Consider polymorphic classes with a base object, a derived interface, and a final object:
// base object
struct object
{
virtual ~object() = default;
};
// interfaces derived from base object
struct interface1 : object
{
virtual void print_hello() const = 0;
template<typename T>
static void on_destruction(object* /*ptr*/)
{
std::cout << "interface1::on_destruction" << std::endl;
}
};
// final object
struct derived1 : interface1
{
virtual void print_hello() const override
{
std::cout << "hello" << std::endl;
}
static std::string get_type_name()
{
return "derived1";
}
};
In the real use case, final objects are defined through a plugin system, but that is not the point. Note that I want to be able to call on_destruction when an object is destroyed (see register_object below). I want to use these classes as follows:
int main()
{
// register derived1 as an instantiable object,
// may be called in a plugin
register_object<derived1>();
// create an instance using the factory system
auto instance = create_unique<interface1>("derived1");
instance->print_hello();
return 0;
}
Using std::unique_ptr to manage the objects, I ended up with the following code for register_object:
template<typename T>
using unique = std::unique_ptr<
T,
std::function<void(object*)> // object deleter
>;
namespace
{
std::map< std::string, std::function<unique<object>(void)> > factory_map;
}
template<typename T>
void register_object()
{
factory_map.emplace(
T::get_type_name(),
[]()
{
unique<T> instance{
new T,
[](object* ptr)
{
T::on_destruction<T>(ptr);
delete ptr;
}
};
return static_move_cast<object>(
std::move(instance)
);
}
);
}
And the create* functions:
unique<object> create_unique_object(const std::string& type_name)
{
auto f = factory_map.at(type_name);
return f();
}
template<typename T>
unique<T> create_unique(const std::string& type_name)
{
return static_move_cast<T>(
create_unique_object(type_name)
);
}
You noticed in register_object and create_unique the call to static_move_cast, which is declared as:
template<typename U, typename T, typename D>
std::unique_ptr<U, D>
static_move_cast
(
std::unique_ptr<T, D>&& to_move_cast
)
{
auto deleter = to_move_cast.get_deleter();
return std::unique_ptr<U, D>{
static_cast<U*>(
to_move_cast.release()
),
deleter
};
}
The goal behind static_move_cast is to allow static_cast on std::unique_ptr while moving the deleter during the cast. The code is working, but I feel like hacking std::unique_ptr. Is there a way to refactor the code to avoid my static_move_cast?
static_move_cast is unnecessary within register_object, since you can just use the converting constructor of unique_ptr template< class U, class E > unique_ptr( unique_ptr<U, E>&& u ):
unique<T> instance{
new T,
// ...
};
return instance;
Or, even simpler, construct and return a unique<object> directly, since T* is convertible to object*:
return unique<object>{
new T,
// ...
};
However for create_unique the use of static_move_cast is unavoidable, since the converting constructor of unique_ptr won't work for downcasts.
Note that shared_ptr has static_pointer_cast, which performs downcasts, but there is no corresponding facility for unique_ptr, presumably because it it is considered straightforward and correct to perform the cast yourself.
I would say it is good solution given the requirements. You transfer the responsibility to the caller of create_unique. He must give correct combination of type and string and string that is in the registry.
auto instance = create_unique<interface1>("derived1");
// ^^^^^^^^^^ ^^^^^^^^
// what if those two don't match?
You could improve it a bit by changing the static_cast to dynamic_cast. And the caller of create_unique should always check that he got non-null pointer before calling anything on it.
Or at least use dynamic_cast with assert in debug mode, so you catch mismatches while developing.
Alternative refactoring: Have separate factory for every existing interface.
Related
I am curious how one would go about storing a parameter pack passed into a function and storing the values for later use.
For instance:
class Storage {
public:
template<typename... Args>
Storage(Args... args) {
//store args somehow
}
}
Basically I am trying to make a class like tuple, but where you don't have to specify what types the tuple will hold, you just pass in the values through the constructor.
So for instance instead of doing something like this:
std::tuple<int, std::string> t = std::make_tuple(5, "s");
You could do this:
Storage storage(5, "s");
And this way you could any Storage objects in the same vector or list. And then in the storage class there would be some method like std::get that would return a given index of an element we passed in.
Since run will return void, I assume all the functions you need to wrap can be functions that return void too.
In that case you can do it like this (and let lambda capture do the storing for you):
#include <iostream>
#include <functional>
#include <string>
#include <utility>
class FnWrapper
{
public:
template<typename fn_t, typename... args_t>
FnWrapper(fn_t fn, args_t&&... args) :
m_fn{ [=] { fn(args...); } }
{
}
void run()
{
m_fn();
}
private:
std::function<void()> m_fn;
};
void foo(const std::string& b)
{
std::cout << b;
}
int main()
{
std::string hello{ "Hello World!" };
FnWrapper wrapper{ foo, hello };
wrapper.run();
return 0;
}
OK, what you're asking is type erasure. Typical way of implementing it is via a virtual function inherited by a class template.
Live demo here: https://godbolt.org/z/fddfTEe5M
I stripped all the forwards, references and other boilerplate for brevity. It is not meant to be production code by any means.
#include<memory>
#include <iostream>
#include <stdexcept>
struct Fn
{
Fn() = default;
template<typename F, typename...Arguments>
Fn(F f, Arguments...arguments)
{
callable =
std::make_unique<CallableImpl<F, Arguments...>>(f, arguments...);
}
void operator()()
{
callable
? callable->call()
: throw std::runtime_error("empty function");
}
struct Callable
{
virtual void call() =0;
virtual ~Callable() = default;
};
template<typename T, typename...Args_>
struct CallableImpl : Callable
{
CallableImpl(T f, Args_...args)
: theCallable(f)
, theArgs(std::make_tuple(args...))
{}
T theCallable;
std::tuple<Args_...> theArgs;
void call() override
{
std::apply(theCallable, theArgs);
}
};
std::unique_ptr<Callable> callable{};
};
void f(int a)
{
std::cout << a << '\n';
}
int main(int, char*[])
{
Fn fx{f, 3};
fx();
char x = 'q';
Fn flambda( [x](){std::cerr << x << '\n';} );
flambda();
}
The "meat" of it lies here:
struct Callable
{
virtual void call() =0;
virtual ~Callable() = default;
};
template<typename T, typename...Args_>
struct CallableImpl : Callable
{
CallableImpl(T f, Args_...args)
: theCallable(f)
, theArgs(std::make_tuple(args...))
