In N3257 I found an example using initializing members without a constructor, which is fine. I guess that is possible, because it is a POD.
template<typename T>
struct adaptor {
NonStdContainer<T>* ptr; // <- data member
T* begin() { return ptr->getFirst(); }
T* end() { return ptr->getLast() + 1; }
};
void f(NonStdContainer<int>& c) {
for (auto i : adaptor<int>{&c}) // <- init
{ /* ... */ }
}
When I played around with this example I replaced the * with a &, because I don't like raw pointers:
template<typename T>
struct adaptor {
NonStdContainer<T>& ptr; // <- data member, now REF
T* begin() { return ptr->getFirst(); }
T* end() { return ptr->getLast() + 1; }
};
void f(NonStdContainer<int>& c) {
for (auto i : adaptor<int>{c}) // <- init
{ /* ... */ }
}
This was fine and compiled without warning with GCC-4.7.0.
Then I got curious about the initialization of PODs and what might have changed with C++0x.
There I found Bjarnes FAQ. He says there that PODs may contain pointers, but no references.
Ops, now I wonder:
Do I have non-POD-object here, which the compiler can initialize without a constructor anyway and I just miss which mechanisms are used here?
or Is the GCC-4.7.0 behaving non-std by letting me initializing the ref this way?
or has there been a change in the std since Bjarnes FAQ that also allows references in PODs?
Update: I found aggregates in the current std (8.5.1 Aggregates [dcl.init.aggr]), but references are not mentioned there, so I am not sure how they relate to this
Quoting the standard [dcl.init.aggr]:
An aggregate is an array or a class (Clause 9) with no user-provided
constructors (12.1), no brace-or-equal- initializers for non-static
data members (9.2), no private or protected non-static data members
(Clause 11), no base classes (Clause 10), and no virtual functions
(10.3).
When an aggregate is initialized by an initializer list, as specified
in 8.5.4, the elements of the initializer list are taken as
initializers for the members of the aggregate, in increasing subscript
or member order. Each member is copy-initialized from the corresponding initializer-clause...
That means you have an aggregate here, aggregates can be initialized how you do it. PODs have nothing to do with it, they are really meant for communication with eg. C.
Copy-initialization of a reference with a variable is certainly legal, because that just means
T& ref = c;
Do I have non-POD-object here, which the compiler can initialize without a constructor anyway and I just miss which mechanisms are used here?
Yes, the object is non-POD.
Is the GCC-4.7.0 behaving non-std by letting me initializing the ref this way?
No.
Related
I am aware of the lack of reflection and basic template mechanics in C++ so the example below can't work. But maybe there's a hack to achieve the intended purpose in another way?
template <typename OwnerClass>
struct Template
{
OwnerClass *owner;
};
struct Base
{
virtual void funct ()
{
Template <decltype(*this)> temp;
// ...
}
};
struct Derived : public Base
{
void whatever ()
{
// supposed to infer this class and use Template<Derived>
// any chance some macro or constexpr magic could help?
funct();
}
};
In the example, Derived::whatever() calls virtual method Base::funct() and wants it to pass its own class name (Derived) to a template. The compiler complains "'owner' declared as a pointer to a reference of type 'Base &'". Not only does decltype(*this) not provide a typename but a reference, the compiler also can't know in advance that funct is called from Derived, which would require funct() to be made a template.
If funct() was a template however, each derived class needs to pass its own name with every call, which is pretty verbose and redundant.
Is there any hack to get around this limitation and make calls to funct() infer the typename of the calling class? Maybe constexpr or macros to help the compiler infer the correct type and reduce verbosity in derived classes?
You should use CRTP Pattern (Curiously Recurring Template Pattern) for inheritance.
Define a base class:
struct CBase {
virtual ~CBase() {}
virtual void function() = 0;
};
Define a prepared to CRTP class:
template<typename T>
struct CBaseCrtp : public CBase {
virtual ~CBaseCrtp() {}
void function() override {
using DerivedType = T;
//do stuff
}
};
Inherit from the CRTP one:
struct Derived : public CBaseCrtp<Derived> {
};
It should work. The only way to know the Derived type is to give it to the base!
