I have an application that consists of multiple tasks that share common data using shared memory. Up to now the data in shared memory look like that:
struct Store = {
int id;
Array<Module, 5> modules;
};
where Module is defined as
struct Module = {
uint32_t a;
char b[64];
Array<Component, 10> components;
};
This Store structure has a fixed size an can be easily used within shared memory.
But now I have to support other Modules, lets say ModuleA and ModuleB. Within the normal C++ context I would model these as:
struct ModuleBase {
// common informations
};
struct ModuleA : public ModuleBase {
// ...
};
struct ModuleB : public ModuleBase {
// ...
};
and replace Module by Module* in the Store.
But within the shared memory this is not so easy.
Accessing data in shared memory is easy for fix structures that's why a compile time array is used. I would like to have this property with my different module's.
Idea 1
union Module {
ModuleType type;
ModuleA moduleA;
ModuleB moduleB;
};
The problem is that my Module classes have constructors and that is not allowed inside the union. Access is easy using the type and then Module.moduleX
fix: remove need of constructors
Idea 2
Using a template that evaluates the maximum size of given classes, e.g.
const size_t max_module_size = MaxTMP<ModuleA, ModuleB>::value;
This is the size of the buffer I need to store the modules:
char ModuleBuffer[max_module_size];
(maybe the ModuleBuffer has to be encapsulated in a struct, for usage with Array)
Access is tricky, the content of ModuleBuffer has to be casted to ModuleBase and according to the type to ModuleX. That for I think I need some reinterpret_cast. And I also need to cast the 'ModuleX' in some way to put into the ModuleBuffer.
Question
I don't like both ideas but I cannot imagine another way to handle this problem. Do you have any comments or - even better - solutions?
Effectively, you are between a rock and a hard place.
I would give a try to boost::variant, because of the facilities it comes with, otherwise it's not too difficult to recreate a similar thing, but it is long...
On top of size, you also need to take care about alignment. It will help to use C++11 here, although it is possible to write this in C++03 with a couple libraries/extensions.
Note that a union is not anything so special, and you can easily implement your own, in a way, and like boost::variant make it "tagged".
A couple helpers will help nicely:
/// Size and Alignment utilties
constexpr size_t max(size_t t) { return t; }
template <typename... U>
constexpr size_t max(size_t l, size_t r, U... tail) {
return l > max(r, tail...) ? l : max(r, tail...);
}
template <typename... T>
struct size { static size_t const value = max(sizeof(T)...); };
template <typename... T>
struct alignment { static size_t const value = max(alignof(T)...); };
/// Position of a type in the list
template <typename...> struct position;
template <typename T>
struct position<T> {
static size_t const value = 0;
};
template <typename T, typename Head, typename... Tail>
struct position<T, Head, Tail...> {
static size_t const value =
std::is_same<T, Head>::value ? 0 : 1 + position<T, Tail...>::value;
};
/// Type at a given position
template <size_t, typename...> struct at;
template <size_t N, typename T, typename... Tail>
struct at<N, T, Tail...> { typedef typename at<N-1, Tail..>::type type; };
template <typename T, typename... Tail>
struct at<0, T, Tail...> { typedef T type; };
Now the true fun starts: how to apply a function in a typesafe manner with a type that may change at runtime :x ?
/// Function application
template <typename...> struct Apply;
template <typename H, typename... Tail>
struct Apply<H, Tail...> {
// Mutable
template <typename Func>
static void Do(Func& f, void* storage, size_t tag) {
if (tag == 0) { f(*reinterpret_cast<H*>(storage)); }
else { Apply<Tail...>::Do(f, storage, tag-1); }
}
template <typename Func>
static void Do(Func const& f, void* storage, size_t tag) {
if (tag == 0) { f(*reinterpret_cast<H*>(storage)); }
else { Apply<Tail...>::Do(f, storage, tag-1); }
}
// Const
template <typename Func>
static void Do(Func& f, void const* storage, size_t tag) {
if (tag == 0) { f(*reinterpret_cast<H const*>(storage)); }
else { Apply<Tail...>::Do(f, storage, tag-1); }
}
template <typename Func>
static void Do(Func const& f, void const* storage, size_t tag) {
if (tag == 0) { f(*reinterpret_cast<H const*>(storage)); }
else { Apply<Tail...>::Do(f, storage, tag-1); }
}
}; // struct Apply
/// We need recursion to end quietly even though `tag` is a runtime argument
/// we place the precondition that `tag` should be a valid index in the type
/// list so this should never be reached.
template <>
struct Apply<> {
template <typename... T>
static void Do(T...&&) { abort(); }
}; // struct Apply
Now we can use this to dynamically dispatch in a type safe manner.