{}
T theCallable;
std::tuple<Args_...> theArgs;
void call() override
{
std::apply(theCallable, theArgs);
}
};
Callable is just the interface to access the object. Enough to store a pointer to it and access desired methods.
The actual storage happens in its derived classes:template<typename T, typename...Args_> struct CallableImpl : Callable. Note the tuple there.
T is for storing the actual object, whatever it is. Note that it has to implement some for of compile-time interface, in C++ terms referred to as a concept. In that case, it has to be callable with a given set of arguments.
Thus it has to be known upfront.
The outer structure holds the unique_ptr to Callable but is able to instantiate the interface thanks to the templated constructor:
template<typename F, typename...Arguments>
Fn(F f, Arguments...arguments)
{
callable =
std::make_unique<CallableImpl<F, Arguments...>>(f, arguments...);
}
What is the main advantage of it?
When done properly, it has value semantics. Effectively, it can be used to represent a sort of polymorphism without derivation, note T doesn't have to have a common base class, it just has to be callable in one way or another; this can be used for addition, subtraction, printing, whatever.
As for the main drawbacks: a virtual function call (CallableImpl stored as Callable) which may hinder performance. Also, getting back the original type is difficult, if not nearly impossible.
I am trying to create a QList of a polymorphic type that still uses Qt's implicit sharing.
My specific use case is passing items held in a QList to QtConcurrent::mapped. The items all descend from a base class which defines a virtual function that QtConcurrent::mapped will call. The majority of the stored data will be child class specific. These items may be edited after the threading begins, leaving me with two main options, locks or copy the data. I do not want to stick locks in, because that would eliminate most of the purpose of using extra threads. Also making full copies of my data also seems quite undesirable. Instead I would like use Qt's implicit sharing to only make copies of the data items that I change, however I can't seem to make a QList of a polymorphic type that still uses implicit sharing.
QList by default uses implicit sharing, so at first glance it would seem that we are already done.
QList<Base> list;
Derived derived_obj;
list.append(derived_obj); // this fails
However appending a child class to a QList of the parent class will not work and the standard answer is to instead use a QList of QSharedPointers to the base class, which will accept appending a pointer to the child class.
QList<QSharedPointer<Base> > pointer_list;
QSharedPointer<Derived> derived_pointer;
pointer_list.append(derived_pointer); // this works but there is no copy-on-write
If I use a QList of QSharedPointers, it is the QSharedPointer that will be be copied rather than my polymorphic class, meaning that I have lost the copy-on-write functionality that I would like.
I have also looked at using a QList of QSharedDataPointers.
QList<QSharedDataPointer<Base> > data_pointer_list;
QSharedDataPointer<Derived> derived_data_pointer;
list.append(derived_data_pointer); // this fails
However like QList itself, QSharedDataPointers do not seem to accept polymorphic types.
This fails:
QList<QSharedDataPointer<Base>> list;
QSharedDataPointer<Derived> derived(new Derived);
list.append(derived);
Note An alternative approach to the below would be to merge the PolymorphicShared and PolymorphicSharedBase to add polymorphism support directly to QSharedDataPointer, without placing special requirements on the QSharedData-derived type (e.g. it wouldn't need to explicitly support clone). This requires a bit more work. The below is just one working approach.
QSharedDataPointer is indeed the answer you seek and can definitely hold polymorphic QSharedData. You need to separate the type into a hierarchy based on QSharedData, and another parallel hierarchy wrapping the QSharedDataPointer. The QSharedDataPointer is not usually meant to be used directly by the end user of a class. It's an implementation detail useful in implementing an implicitly shared class.
For efficiency's sake, a QSharedDataPointer is a small type that can be moved at the bit level. It's quite efficient when stored in containers of all sorts - especially in Qt containers that can utilize the type traits to be aware of this property. The size of a class using a QSharedDataPointer will usually double if we make it polymorphic itself, thus it helps not to do it. The pointed-to data type can be polymorphic, of course.
First, let's define a rather univeral base class PIMPL that you'll build the hierarchy on. The PIMPL class can be dumped into the debug stream, and cloned.
// https://github.com/KubaO/stackoverflown/tree/master/questions/implicit-list-44593216
#include <QtCore>
#include <type_traits>
class PolymorphicSharedData : public QSharedData {
public:
virtual PolymorphicSharedData * clone() const = 0;
virtual QDebug dump(QDebug) const = 0;
virtual ~PolymorphicSharedData() {}
};
The xxxData types are PIMPLs and are not meant for use by the end-user. The user is meant to use the xxx type itself. This shared type then wraps the polymorphic PIMPL and uses the QSharedDataPointer for storage of the PIMPL. It exposes the methods of the PIMPL.
The type itself is not polymorphic, to save on the size of the virtual table pointer. The as() function acts as dynamic_cast() would, by redirecting polymorphism to the PIMPL.
