C++ multimethods and compile time detection - c++

I have the following code :
class A{};
class B: public A{};
class C: public A{};
class MyVisitor
{
public:
void visit(B*);
void visit(C*);
};
And then collection of A* objects, I want to achieve the following :
1)
MyVisitor visitor;
for(vector<A*>::iterator it = vec.begin(); it!= vec.end();it++)
visitor->visit(a);
2) Somehow determine at compile time, if A* points to derived object D,and give compiler error, if MyVisitor::visit(D*) function is not present
I know that 1) is achievable with some multimethods implementation, I guess I can find some implementations of multimethods for c++ . But is 2) somehow possible ?

You could use dynamic_cast like this (inside the body of your for loop) since the behavior should vary at run-time (according to the actual type of data).
ClassB* ba = dynamic_cast<ClassB*>(a);
if (ba)
visitor->visit(ba);
ClassC* ca = dynamic_cast<ClassC*>(a);
if (ca)
visitor->visit(ca);
Maybe your visit functions might be declared virtual (for your ClassD thing).
Otherwise organize your classes as a tree (not a forest) of classes, and have your topmost root class
class Topmost {
virtual int classnum() const;
and adopt the convention that each non-abstract class gives its unique classnum etc... Or have a metaclass mechanism (like e.g. Qt has)

You can try something like this.
#include <iostream>
class A
{
virtual void visit() = 0;
};
class B: private A
{
public:
void visit()
{
std::cout << __PRETTY_FUNCTION__ << "\n";
}
};
class C: private A
{
public:
void visit()
{
std::cout << __PRETTY_FUNCTION__ << "\n";
}
};
template <typename... Args>
class MyVisitor : public Args...
{
public:
template <typename T>
void visit(T* t)
{
t->visit();
}
};
int main()
{
MyVisitor<B, C> visitor;
B b;
B* bp = &b;
visitor.visit(bp);
return 0;
}
Live example

You may apply the visitor pattern completly:
class B;
class C;
class IVisitor
{
public:
void visit(B&) = 0;
void visit(C&) = 0;
};
class A
{
virtual ~A() = default;
virtual void accept(IVisitor& v) = 0;
};
class B: public A{ void accept(IVisitor& v) override { v.visit(*this); } };
class C: public A{ void accept(IVisitor& v) override { v.visit(*this); } };

Related

Best way to achieve late-stage polymorphism

I have several disparate templated pure abstract classes. I derive from these to get a bunch of classes, and from there, I can use those to make a bunch of objects. I would like to put all of these objects into a container. However, they are all of different types. I am wondering how to accomplish this late-stage polymorphism.
Say this is my pre-existing code that I have right now:
#include <iostream>
template<typename T>
class A{
public:
A() : m_num(1.0) {};
virtual ~A() {};
virtual void printNum() const = 0;
protected:
T m_num;
};
template<typename T>
class B{
public:
B() : m_num(2.0) {};
virtual ~B() {};
virtual void printTwiceNum() const = 0;
protected:
T m_num;
};
class A_example : public A<int>
{
public:
A_example() : A<int>() {};
void printNum() const { std::cout << m_num << "\n"; };
};
class B_example : public B<int>
{
public:
B_example() : B<int>() {};
void printTwiceNum() const { std::cout << 2*m_num << "\n"; };
};
int main(){
A_example first;
B_example second;
first.printNum();
second.printTwiceNum();
return 0;
}
With more classes, it could get pretty messy inside of main(). Ideally I could jut iterate over the container and call print() on each element. My first thought is to use a std::vector<unique_ptr<Base>>. This seems to work:
#include <iostream>
#include <vector> // new include
#include <memory> // new include
#include <utility> // new include
// new Base class here
class Base{
public:
virtual ~Base(){};
};
template<typename T>
class A : public Base{ // new inheritance here
public:
A() : m_num(1.0) {};
virtual ~A() {};
virtual void printNum() const = 0;
protected:
T m_num;
};
template<typename T>
class B : public Base{ // new inheritance here as well
public:
B() : m_num(2.0) {};
virtual ~B() {};
virtual void printTwiceNum() const = 0;
protected:
T m_num;
};
class A_example : public A<int>
{
public:
A_example() : A<int>() {};
void printNum() const { std::cout << m_num << "\n"; };
};
class B_example : public B<int>
{
public:
B_example() : B<int>() {};
void printTwiceNum() const { std::cout << 2*m_num << "\n"; };
};
int main(){
std::vector<std::unique_ptr<Base>> v;
v.emplace_back( new A_example() );
v.emplace_back( new B_example() );
//v[0]->printNum(); // nope
//v[1]->printTwiceNum(); // nope
return 0;
}
This is cool because I didn't have to change A_example or B_example, and all I changed in A and B was that I added : public Base. However, I have no idea how to call each elements print*** function. Is there any way to call the printNum() and printTwiceNum() functions, and for them to be automatically recognized?
The simplest approach is to just make a virtual function Base::print and have your derived classes implement it. But that's not always appropriate.
Another approach is to branch on dynamic_cast conversions. The premise there is that some functions are only available on some classes. But this can get hairy especially when using class templates, as you must handle all expected template parameters.
To generalize this, you can use interface classes. Let's say you have lots of different classes but only a small number of print variations. In that case, it may make sense to do this:
class PrintNumInterface {
public:
virtual void printNum() const = 0;
};
class PrintTwiceNumInterface {
public:
virtual void printTwiceNum() const = 0;
};
template<typename T> class A : public Base, public PrintNumInterface { ... };
template<typename T> class B : public Base, public PrintTwiceNumInterface { ... };
And now, no matter how many additional classes or template expansions you have to deal with, you only need to handle these interfaces:
for (auto& p : v)
{
if (PrintNumInterface* iface = dynamic_cast<PrintNumInterface*>(p.get())
iface->printNum();
else if (PrintTwiceNumInterface* iface = dynamic_cast<PrintTwiceNumInterface*>(p.get())
iface->printTwiceNum();
}

