VISITOR_WITH_SUPERCLASS(CommonBase)
};
class MyVisitor
: public Visitor
, public VTarget<CommonBase>
, public VTarget<A>
, public VTarget<B>
{
public:
virtual void dispatch(CommonBase& obj);
virtual void dispatch(A& obj);
virtual void dispatch(B& obj);
};
Using the visitor ultimately results in dynamic_cast<>'s from Visitor to VTarget<T>, which is a cross-cast.
The other way this could be implemented is to make Visitor a virtual base of VTarget<T> - MyVisitor would then not need to inherit directly from Visitor anymore.
The dynamic_cast<> in the Visitor::visit code would then result in a down-cast from the virtual base, Visitor.
Is one method faster than the other when performing the casts? Or do you only get a size penalty for having the virtual base?
Well, it looks like the cross-cast method is faster than the virtual base method.
With a visitation that requires 1 fallback to a superclass, over 100000000 iterations, the cross-cast method took 30.2747 seconds, and the virtual base method took 41.3999 - about 37% slower.
With no fallback to an overload for a superclass, the cross-cast was 10.733 seconds and the virtual base 19.9982 (86% slower).
I was more interested to know how the dynamic_cast would operate in either mode, really.