class PolymorphicShared {
protected:
QSharedDataPointer<PolymorphicSharedData> d_ptr;
PolymorphicShared(PolymorphicSharedData * d) : d_ptr(d) {}
public:
PolymorphicShared() = default;
PolymorphicShared(const PolymorphicShared & o) = default;
PolymorphicShared & operator=(const PolymorphicShared &) = default;
QDebug dump(QDebug dbg) const { return d_ptr->dump(dbg); }
template <class T> typename
std::enable_if<std::is_pointer<T>::value, typename
std::enable_if<!std::is_const<typename std::remove_pointer<T>::type>::value, T>::type>
::type as() {
if (dynamic_cast<typename std::remove_pointer<T>::type::PIMPL*>(d_ptr.data()))
return static_cast<T>(this);
return {};
}
template <class T> typename
std::enable_if<std::is_pointer<T>::value, typename
std::enable_if<std::is_const<typename std::remove_pointer<T>::type>::value, T>::type>
::type as() const {
if (dynamic_cast<const typename std::remove_pointer<T>::type::PIMPL*>(d_ptr.data()))
return static_cast<T>(this);
return {};
}
template <class T> typename
std::enable_if<std::is_reference<T>::value, typename
std::enable_if<!std::is_const<typename std::remove_reference<T>::type>::value, T>::type>
::type as() {
Q_UNUSED(dynamic_cast<typename std::remove_reference<T>::type::PIMPL&>(*d_ptr));
return static_cast<T>(*this);
}
template <class T> typename
std::enable_if<std::is_reference<T>::value, typename
std::enable_if<std::is_const<typename std::remove_reference<T>::type>::value, T>::type>
::type as() const {
Q_UNUSED(dynamic_cast<const typename std::remove_reference<T>::type::PIMPL&>(*d_ptr));
return static_cast<T>(*this);
}
int ref() const { return d_ptr ? d_ptr->ref.load() : 0; }
};
QDebug operator<<(QDebug dbg, const PolymorphicShared & val) {
return val.dump(dbg);
}
Q_DECLARE_TYPEINFO(PolymorphicShared, Q_MOVABLE_TYPE);
#define DECLARE_TYPEINFO(concreteType) Q_DECLARE_TYPEINFO(concreteType, Q_MOVABLE_TYPE)
template <> PolymorphicSharedData * QSharedDataPointer<PolymorphicSharedData>::clone() {
return d->clone();
}
A helper to makes it easy to use the abstract base class with derived data types. It casts the d-ptr to a proper derived PIMPL type, and forwards the constructor arguments to the PIMPL's constructor.
template <class Data, class Base = PolymorphicShared> class PolymorphicSharedBase : public Base {
friend class PolymorphicShared;
protected:
using PIMPL = typename std::enable_if<std::is_base_of<PolymorphicSharedData, Data>::value, Data>::type;
PIMPL * d() { return static_cast<PIMPL*>(&*this->d_ptr); }
const PIMPL * d() const { return static_cast<const PIMPL*>(&*this->d_ptr); }
PolymorphicSharedBase(PolymorphicSharedData * d) : Base(d) {}
template <typename T> static typename std::enable_if<std::is_constructible<T>::value, T*>::type
construct() { return new T(); }
template <typename T> static typename std::enable_if<!std::is_constructible<T>::value, T*>::type
construct() { return nullptr; }
public:
using Base::Base;
template<typename ...Args,
typename = typename std::enable_if<std::is_constructible<Data, Args...>::value>::type
> PolymorphicSharedBase(Args&&... args) :
Base(static_cast<PolymorphicSharedData*>(new Data(std::forward<Args>(args)...))) {}
PolymorphicSharedBase() : Base(construct<Data>()) {}
};
It's now a simple matter to have the parallel hierarchy of PIMPL types and their carriers. First, a basic abstract type in our hierarchy that adds two methods. Note how PolymorphicSharedBase adds the d() accessor of the correct type.
class MyAbstractTypeData : public PolymorphicSharedData {
public:
virtual void gurgle() = 0;
virtual int gargle() const = 0;
};
class MyAbstractType : public PolymorphicSharedBase<MyAbstractTypeData> {
public:
using PolymorphicSharedBase::PolymorphicSharedBase;
void gurgle() { d()->gurgle(); }
int gargle() const { return d()->gargle(); }
};
DECLARE_TYPEINFO(MyAbstractType);
Then, a concrete type that adds no new methods:
class FooTypeData : public MyAbstractTypeData {
protected:
int m_foo = 0;
public:
FooTypeData() = default;
FooTypeData(int data) : m_foo(data) {}
void gurgle() override { m_foo++; }
int gargle() const override { return m_foo; }
MyAbstractTypeData * clone() const override { return new FooTypeData(*this); }
QDebug dump(QDebug dbg) const override {
return dbg << "FooType-" << ref << ":" << m_foo;
}
};
using FooType = PolymorphicSharedBase<FooTypeData, MyAbstractType>;
DECLARE_TYPEINFO(FooType);
And another type that adds methods.
class BarTypeData : public FooTypeData {
protected:
int m_bar = 0;
public:
BarTypeData() = default;
BarTypeData(int data) : m_bar(data) {}
MyAbstractTypeData * clone() const override { return new BarTypeData(*this); }
QDebug dump(QDebug dbg) const override {
return dbg << "BarType-" << ref << ":" << m_foo << "," << m_bar;
}
virtual void murgle() { m_bar++; }
};
class BarType : public PolymorphicSharedBase<BarTypeData, FooType> {
public:
using PolymorphicSharedBase::PolymorphicSharedBase;
void murgle() { d()->murgle(); }
};
DECLARE_TYPEINFO(BarType);
We'll want to verify that the as() method throws as needed:
template <typename F> bool is_bad_cast(F && fun) {
try { fun(); } catch (std::bad_cast) { return true; }
return false;
}
The use of the implicitly shared types is no different than the use of Qt's own such types. We can also cast using as instead of dynamic_cast.
int main() {
Q_ASSERT(sizeof(FooType) == sizeof(void*));
MyAbstractType a;
Q_ASSERT(!a.as<FooType*>());
FooType foo;
Q_ASSERT(foo.as<FooType*>());
a = foo;
Q_ASSERT(a.ref() == 2);
Q_ASSERT(a.as<const FooType*>());
Q_ASSERT(a.ref() == 2);
Q_ASSERT(a.as<FooType*>());
Q_ASSERT(a.ref() == 1);
MyAbstractType a2(foo);
Q_ASSERT(a2.ref() == 2);
QList<MyAbstractType> list1{FooType(3), BarType(8)};
auto list2 = list1;
qDebug() << "After copy: " << list1 << list2;
list2.detach();
qDebug() << "After detach: " << list1 << list2;
list1[0].gurgle();
qDebug() << "After list1[0] mod: " << list1 << list2;
Q_ASSERT(list2[1].as<BarType*>());
list2[1].as<BarType&>().murgle();
qDebug() << "After list2[1] mod: " << list1 << list2;
Q_ASSERT(!list2[0].as<BarType*>());
Q_ASSERT(is_bad_cast([&]{ list2[0].as<BarType&>(); }));
auto const list3 = list1;
Q_ASSERT(!list3[0].as<const BarType*>());
Q_ASSERT(is_bad_cast([&]{ list3[0].as<const BarType&>(); }));
}
Output:
After copy: (FooType-1:3, BarType-1:0,8) (FooType-1:3, BarType-1:0,8)
After detach: (FooType-2:3, BarType-2:0,8) (FooType-2:3, BarType-2:0,8)
After list1[0] mod: (FooType-1:4, BarType-2:0,8) (FooType-1:3, BarType-2:0,8)
After list2[1] mod: (FooType-1:4, BarType-1:0,8) (FooType-1:3, BarType-1:0,9)
The list copy was shallow and the items themselves weren't copied: the reference counts are all 1. After the detach, all data items were copied but because they are implicitly shared, they only incremented their reference counts. Finally, after an item is was modified, it is automatically detached, and the reference counts drop back to 1.