Currently, this can't be done. Base is a Base and nothing else at the time Template <decltype(*this)> is instantiated. You are trying to mix the static type system for an inheritance hierarchy inherently not resolved before runtime. This very same mechanism is the reason for not calling virtual member functions of an object during its construction.
At some point, this limitation might change in the future. One step towards this is demonstrated in the Deducing this proposal.
I know that additional initialization methods are evil, as they leave a very nasty option for having object half-constructed and as result all methods needs to check for this. But what about this situation?
class config;
class cfg_item final
{
private:
friend class config;
cfg_item(std::weak_ptr<config> owner) : owner(owner) { }
std::weak_ptr<config> owner;
}
class config final : private std::enable_shared_from_this<config>
{
public:
config()
{
items.emplace(std::make_shared<cfg_item>(weak_from_this())); // Will crash!
}
private:
std::vector<std::shared_ptr<cfg_item>> items;
}
int main(int argc, char * argv[])
{
std::shared_ptr<config> cfg = std::make_shared<config>();
}
I KNOW WHY IT CRASHES. The std::shared_ptr in the main is not yet initialized with shared pointer to config instance, so constructor does not know how to make weak_from_this and just raises std::bad_weak_ptr exception because there are no valid std::shared_ptr pointing to this at constructor's call time.
The question is: how can I avoided the whole thing? I believe the only way I see would be to add separate initialization method, which is evil as I've already mentioned...
As note about real code: the constructors loads cfg_item from external source. It is assumed that all cfg_items are available for the entire lifetime of config. The weak pointers back to config are mandatory, as cfg_item must push all changes done to it back to config to save to external source
If you look at the answers to this question, there are strong arguments why an external initialization function is necessary. However, you rightfully write
I know that additional initialization methods are evil, as they leave a very nasty option for having object half-constructed and as result all methods needs to check for this.
it's possible to reduce this problem. Say you have a class foo, with the protocol that each time a foo object is constructed, foo::init() needs to be called. Obviously, this is a brittle class (client code will eventually omit calls to init()).
So, one way is to make the (non-copy / non-move) constructors of foo private, and create a variadic static factory method that creates objects, then calls init():
#include <utility>
class foo {
private:
foo() {}
foo(int) {}
void init() {}
public:
template<typename ...Args>
static foo create(Args &&...args) {
foo f{std::forward<Args>(args)...};
f.init();
return f;
}
};
In the following code
template<typename ...Args>
static foo create(Args &&...args) {
foo f{std::forward<Args>(args)...};
f.init();
return f;
}
note that this single method can be used for all constructors, regardless of their signature. Furthermore, since it is static, it is external to the constructor, and doesn't have the problems in your question.
You can use it as follows:
int main() {
auto f0 = foo::create();
auto f1 = foo::create(2);
// Next line doesn't compile if uncommented
// foo f2;
}
Note that it's impossible to create an object without this method, and the interface doesn't even contain init.
Consider this class:
template<typename T> struct pooled_resource: T{
template<typename... Args> pooled_resource(std::list<T>& pool, Args&&... args):
T(select_resource(pool, std::forward<Args>(args)...)), pool(pool){
if(!pool.empty()) pool.pop_front();
}
~pooled_resource(){
pool.push_front(static_cast<T&&>(*this));
}
private:
std::list<T>& pool;
template<typename... Args> static T select_resource(std::list<T>& pool, Args&&... args){
if(pool.empty())
return T(std::forward<Args>(args)...);
else
return std::move(pool.front());
}
};
It allows to create variables that have pooled resources but are semantically equivalent to the non-pooled version:
std::list<std::vector<int>> pool;
using pooled_vector = pooled_resource<std::vector<int>>;
{
pooled_vector a(pool); //pool is empty, allocate new resources
{ pooled_vector b(pool, 100); } //allocate again, but then release to pool as b goes out of scope
pooled_vector c(pool); //reuse b's resources
assert(c.size() == 100);
}
My actual problem is that for the simple case above, everything works and the move constructor of the pooled resource is called. However, I'm getting for another class that the move constructor is not called, but rather the copy constructor is. Precisely, the class is boost::compute::vector, that does declare a move constructor which seems to work in simple cases such as boost::compute::vector<int> a; boost::compute::vector<int> b(std::move(a));.