multiple inheritance for function only - without virtual and CRTP

How to do multiple inheritance just for function?
must share data of the base class
no virtual function (assume that vtable is expensive)
avoid virtual inheritance
implementation must be able to reside in .cpp
c++14 is allowed
Here are similar questions :-
Multiple inheritance in diamond shape with functions only - use virtual inheritance. Virtual inheritance is generally bad and expensive.
multiple inheritance without virtual inheritance - focuses on syntax and compiling rather than programming technique.
Multilevel inheritance in c++ (CRTP) , CRTP and multilevel inheritance , Eliminate redundancy with CRTP and multiple inheritance (C++03) and Using CRTP with virtual inheritance - implementation must be in header
Here is a sample code (coliru demo) :-
class O{
protected: int database=0;
};
class A : public O{
public: void print(){
std::cout<<database<<std::endl;
}
};
class B : public O{
public: void set(int s){
database=s+1;
}
};
class AB : public O{
public: void print(){//duplicate
std::cout<<database<<std::endl;
}
public: void set(int s){//duplicate
database=s+1;
}
};
//AB ab; ab.set(1); ab.print(); // would print 2
Here is my attempt (wandbox demo). I abuse CRTP :( :-
class O{
public: int database=0;
};
template<class T>class OA{
public: void print(){
std::cout<<static_cast<T*>(this)->database<<std::endl;
}
};
template<class T>class OB{
public: void set(int s){
static_cast<T*>(this)->database=s+1;
}
};
class A :public O,public OA<A>{};
class B :public O,public OB<B>{};
class AB :public O,public OA<AB>,public OB<AB>{};
It works, but it looks inelegant.
Furthermore, implementation must be in header (because OA and OB are template classes).
Are there better approaches? Or is this the way to go?
Sorry if it is too newbie question or already asked. I am a C++ beginner.
Edit
Give extended example of using please.
In ECS, it would be useful in some cases :-
class O{
protected: EntityHandle e;
};
class ViewAsPhysic : public O{ //A
public: void setTransform(Transformation t){
Ptr<PhysicTransformComponent> g=e;
g->transform=t;
}
};
class ViewAsLight : public O{ //B
public: void setBrightness(int t){
Ptr<LightComponent> g=e;
g->clan=t;
}
};
class ViewAsLightBlock : public O{ //AB
//both functions
};
The problem here is that the database field is member of class O. So without virtual inheritance, A and B will have each their own copy of database. So you must find a way to force A and B to share same value. You could for example use a reference field initialized in a protected constructor:
#include <iostream>
class O{
int _db;
protected: int &database;
O(): database(_db) {};
O(int &db): database(db) {};
};
class A : public O{
public: void print(){
std::cout<<database<<std::endl;
}
A() {} // public default ctor
protected: A(int& db): O(db) {}; // protectect ctor
};
class B : public O{
public: void set(int s){
database=s+1;
}
B() {} // public default ctor
protected: B(int& db): O(db) {}; // protectect ctor
};
class AB : public A, public B {
int _db2;
public: AB(): A(_db2), B(_db2) {}; // initialize both references to same private var
};
int main() {
AB ab;
ab.set(1);
ab.print();
return 0;
}
displays as expected:
2
Above code uses no virtual inheritance, no virtual function and no templates, so method can safely implemented in cpp files. The class AB actually uses methods from its both parents and has still a coherent view on its underlying data. In fact it simulates an explicit virtual inheritance by building the common data in the most derived class and injecting in through protected constructors in its parents.
Something like this?
must share data of the base class - check
no virtual function (assume that vtable is expensive) - check
avoid virtual inheritance - check
implementation must be able to reside in .cpp- check
c++14 is allowed - check. c++11 used.
#include <iostream>
class O {
protected:
int database = 0;
};
/*
* the concept of implementing print for a base class
*/
template<class...Bases>
struct implements_print : Bases... {
void print() const {
std::cout << this->database << std::endl;
}
};
/*
* The concept of implementing set for a base class
*/
template<class...Bases>