I am trying to create my own boost::adaptors::transformed.
Here is the related boost code.
Here is its usage (modified from a SO answer by LogicStuff):-
C funcPointer(B& b){
//"funcPointer" is function convert from "B" to "C"
return instance-of-C
}
MyArray<B> test; //<-- any type, must already have begin() & end()
for(C c : test | boost::adaptor::transformed(funcPointer)) {
//... something ....
}
The result will be the same as :-
for(auto b : test) {
C c = funcPointer(b);
//... something ...
}
My Attempt
I created CollectAdapter that aim to work like boost::adaptor::transformed.
It works OK in most common cases.
Here is the full demo and back up. (same as below code)
The problematic part is CollectAdapter - the core of my library.
I don't know whether I should cache the collection_ by-pointer or by-value.
CollectAdapter encapsulates underlying collection_ (e.g. pointer to std::vector<>) :-
template<class COLLECTION,class ADAPTER>class CollectAdapter{
using CollectAdapterT=CollectAdapter<COLLECTION,ADAPTER>;
COLLECTION* collection_; //<---- #1 problem? should cache by value?
ADAPTER adapter_; //<---- = func1 (or func2)
public: CollectAdapter(COLLECTION& collection,ADAPTER adapter){
collection_=&collection;
adapter_=adapter;
}
public: auto begin(){
return IteratorAdapter<
decltype(std::declval<COLLECTION>().begin()),
decltype(adapter_)>
(collection_->begin(),adapter_);
}
public: auto end(){ ..... }
};
IteratorAdapter (used above) encapsulates underlying iterator, change behavior of operator* :-
template<class ITERATORT,class ADAPTER>class IteratorAdapter : public ITERATORT {
ADAPTER adapter_;
public: IteratorAdapter(ITERATORT underlying,ADAPTER adapter) :
ITERATORT(underlying),
adapter_(adapter)
{ }
public: auto operator*(){
return adapter_(ITERATORT::operator*());
}
};
CollectAdapterWidget (used below) is just a helper class to construct CollectAdapter-instance.
It can be used like:-
int func1(int i){ return i+10; }
int main(){
std::vector<int> test; test.push_back(5);
for(auto b:CollectAdapterWidget::createAdapter(test,func1)){
//^ create "CollectAdapter<std::vector<int>,func1>" instance
//here, b=5+10=15
}
}
Problem
The above code works OK in most cases, except when COLLECTION is a temporary object.
More specifically, dangling pointer potentially occurs when I create adapter of adapter of adapter ....
int func1(int i){ return i+10; }
int func2(int i){ return i+100; }
template<class T> auto utilityAdapter(const T& t){
auto adapter1=CollectAdapterWidget::createAdapter(t,func1);
auto adapter12=CollectAdapterWidget::createAdapter(adapter1,func2);
//"adapter12.collection_" point to "adapter1"
return adapter12;
//end of scope, "adapter1" is deleted
//"adapter12.collection_" will be dangling pointer
}
int main(){
std::vector<int> test;
test.push_back(5);
for(auto b:utilityAdapter(test)){
std::cout<< b<<std::endl; //should 5+10+100 = 115
}
}
This will cause run time error. Here is the dangling-pointer demo.
In the real usage, if the interface is more awesome, e.g. use | operator, the bug will be even harder to be detected :-
//inside "utilityAdapter(t)"
return t|func1; //OK!
return t|func1|func2; //dangling pointer
Question
How to improve my library to fix this error while keeping performance & robustness & maintainablilty near the same level?
In other words, how to cache data or pointer of COLLECTION (that can be adapter or real data-structure) elegantly?
Alternatively, if it is easier to answer by coding from scratch (than modifying my code), go for it. :)
My workarounds
The current code caches by pointer.
The main idea of workarounds is to cache by value instead.
Workaround 1 (always "by value")
Let adapter cache the value of COLLECTION.
Here is the main change:-
COLLECTION collection_; //<------ #1
//changed from .... COLLECTION* collection_;
Disadvantage:-
Whole data-structure (e.g. std::vector) will be value-copied - waste resource.
(when use for std::vector directly)
Workaround 2 (two versions of library, best?)
I will create 2 versions of the library - AdapterValue and AdapterPointer.
I have to create related classes (Widget,AdapterIterator,etc.) as well.
AdapterValue - by value. (designed for utilityAdapter())
AdapterPointer - by pointer. (designed for std::vector)
Disadvantage:-
Duplicate code a lot = low maintainability
Users (coders) have to be very conscious about which one to pick = low robustness
Workaround 3 (detect type)
I may use template specialization that do this :-
If( COLLECTION is an "CollectAdapter" ){ by value }
Else{ by pointer }
Disadvantage:-
Not cooperate well between many adapter classes.
They have to recognize each other : recognized = should cache by value.
Sorry for very long post.
I personally would go with template specialisation – however, not specialise the original template, but a nested class instead:
template<typename Collection, typename Adapter>
class CollectAdapter
{
template<typename C>
class ObjectKeeper // find some better name yourself...