I have no clue why this happens and I don't know how to diagnose it: what am I missing regarding the rules for a move constructor to be actually used?
So the problem was a silly mistake in the class of the pooled resource. This is a sketch of it:
template<typename T, typename Allocator = DefaultAllocator> struct Resource{
/*...*/
Resource(Resource<T>&& o){ /*...*/ }
}
The problem was that the argument of the move constructor is of type Resource<T, DefaultAllocator>. Since I was using some custom allocator, there was actually no template move constructor available with a generic Allocator, but just with a generic type T and allocator DefaultAllocator. The fix just requires leaving out all template specifications of the move constructor:
Resource(Resource&& o){ /*...*/ }
Hope this can save somebody's else afternoon.
I searched many pages, and I think I have known how to write the std::hash. But I don't know where to put it.
An example is presented here http://en.cppreference.com/w/cpp/utility/hash .
However, I defined my type Instance in namespace ca in file instance_management.h. I want to use unordered_set<Instance> in the same file in another class InstanceManager. So I write the following code:
namespace std
{
template <> struct hash<ca::Instance>
{
size_t operator()(const ca::Instance & instance) const
{
std::size_t seed = 0;
// Some hash value calculation here.
return seed;
}
};
} // namespace std
But where should I put it? I tried many locations but all failed.
I am using visual studio 2013. I tried to put the previous code in some locations but all failed to compile it.
// location 1
namespace ca
{
class Instance {...}
class InstanceManager
{
// ... some other things.
private unordered_set<Instance>;
}
}
// location 2
There are several ways.
Specializing std::hash
In your code make sure that your std::hash<Instance> specialization is preceded immediately by the Instance class definition, and followed by the use of the unordered_set container that uses it.
namespace ca
{
class Instance {...};
}
namespaces std {
template<> hash<Instance> { ... };
}
namespace ca {
class InstanceManager
{
// ... some other things.
private unordered_set<Instance>;
}
}
One drawback is that you can have funny name lookup interference when passing a std::hash<ca::Instance> to other functions. The reason is that the associated namespace (ca) of all the template arguments of std::hash can be used during name lookup (ADL). Such errors are a bit rare, but if they occur they can be hard to debug.
See this Q&A for more details.
Passing your hash to unordered_set
struct MyInstanceHash { ... };
using MyUnorderedSet = std:unordered_set<Instance, MyInstanceHash>;
Here, you simply pass your own hash function to the container and be done with it. The drawback is that you have to explicitly type your own container.
Using hash_append
Note, however, there is the N3980 Standard proposal is currently pending for review. This proposal features a much superior design that uses a universal hash function that takes an arbitrary byte stream to be hashed by its template parameter (the actual hashing algorithm)
template <class HashAlgorithm>
struct uhash
{
using result_type = typename HashAlgorithm::result_type;
template <class T>
result_type
operator()(T const& t) const noexcept
{
HashAlgorithm h;
using std::hash_append;
hash_append(h, t);
return static_cast<result_type>(h);
}
};
A user-defined class X then has to provide the proper hash_append through which it presents itself as a byte stream, ready to be hashed by the univeral hasher.
class X
{
std::tuple<short, unsigned char, unsigned char> date_;
std::vector<std::pair<int, int>> data_;
public:
// ...
friend bool operator==(X const& x, X const& y)
{
return std::tie(x.date_, x.data_) == std::tie(y.date_, y.data_);
}
// Hook into the system like this
template <class HashAlgorithm>
friend void hash_append(HashAlgorithm& h, X const& x) noexcept
{
using std::hash_append;
hash_append(h, x.date_);
hash_append(h, x.data_);
}
}
For more details, see the presentation by the author #HowardHinnant at CppCon14 (slides, video). Full source code by both the author and Bloomberg is available.
Do not specialise std::hash, instead write your own hash function object (see Edge_Hash below) and declare your unordered_set with two template arguments.