struct implements_set : Bases... {
void set() {
++this->database;
}
};
struct B : implements_set<O> {
};
struct A : implements_print<O> {
};
struct AB : implements_set<implements_print<O>> {
};
int main() {
A a;
a.print();
B b;
b.set();
AB ab;
ab.set();
ab.print();
}
Another way, using composition and an access class to provide access to the protected member. This example shows how to defer the work on database to another compilation unit:
#include <iostream>
/*
* this stuff in cpp
*/
namespace implementation
{
void print(const int& database) {
std::cout << database << std::endl;
}
void set(int& database) {
++database;
}
}
/*
* this stuff in header
*/
struct OAccess;
class O {
private:
int database = 0;
friend OAccess;
};
struct OAccess {
template<class Host>
constexpr decltype(auto) database(Host &host) const { return (host.database); } // note: () makes reference
template<class Host>
constexpr decltype(auto) database(Host const &host) const { return (host.database); } // note: () makes reference
};
/*
* the concept of implementing print for a derived class
*/
template<class Host>
struct implements_print {
void print() const {
OAccess access;
implementation::print(access.database(self()));
}
private:
decltype(auto) self() const { return static_cast<Host const &>(*this); }
};
/*
* The concept of implementing set for a derived class
*/
template<class Host>
struct implements_set {
void set() {
OAccess access;
implementation::set(access.database(self()));
}
private:
decltype(auto) self() { return static_cast<Host &>(*this); }
};
template<template<class> class...Impls>
struct OImpl : Impls<OImpl<Impls...>> ..., O {
};
using B = OImpl<implements_set>;
using A = OImpl<implements_print>;
using AB = OImpl<implements_print, implements_set>;
int main() {
A a;
a.print();
B b;
b.set();
AB ab;
ab.set();
ab.print();
}
We start with defining the concepts of things that can print and things that can be set:
namespace util {
template<class Base, class Derived, class R=void>
using if_base = std::enable_if_t< std::is_base_of< std::decay_t<Base>, std::decay_t<Derived>>::value, R >;
struct stub {};
}
namespace concepts {
template<class Token>
void do_print(Token, util::stub const&)=delete;
template<class Token>
void do_set(Token, util::stub&, int)=delete;
struct has_print {
struct token { friend struct has_print; private: token(int){} };
template<class T>
friend util::if_base<has_print, T> print(T const& t) {
do_print(get_token(), t);
}
private: static token get_token() { return 0; }
};
struct has_set {
struct token { friend struct has_set; private: token(int){} };
template<class T>
friend util::if_base<has_set, T> set(T& t, int x) {
do_set(get_token(),t, x);
}
private: static token get_token() { return 0; }
};
}
We then declare O and the operations you can support on it:
namespace DB {
class O;
void do_print(::concepts::has_print::token, O const& o);
void do_set(::concepts::has_set::token, O& o, int);
class O{
protected: int database=0;
friend void do_print(::concepts::has_print::token, O const&);
friend void do_set(::concepts::has_set::token, O&, int);
};
class A : public O, public concepts::has_print {
};
class B : public O, public concepts::has_set {
};
class AB : public O, public concepts::has_print, concepts::has_set {
};
}
void DB::do_print(::concepts::has_print::token, O const& o ) { std::cout << o.database << std::endl; }
void DB::do_set(::concepts::has_set::token, O& o, int x) { o.database = x+1; }
The hard part of this is the access control.
I ensure it isn't possible to call do_set except through has_set::set.
That is what all those tokens are about. You can strip them out and their overhead if you are willing to just say "don't call the do_ functions" (and maybe give them another name, like private_impl_set).
Live example.
To start discussion.
class O
{
// no virtual destructor. So cant use polymorphic deletion
// like :
// O *o = new AB;
// delete o;
protected: int database=0;
};
class A : virtual public O{
public: void print(){
std::cout<<database<<std::endl;
}
};
class B : virtual public O{
public: void set(int s){
database=s+1;
}
};
class AB : protected A, protected B{}; // no vtable
void foo() {
AB ab;
ab.print(); // won't perform virtual call.
}