{
C* object;
public:
C* operator*() { return object; };
C* operator->() { return object; };
};
template<typename C, typename A>
class ObjectKeeper <CollectAdapter<C, A>>
{
CollectAdapter<C, A> object;
public:
CollectAdapter<C, A>* operator*() { return &object; };
CollectAdapter<C, A>* operator->() { return &object; };
};
ObjectKeeper<Collection> keeper;
// now use *keeper or keeper-> wherever needed
};
The outer class then covers both cases by just always using pointers while the nested class hides the differences away.
Sure, incomplete (you yet need to add appropriate constructors, for instance, both to outer and inner class), but it should give you the idea...
You might even allow the user to select if she/he wants to copy:
template<typename Collection, typename Adapter, bool IsAlwaysCopy = false>
class CollectAdapter
{
template<typename C, bool IsCopy>
class ObjectWrapper // find some better name yourself...
{
C* object;
public:
C* operator*() { return object; };
C* operator->() { return object; };
};
template<typename C>
class ObjectWrapper<C, true>
{
C object;
public:
C* operator*() { return &object; };
C* operator->() { return &object; };
};
// avoiding code duplication...
template<typename C, bool IsCopy>
class ObjectKeeper : public ObjectWrapper<C, IsCopy>
{ };
template<typename C, typename A, bool IsCopy>
class ObjectKeeper <CollectAdapter<C, A>, IsCopy>
: public ObjectWrapper<CollectAdapter<C, A>, true>
{ };
ObjectKeeper<Collection> keeper;
};
In my indexed_view I store the value of the collection if it is an rvalue, and store a reference if it is an lvalue. You could do the same here: overload your operator| for both rvalues and lvalues.
template<typename Collection,typename Filter>
auto operator|(Collection&& collection,Filter filter){
return create_adapter_for_rvalue_collection(collection,filter);
}
template<typename Collection,typename Filter>
auto operator|(Collection const& collection,Filter filter){
return create_adapter_for_const_lvalue_collection(collection,filter);
}
template<typename Collection,typename Filter>
auto operator|(Collection & collection,Filter filter){
return create_adapter_for_non_const_lvalue_collection(collection,filter);
}
For example I have some function pet_maker() that creates and returns a Cat or a Dog as a base Pet. I want to call this function many many times, and do something with the Pet returned.
Traditionally I would new the Cat or Dog in pet_maker() and return a pointer to it, however the new call is much slower than doing everything on the stack.
Is there a neat way anyone can think of to return as an abstraction without having to do the new every time the function is called, or is there some other way that I can quickly create and return abstractions?
Using new is pretty much inevitable if you want polymorphism. But the reason new works slowly is because it looks for free memory every time. What you could do is write your own operator new, which could, in theory, for example use pre-allocated memory chunks and be very fast.
This article covers many aspects of what you might need.
Each allocation is an overhead so you may get benefits by allocating whole arrays of objects rather than one object at a time.
You could use std::deque to achieve this:
class Pet { public: virtual ~Pet() {} virtual std::string talk() const = 0; };
class Cat: public Pet { std::string talk() const override { return "meow"; }};
class Dog: public Pet { std::string talk() const override { return "woof"; }};
class Pig: public Pet { std::string talk() const override { return "oink"; }};
class PetMaker
{
// std::deque never re-allocates when adding
// elements which is important when distributing
// pointers to the elements
std::deque<Cat> cats;
std::deque<Dog> dogs;
std::deque<Pig> pigs;
public:
Pet* make()
{
switch(std::rand() % 3)
{
case 0:
cats.emplace_back();
return &cats.back();
case 1:
dogs.emplace_back();
return &dogs.back();
}
pigs.emplace_back();
return &pigs.back();
}
};
int main()
{
std::srand(std::time(0));
PetMaker maker;
std::vector<Pet*> pets;
for(auto i = 0; i < 100; ++i)
pets.push_back(maker.make());
for(auto pet: pets)
std::cout << pet->talk() << '\n';
}
The reason to use a std::deque is that it never reallocates its elements when you add new ones so the pointers that you distribute always remain valid until the PetMaker itself is deleted.
An added benefit to this over allocating objects individually is that they don't need to be deleted or placed in a smart pointer, the std::deque manages their lifetime.
Is there a neat way anyone can think of to return as an abstraction without having to do the new every time the function is called, or is there some other way that I can quickly create and return abstractions?
TL;DR: The function need not allocate if there is already sufficient memory to work with.
A simple way would be to create a smart pointer that is slightly different from its siblings: it would contain a buffer in which it would store the object. We can even make it non-nullable!
Long version:
I'll present the rough draft in reverse order, from the motivation to the tricky details:
class Pet {
public:
virtual ~Pet() {}
virtual void say() = 0;
};
class Cat: public Pet {
public:
virtual void say() override { std::cout << "Miaou\n"; }
};
class Dog: public Pet {
public:
virtual void say() override { std::cout << "Woof\n"; }
};
template <>
struct polymorphic_value_memory<Pet> {
static size_t const capacity = sizeof(Dog);
static size_t const alignment = alignof(Dog);
};
typedef polymorphic_value<Pet> any_pet;
any_pet pet_factory(std::string const& name) {
if (name == "Cat") { return any_pet::build<Cat>(); }
if (name == "Dog") { return any_pet::build<Dog>(); }
throw std::runtime_error("Unknown pet name");
}
int main() {
any_pet pet = pet_factory("Cat");
pet->say();
pet = pet_factory("Dog");
pet->say();
pet = pet_factory("Cat");
pet->say();
}
The expected output:
Miaou
Woof
Miaou
which you can find here.
Note that it is required to specify the maximum size and alignment of the derived values that can be supported. No way around that.
Of course, we statically check whether the caller would attempt to build a value with an inappropriate type to avoid any unpleasantness.
The main disadvantage, of course, is that it must be at least as big (and aligned) as its largest variant, and all this must be predicted ahead of time. This is thus not a silver bullet, but performance-wise the absence of memory-allocation can rock.