#include <unordered_set>
#include <functional>
namespace foo
{
// an edge is a link between two nodes
struct Edge
{
size_t src, dst;
};
// this is an example of symmetric hash (suitable for undirected graphs)
struct Edge_Hash
{
inline size_t operator() ( const Edge& e ) const
{
static std::hash<size_t> H;
return H(e.src) ^ H(e.dst);
}
};
// this keeps all edges in a set based on their hash value
struct Edge_Set
{
// I think this is what you're trying to do?
std::unordered_set<Edge,Edge_Hash> edges;
};
}
int main()
{
foo::Edge_Set e;
}
Related posts are, eg:
Inserting in unordered_set using custom hash function
Trouble creating custom hash function unordered_map
Thanks to everyone.
I have found the reason and solved the problem somehow: visual studio accepted the InstanceHash when I was defining instances_. Since I was changing the use of set to unordered_set, I forgot to specify InstanceHash when I tried to get the const_iterator, so this time the compiler tried to use the std::hash<> things and failed. But the compiler didn't locate the line using const_iterator, so I mistakenly thought it didn't accept InstanceHash when I was defining instances_.
I also tried to specialize the std::hash<> for class Instance. However, this specialization requires at least the declaration of class ca::Instance and some of its member functions to calculate the hash value. After this specialization, the definition of class ca::InstanceManage will use it.
I now generally put declarations and implementations of almost every classes and member functions together. So, the thing I need to do is probably to split the ca namespace scope to 2 parts and put the std{ template <> struct hash<ca::Instance>{...}} in the middle.
Sorry about the title, I couldn't come with a better one.
Suppose that I have a class with special delete semantics, which needs to call a function instead of been deleted by delete, let's call it releaseable_object:
struct releaseable_object
{
releaseable_object() : dummy_ptr(new int) {}
void Release()
{
std::cout << "Releasing releaseable object\n";
delete dummy_ptr;
}
int *const dummy_ptr;
};
And this releaseable_object is the base class of a bunch of other objects, each of them constructed by a factory which only returns pointers.
I'm trying to wrap each class into a std::unique_ptr with a custom deleter which call the releaseable_object::Release() function, so I've created a helper struct to handle some of the generic stuff:
// std::is_base_of<releaseable_object, T>::value must be true
template <typename T> struct Managed
{
using type = T;
static void deleter(type *object)
{
std::cout << "Release!\n";
object->Release();
};
using pointer = std::unique_ptr<T, decltype(deleter)>;
};
And then, a bunch of derived classes which does all the specific initializations and calls to te factory:
struct ManagedA : Managed<A>
{
using base = Managed<A>;
using base::pointer;
using base::deleter;
ManagedA(/* lots of parameters */) :
m_pointer(nullptr, deleter)
{
// do A specific stuff...
A *a = factory::CreateA(/* lots of parameters */);
// more A specific stuff...
// wrap the pointer:
m_pointer.reset(a);
}
pointer m_pointer;
};
If I try to compile the code above, it complains about the unique_ptr (demo here), I don't know what I'm doing wrong there, the error is about the instantiation of a tuple (the complete error log is in the ideone demo):
tuple: In instantiation of ‘struct std::_Head_base<1u, void(A*), false>’:
tuple:229:12: recursively required from ‘struct std::_Tuple_impl<1u, void(A*)>’
tuple:229:12: required from ‘struct std::_Tuple_impl<0u, A*, void(A*)>’
tuple:521:11: required from ‘class std::tuple<A*, void(A*)>’
bits/unique_ptr.h:127:57: required from ‘class std::unique_ptr<A, void(A*)>’
If I get rid of the m_pointer member then the compilation succeeds. I'm pretty lost with this, I'll be grateful of any hints about how to fix the compilation error.
Thanks for your attention.
The problem is that decltype(deleter) is a function type instead of a pointer-to-function type. Changing the pointer declaration to
using pointer = std::unique_ptr<T, decltype(deleter)*>; // or spell out void(*)(T*)
will fix it.
Be aware that a function object type is usually preferable to a function pointer type for a unique pointer deleter, since the function pointer must be stored in the object itself. i.e.,
sizeof(std::unique_ptr<foo*,void(*)(foo*)>) == sizeof(foo*) + sizeof(void(*)(foo*))
but most implementations will take advantage of the Empty Base Optimization if you use an empty deleter type:
struct deleter_type {
void operator () (foo*) {
// ...
}
};
sizeof(std::unique_ptr<foo*,deleter_type>) == sizeof(foo*)
Here's how your sample code would be written using a deleter type..