Acyclic Visitor C++

I'm reading the book by Alexandrescu, and I've run into the Acyclic Visitor pattern. I think that it's possible to get rid of the macross that calls AcceptImpl method of the BaseVisitable class.
Could you tell me, whether the following implementation bellow conforms the standard?
class BaseVisitor
{
public:
virtual ~BaseVisitor() {}
};
template <class SpecificVisitable>
class SpecificVisitor
{
public:
virtual void Visit(SpecificVisitable& t) = 0;
protected:
~SpecificVisitor() {}
};
template <class SpecificVisitable>
class BaseVisitable
{
public:
void Accept(BaseVisitor& visitor)
{
SpecificVisitor<SpecificVisitable>& specificVisitor = dynamic_cast<SpecificVisitor<SpecificVisitable>&>(visitor);
specificVisitor.Visit(static_cast<SpecificVisitable&>(*this));
}
protected:
~BaseVisitable() {}
};
class A : public BaseVisitable<A>
{
public:
void PrintA() { std::cout << "A\n"; }
};
class B : public BaseVisitable<B>
{
public:
void PrintB() { std::cout << "B\n"; }
};
class PrintVisitor final:
public BaseVisitor,
public SpecificVisitor<A>,
public SpecificVisitor<B>
{
public:
virtual void Visit(A& a) override
{
a.PrintA();
}
virtual void Visit(B& b) override
{
b.PrintB();
}
};
int main()
{
A a;
B b;
PrintVisitor visitor;
a.Accept(visitor);
b.Accept(visitor);
}
I agree with the commenters to the original post that state that a dynamic_cast to a reference will fail. You are not testing that situation, so your code will of course work just fine. But, in the general case, it's wrong.
I recommend that you instead rewrite BaseVisitable to instead cast to a pointer and check that pointer before dispatching on it. Something like
if (auto sv = dynamic_cast<SpecificVisitor<SpecificVisitable>*>(visitor))
sv->visit(*this)