How does it work? Using this high-level class (and the helper):
// To be specialized for each base class:
// - provide capacity member (size_t)
// - provide alignment member (size_t)
template <typename> struct polymorphic_value_memory;
template <typename T,
typename CA = CopyAssignableTag,
typename CC = CopyConstructibleTag,
typename MA = MoveAssignableTag,
typename MC = MoveConstructibleTag>
class polymorphic_value {
static size_t const capacity = polymorphic_value_memory<T>::capacity;
static size_t const alignment = polymorphic_value_memory<T>::alignment;
static bool const move_constructible = std::is_same<MC, MoveConstructibleTag>::value;
static bool const move_assignable = std::is_same<MA, MoveAssignableTag>::value;
static bool const copy_constructible = std::is_same<CC, CopyConstructibleTag>::value;
static bool const copy_assignable = std::is_same<CA, CopyAssignableTag>::value;
typedef typename std::aligned_storage<capacity, alignment>::type storage_type;
public:
template <typename U, typename... Args>
static polymorphic_value build(Args&&... args) {
static_assert(
sizeof(U) <= capacity,
"Cannot host such a large type."
);
static_assert(
alignof(U) <= alignment,
"Cannot host such a largely aligned type."
);
polymorphic_value result{NoneTag{}};
result.m_vtable = &build_vtable<T, U, MC, CC, MA, CA>();
new (result.get_ptr()) U(std::forward<Args>(args)...);
return result;
}
polymorphic_value(polymorphic_value&& other): m_vtable(other.m_vtable), m_storage() {
static_assert(
move_constructible,
"Cannot move construct this value."
);
(*m_vtable->move_construct)(&other.m_storage, &m_storage);
m_vtable = other.m_vtable;
}
polymorphic_value& operator=(polymorphic_value&& other) {
static_assert(
move_assignable || move_constructible,
"Cannot move assign this value."
);
if (move_assignable && m_vtable == other.m_vtable)
{
(*m_vtable->move_assign)(&other.m_storage, &m_storage);
}
else
{
(*m_vtable->destroy)(&m_storage);
m_vtable = other.m_vtable;
(*m_vtable->move_construct)(&other.m_storage, &m_storage);
}
return *this;
}
polymorphic_value(polymorphic_value const& other): m_vtable(other.m_vtable), m_storage() {
static_assert(
copy_constructible,
"Cannot copy construct this value."
);
(*m_vtable->copy_construct)(&other.m_storage, &m_storage);
}
polymorphic_value& operator=(polymorphic_value const& other) {
static_assert(
copy_assignable || (copy_constructible && move_constructible),
"Cannot copy assign this value."
);
if (copy_assignable && m_vtable == other.m_vtable)
{
(*m_vtable->copy_assign)(&other.m_storage, &m_storage);
return *this;
}
// Exception safety
storage_type tmp;
(*other.m_vtable->copy_construct)(&other.m_storage, &tmp);
if (move_assignable && m_vtable == other.m_vtable)
{
(*m_vtable->move_assign)(&tmp, &m_storage);
}
else
{
(*m_vtable->destroy)(&m_storage);
m_vtable = other.m_vtable;
(*m_vtable->move_construct)(&tmp, &m_storage);
}
return *this;
}
~polymorphic_value() { (*m_vtable->destroy)(&m_storage); }
T& get() { return *this->get_ptr(); }
T const& get() const { return *this->get_ptr(); }
T* operator->() { return this->get_ptr(); }
T const* operator->() const { return this->get_ptr(); }
T& operator*() { return this->get(); }
T const& operator*() const { return this->get(); }
private:
polymorphic_value(NoneTag): m_vtable(0), m_storage() {}
T* get_ptr() { return reinterpret_cast<T*>(&m_storage); }
T const* get_ptr() const { return reinterpret_cast<T const*>(&m_storage); }
polymorphic_value_vtable const* m_vtable;
storage_type m_storage;
}; // class polymorphic_value
Essentially, this is just like any STL container. The bulk of the complexity is in redefining the construction, move, copy and destruction. It's otherwise quite simple.
There are two points of note:
I use a tag-based approach to handling capabilities:
for example, a copy constructor is only available if the CopyConstructibleTag is passed
if the CopyConstructibleTag is passed, all types passed to build must be copy constructible
Some operations are provided even if the objects do not have the capability, as long as some alternative way of providing them exist
Obviously, all methods preserve the invariant that the polymorphic_value is never empty.
There is also a tricky detail related to assignments: assignment is only well-defined if both objects are of the same dynamic type, which we check with the m_vtable == other.m_vtable checks.
For completeness, the missing pieces used to power up this class:
//
// VTable, with nullable methods for run-time detection of capabilities
//
struct NoneTag {};
struct MoveConstructibleTag {};
struct CopyConstructibleTag {};
struct MoveAssignableTag {};
struct CopyAssignableTag {};
struct polymorphic_value_vtable {
typedef void (*move_construct_type)(void* src, void* dst);
typedef void (*copy_construct_type)(void const* src, void* dst);
typedef void (*move_assign_type)(void* src, void* dst);
typedef void (*copy_assign_type)(void const* src, void* dst);
typedef void (*destroy_type)(void* dst);
move_construct_type move_construct;
copy_construct_type copy_construct;
move_assign_type move_assign;
copy_assign_type copy_assign;
destroy_type destroy;
};
template <typename Base, typename Derived>
void core_move_construct_function(void* src, void* dst) {
Derived* derived = reinterpret_cast<Derived*>(src);
new (reinterpret_cast<Base*>(dst)) Derived(std::move(*derived));
} // core_move_construct_function
template <typename Base, typename Derived>
void core_copy_construct_function(void const* src, void* dst) {
Derived const* derived = reinterpret_cast<Derived const*>(src);
new (reinterpret_cast<Base*>(dst)) Derived(*derived);
} // core_copy_construct_function
template <typename Derived>
void core_move_assign_function(void* src, void* dst) {
Derived* source = reinterpret_cast<Derived*>(src);
Derived* destination = reinterpret_cast<Derived*>(dst);
*destination = std::move(*source);
} // core_move_assign_function
template <typename Derived>
void core_copy_assign_function(void const* src, void* dst) {
Derived const* source = reinterpret_cast<Derived const*>(src);
Derived* destination = reinterpret_cast<Derived*>(dst);
*destination = *source;
} // core_copy_assign_function
template <typename Derived>
void core_destroy_function(void* dst) {
Derived* d = reinterpret_cast<Derived*>(dst);
d->~Derived();
} // core_destroy_function
template <typename Tag, typename Base, typename Derived>
typename std::enable_if<
std::is_same<Tag, MoveConstructibleTag>::value,
polymorphic_value_vtable::move_construct_type
>::type
build_move_construct_function()
{
return &core_move_construct_function<Base, Derived>;
} // build_move_construct_function
template <typename Tag, typename Base, typename Derived>
typename std::enable_if<
std::is_same<Tag, CopyConstructibleTag>::value,
polymorphic_value_vtable::copy_construct_type
>::type
build_copy_construct_function()
{
return &core_copy_construct_function<Base, Derived>;
} // build_copy_construct_function
template <typename Tag, typename Derived>
typename std::enable_if<
std::is_same<Tag, MoveAssignableTag>::value,
polymorphic_value_vtable::move_assign_type
>::type
build_move_assign_function()
{
return &core_move_assign_function<Derived>;
} // build_move_assign_function
template <typename Tag, typename Derived>
typename std::enable_if<
std::is_same<Tag, CopyAssignableTag>::value,
polymorphic_value_vtable::copy_construct_type
>::type
build_copy_assign_function()
{
return &core_copy_assign_function<Derived>;
} // build_copy_assign_function
template <typename Base, typename Derived,
typename MC, typename CC,
typename MA, typename CA>
polymorphic_value_vtable const& build_vtable() {
static polymorphic_value_vtable const V = {
build_move_construct_function<MC, Base, Derived>(),
build_copy_construct_function<CC, Base, Derived>(),
build_move_assign_function<MA, Derived>(),
build_copy_assign_function<CA, Derived>(),
&core_destroy_function<Derived>
};
return V;
} // build_vtable
The one trick I use here is to let the user configure whether the types he will use in this container can be move constructed, move assigned, ... via capability tags. A number of operations are keyed on these tags and will either be disabled or less efficient if the requested capability
You can create a stack allocator instance (with some max limit of course) and pass that as an argument to your pet_maker function. Then instead of regular new do a placement new on the address provided by the stack allocator.
You can probably also default to new on exceeding max_size of the stack allocator.
One way is to work out, in advance through analysis, how many of each type of object is needed by your program.
Then you can allocate arrays of an appropriate size in advance, as long as you have book-keeping to track the allocation.
For example;
#include <array>
// Ncats, Ndogs, etc are predefined constants specifying the number of cats and dogs
std::array<Cat, Ncats> cats;
std::array<Dog, Ndogs> dogs;
// bookkeeping - track the returned number of cats and dogs
std::size_t Rcats = 0, Rdogs = 0;
Pet *pet_maker()
{
// determine what needs to be returned
if (return_cat)
{
assert(Rcats < Ncats);
return &cats[Rcats++];
}
else if (return_dog)
{
assert(Rdogs < Ndogs);
return &dogs[Rdogs++];
}
else
{
// handle other case somehow
}
}
Of course, the big trade-off in is the requirement to explicitly determine the number of each type of animal in advance - and separately track each type.
However, if you wish to avoid dynamic memory allocation (operator new) then this way - as draconian as it might seem - provides an absolute guarantee. Using operator new explicitly allows the number of objects needed to be determined at run time. Conversely, to avoid using operator new but allow some function to safely access a number of objects it is necessary to predetermine the number of objects.
It depends on the exact use case you have, and what restrictions you are willing to tolerate. For example, if you are OK with re-using the same objects rather than having new copies every time, you could return references to static objects inside the function:
Pet& pet_maker()
{
static Dog dog;
static Cat cat;
//...
if(shouldReturnDog) {
//manipulate dog as necessary
//...
return dog;
}
else
{
//manipulate cat as necessary
//...
return cat;
}
}
This works if the client code accepts that it doesn't own the object returned and that the same physical instances are reused.
There are other tricks possible if this particular set of assumptions is unsuitable.
At some point somebody is going to have to allocate the memory and initialize the objects. If doing them on demand, using the heap via new is taking too long, then why no pre-allocate a number of then in a pool. Then you can initialize each individual object on an as needed basis. The downside is that you might have a bunch of extra objects laying around for a while.
If actually initializing the object is the problem, and not memory allocation, then you can consider keeping a pre-built object around and using the Pototype pattern for quicker initialization.
For best results, memory allocation is problem and initialization time, you can combine both strategies.
You may want to consider using a (Boost) variant. It will require an extra step by the caller, but it might suit your needs:
#include <boost/variant/variant.hpp>
#include <boost/variant/get.hpp>
#include <iostream>
using boost::variant;
using std::cout;
struct Pet {
virtual void print_type() const = 0;
};
struct Cat : Pet {
virtual void print_type() const { cout << "Cat\n"; }
};
struct Dog : Pet {
virtual void print_type() const { cout << "Dog\n"; }
};
using PetVariant = variant<Cat,Dog>;
enum class PetType { cat, dog };
PetVariant make_pet(PetType type)
{
switch (type) {
case PetType::cat: return Cat();
case PetType::dog: return Dog();
}
return {};
}
Pet& get_pet(PetVariant& pet_variant)
{
return apply_visitor([](Pet& pet) -> Pet& { return pet; },pet_variant);
}
int main()
{
PetVariant pet_variant_1 = make_pet(PetType::cat);
PetVariant pet_variant_2 = make_pet(PetType::dog);
Pet& pet1 = get_pet(pet_variant_1);
Pet& pet2 = get_pet(pet_variant_2);
pet1.print_type();
pet2.print_type();
}
Output:
Cat
Dog
For example I have some function pet_maker() that creates and returns a Cat or a Dog as a base Pet. I want to call this function many many times, and do something with the Pet returned.