Iterate over different CRTP Derived class methods

In an example below I have a pretty typical CRTP example, two different derived classes that both have a method bar. The base class has a method foo which just forwards to some derived bar method
#include <iostream>
template<typename Derived>
class Base {
public:
void foo() {
static_cast<Derived*>(this)->bar();
}
};
class DerivedA : public Base<DerivedA> {
public:
void bar() {
::std::cout << "A\n";
}
};
class DerivedB : public Base<DerivedB> {
public:
void bar() {
::std::cout << "B\n";
}
};
int main() {
DerivedA a;
DerivedB b;
a.foo();
b.foo();
}
It doesn't seem like I can have an array / vector / etc. of the base class because it would have to have a type along the lines of Base<T> where T is different
Is there some kind of convention without virtual for being able to iterate over different derived classes assuming they all have the same method (bar in this case)?
You can use Boost.Variant. For example:
typedef boost::variant<DerivedA, DerivedB> Derived;
struct BarCaller : public boost::static_visitor<void> {
template <class T>
void operator()(T& obj) {
obj.bar();
}
};
int main() {
std::vector<Derived> vec{DerivedA(), DerivedB(), DerivedA()};
BarCaller bar;
for (Derived& obj : vec) {
obj.apply_visitor(bar);
}
}
This lets you store heterogeneous types in a vector or other STL container (by using a "discriminated union"), and lets you call a specific function on all of them regardless of their not having a common ancestor or any virtual methods.
It doesn't seem like I can have an array / vector / etc. of the base class because it would have to have a type along the lines of Base<T> where T is different.
You can have a base class of Base<T> for all T, then, you can have a list/vector/array of pointers to the base class, if that works for you.
struct BaseOne
{
virtual void foo() = 0;
virtual ~BaseOne() {}
};
template<typename Derived>
class Base : struct BaseOne {
public:
void foo() {
static_cast<Derived*>(this)->bar();
}
};
and then,
int main() {
std::vector<BaseOne*> v {new DerivedA, new DerivedB };
for ( auto item : v )
item->bar();
for ( auto item : v )
delete item;
}
Is there some kind of convention without virtual for being able to iterate over different derived classes assuming they all have the same method (bar in this case)?
No, there isn't.
As per now, variant has became part of the C++17 standard and the solution to the problem can be solved by std::variant and std::visit as follows.
The template class in the example is Interface<> and use the CRTP idiom to force derived class to implement helloImpl():
#include <iostream>
#include <vector>
#include <variant>
template<typename Implementer>
struct Interface {
void hello() const {
static_cast<Implementer const *>(this)->helloImpl();
}
};
A couple of class examples with different implementations of helloImpl()
struct Hello1 : public Interface<Hello1> {
void helloImpl() const {
std::cout << "Hello1" << std::endl;
}
};
struct Hello2 : public Interface<Hello2> {
void helloImpl() const {
std::cout << "Hello2" << std::endl;
}
};
And here is how to use it to store data in a vector<> container and its traversal:
int main() {
using var_t = std::variant<Hello1, Hello2>;
std::vector<var_t> items{Hello1(), Hello1(), Hello2()};
for(auto &item: items) {
std::visit([](auto &&arg) {
arg.hello();
}, item);
}
return 0;
}

Avoid new when storing base-typed member variable that could be initialized with different derived types

My code structure is like below where multiple classes implement Interface. In Example class I store a pointer to the Interface and new() it in the constructor appropriately (depending on constructor parameters not shown here). I'm looking for ways to avoid using new() in this scenario but haven't got a solution yet. What's the best practice for something like this?
class Interface
{
virtual void Foo() = 0;
};
class A : public Interface
{
void Foo() { ... }
};
class B : public Interface
{
void Foo() { ... }
};
class Example
{
private:
Interface* m_bar;
public:
Example()
{
m_bar = new A(); // deleted in destructor
}
};
There are two ways this is typically done, each with their own merits.
If A is truely defined at compile time, than a typical way to handle this is to simply use a template type:
template <typename T>
class TemplateExample
{
T m_bar;
public:
TemplateExample() : m_bar() {};
}
This has some downsides. TemplateExample<A> becomes unrelated to TemplateExample<B>, the error messages when T doesn't follow the correct interface are pretty obtuse, ect. The upside is this may use duck typing rather than interface typing, and m_bar is a concrete instance.
The other (arguable more common) way is to do the following
class UniquePtrExample
{
std::unique_ptr<Interface> m_bar;
public:
UniquePtrExample() : m_bar(new A()){}
};
This has the benefit of being able to be run time configuratble if you follow a cloable pattern:
class Interface
{
public:
virtual void Foo() = 0;
virtual Interface* clone() const = 0;
};
template <typename T>
class CloneHelper : public Interface
{
public:
virtual Interface* clone() const { return new T(static_cast<const T&>(*this));}
};
class A : public CloneHelper<A>
{
virtual void Foo() { std::cout << 'A' << std::endl; }
};
class B : public CloneHelper<B>
{
virtual void Foo() { std::cout << 'B' << std::endl; }
};
class UniquePtrExample
{
std::unique_ptr<Interface> m_bar;
public:
UniquePtrExample() : m_bar(new A()){}
UniquePtrExample(const Interface& i) : m_bar(i.clone());
};
Note you can further extend the above to have a move variant of the clone function.