If you are going to discard the pet immediately after you have done something with it, you can use the technique shown in the following example:
#include<iostream>
#include<utility>
struct Pet {
virtual ~Pet() = default;
virtual void foo() const = 0;
};
struct Cat: Pet {
void foo() const override {
std::cout << "cat" << std::endl;
}
};
struct Dog: Pet {
void foo() const override {
std::cout << "dog" << std::endl;
}
};
template<typename T, typename F>
void factory(F &&f) {
std::forward<F>(f)(T{});
}
int main() {
auto lambda = [](const Pet &pet) { pet.foo(); };
factory<Cat>(lambda);
factory<Dog>(lambda);
}
No allocation required at all. The basic idea is to revert the logic: the factory no longer returns an object. Instead it calls a function providing the right instance as a reference.
The problem with this approach arises if you want to copy and store the object somewhere.
For it is not clear from the question, it's worth to propose also this solution.
In C++03, when you were to wrap a bunch of C functions in a class to create an 'auto object', you had to customize the object to the type of functions it encapsulated. As an example, to wrap a windows file HANDLE, you needed to call CloseHandle() in the destructor and CreateFile() in the constructor. The constructor would need to mimic the function signature of the CreateFile() function, sans the file HANDLE variable (since it's being managed).
Anyway, what I'd like to know is if it's possible to use the new features of C++11 to create a single generic class that can be used to wrap any type of resource by only providing an implementation for creation and deletion?
One problem I foresee is that the creation function, such as noted above with CreateFile(), can taken any number of parameters. Is there a way to auto-magically generate a templated constructor that mimics the signature of the function? Variadic Parameters come to mind, but I have not yet used them.
Has anyone tried writing something like this?
EDIT: Some code to help illustrate (pseudo):
template<typename Res, FunctionPtrToCreatorFunc Func, typename... Arguments>
class creator
{
public:
operator()(Res &r, Arguments... Args)
{
Func(r, /*use args?*/ Args); // Allocate resource, ie. CreateFile(r, args)
}
};
template<typename Res, FunctionPtrToDeleterFunc Func>
class deleter
{
operator()(Res &r)
{
Func(r); // delete the resource, ie. CloseHandle(r)
}
};
Then this will be the implementation of my super auto object:
template<typename Res, typename Creator, typename Deleter>
class auto_obj
{
public:
auto_obj(/*somehow copy Args from Creator class?*/)
{
Creator(_res, /*args?*/);
}
~auto_obj()
{
deleter(_res);
}
Res _res;
};
Yes, this has a similar structure to shared_ptr or unique_ptr, but instead the constructor will be the one that creates the resources by developer written creator and deleter classes. I have a feeling that std::bind may play a role in this, but I have never used it.
Here is a stab at it:
#include <utility>
#include <type_traits>
#include <cstddef>
A more friendly way to wrap up a function. I move the signature boilerplate to this template, instead of messing up the actual RAII class below. This also allows full fledged function objects to be used, as well as functions, in the RAII class below:
template< typename FuncSig, FuncSig func >
struct Functor {
template<typename... Args>
auto operator()(Args&&... args) const
-> decltype( func(std::forward<Args>(args)...) )
{ return ( func(std::forward<Args>(args)...) ); }
};
One operation that is needed for more than basic functionality is the ability to "null" a handle, allowing invalid handles to exist, and allowing handles to be moved around. Zeroer is my default function object for "null"ing a handle:
struct Zeroer {
template<typename T>
void operator()( T& t ) const {
t = 0;
}
};
RAII_handle herself. You pack the creation and destruction signatures into it, and it forwards construction to the underlying data. .close() lets you close the RAII_handle early, which is a common requirement in practice. You access the underlying data via operator* or operator->, and while this makes it look pointer-like, RAII_handle does not obey pointer semantics. It is a move-only type.
template< typename T, typename Creator, typename Destroyer, typename Nuller=Zeroer >
struct RAII_handle {
RAII_handle( std::nullptr_t ):
data()
{
Nuller()(data);
}
RAII_handle( RAII_handle const& ) = delete;
RAII_handle( RAII_handle && o ):data(std::move(o.data)) {
Nuller()(o.data);
}
RAII_handle& operator=( RAII_handle const& ) = delete;
RAII_handle& operator=( RAII_handle && o ) {
data = std::move(o.data);
Nuller()(o.data);
return *this;
}
template<typename... Args>
RAII_handle( Args&&... args ):
data( Creator()(std::forward<Args>(args)...) )
{}
auto close()->decltype( Destroyer()(std::declval<T&>()) ) {
auto retval = Destroyer()(data);
Nuller()(data);
return retval;
}
~RAII_handle() {
close();
}
T& get() { return data; }
T const& get() const { return data; }
T& operator*() { return get(); }
T const& operator*() const { return get(); }
T* operator->() { return &get(); }
T const* operator->() const { return &get(); }
private:
T data;
};
Now, some test code. My file handles will be unsigned char, and opening/closing will simply test if things are not working right.
#include <iostream>
typedef unsigned char HANDLE;
HANDLE CreateFile( char const* name ) {
std::cout << name << "\n";
return 7;
}
bool CloseFile( HANDLE h ) {
if (h) {
--h;
std::cout << (int)h << "\n";
return true;
} else {
std::cout << "already closed\n";
return true;
}
}
Once you have your open/close functions or function objects, here is how you make the type of the FileHandle:
typedef RAII_handle< HANDLE, Functor< HANDLE(*)( char const* ), CreateFile >, Functor< bool(*)(HANDLE), CloseFile > > FileHandle;
You can support entire overload sets by simply creating a function object that forwards to a fixed function name, instead of to a fixed function pointer. Basically take Functor above, remove the template signature and pointer, and replace the use of func with actual use of your function name.
Suddenly your function object represents not calling one function, but calling an entire overload set.
Fancier work can even support multiple functions, allowing one function object to support calling either CreateFile or CreateFileEx depending on what arguments are passed in.
And here is our trivial test code:
int main() {
FileHandle bob("hello.txt");
HANDLE value = *bob; // get the HANDLE out of the FileHandle
bob.close(); // optional, to close early
}
Requirements: your CloseFile must accept Nuller()(std::declval<T&>()) and not behave badly. The default Nuller()(...) just assigns zero to your T, which works for many handle types.
It supports move semantics, allowing you to return these from a function, but I didn't include a Copier argument (which I'd expect would be required for any RAII objects that can be copied).
Live example with very slightly different code.