/// Variant itself
template <typename... List>
class Variant {
public:
/// Constructor & co
Variant() {
typedef typename at<0, List...>::type First;
new (&_storage) First();
}
Variant(Variant const& other) {
this->initialize(other);
}
Variant& operator=(Variant const& other) {
this->destroy();
this->initialize(other);
return *this;
}
~Variant() { this->destroy(); }
/// Conversions
template <typename T>
explicit Variant(T const& t) {
_tag = position<T, List...>::value;
new (&_storage) T(t);
}
template <typename T>
Variant& operator=(T const& t) {
_tag = position<T, List...>::value;
this->destroy();
new (&_storage) T(t);
return *this;
}
/// Applying a func
template <typename Func>
void apply(Func& f) { Apply<List...>::Do(f, &_storage, _tag); }
template <typename Func>
void apply(Func& f) const { Apply<List...>::Do(f, &_storage, _tag); }
template <typename Func>
void apply(Func const& f) { Apply<List...>::Do(f, &_storage, _tag); }
template <typename Func>
void apply(Func const& f) const { Apply<List...>::Do(f, &_storage, _tag); }
private:
void initialize(Variant const& v) {
struct {
template <typename T>
void operator()(T& t) const { new (_storage) T(t); }
void* _storage;
} copier = { &_storage };
v.apply(copier);
_tag = v._tag;
}
void destroy() {
struct {
template <typename T>
void operator()(T& t) const { t.~T(); }
} eraser;
this->apply(eraser);
}
std::aligned_storage<size<List...>::value,
alignment<List...>::value> _storage;
size_t _tag;
}; // class Variant
Did I say easy ?
Well, there is a subtle issue still: the operator= implementations are not exception safe. In your case it should not be an issue, since you do not have dynamic memory allocation in those types.
References:
std::aligned_storage
Related
I'm making a struct Box<T> that handles some data. The specifics are unimportant.
An important note however is that Box<T> can store a pointer, but it might not. So both Box<int> and Box<int *> are valid. Obviously, if we own Box.data, we're going to need to delete data if it is a pointer type.
Here's a solution I came up with that works in C++11:
template <typename T> struct BoxTraits;
template <typename T> struct Box {
using traits_t = BoxTraits<T>;
T data;
~Box() = default; // not required, I know
T get_data() { return traits_t::get_data(this); }
};
template <typename T> struct Box<T *> {
using traits_t = BoxTraits<T *>;
T *data;
~Box() { delete data; }
T *get_data() { return traits_t::get_data(this); }
};
template <typename T> struct BoxTraits {
static T get_data(Box<T> *const box) { return box->data; }
};
Box::get_data is here to illustrate an issue with this design pattern. For every single method I want to add to Box, I need to add some boiler plate in each specialisation. Note that I would also need a Box<T *const> specialisation.
This seems like quite a rubbish solution. In C++14, I could use if constexpr with a is_ptr<T> trait and only have to write extra code in the methods that need specialising... Is there any way I can do this is in C++11?
This solution is shorter, cleaner and works for Box<U *const>!
template <typename T> struct is_ptr { static const bool value = false; };
template <typename U> struct is_ptr<U *> { static const bool value = true; };
template <typename U> struct is_ptr<U *const> {
static const bool value = true;
};
template <typename T> struct Box {
T data;
~Box() {
if constexpr (is_ptr<T>::value) {
delete data;
}
}
T get_data() { return data; }
};
First off, C++11 already has std::is_pointer, no need to roll your own. You can see that it inherits from std::true_type or std::false_type instead of defining its own value member. The reason for that is tag dispatching, that can effectively replace if constexpr in this situation:
template <typename T> struct Box {
T data;
~Box() {
destroy(std::is_pointer<T>{});
}
private:
void destroy(std::true_type) {
delete data;
}
void destroy(std::false_type) {} // nothing to do
};
Demo
I think this is the most idiomatic way in C++11 for delegating to different implementations based on type traits. In many situations, tag dispatching can replace if constexpr (from C++17, not C++14), and I believe the latter always replaces the former in addition to being clearer. Tag dispatching can also be used before C++11 if you roll your own type traits.
Last note: you don't need to use the standard type traits, you can do something like this:
template <typename T> struct is_ptr { static const bool value = false; };
template <typename T> struct is_ptr<T*> { static const bool value = true; };
template <typename T> struct is_ptr<T* const> { static const bool value = true; };
template <typename T> struct is_ptr<T* volatile> { static const bool value = true; };
template <typename T> struct is_ptr<T* const volatile> { static const bool value = true; };
template<bool b>
struct bool_constant {};
template<typename T>
struct Box {
T data;
~Box() {
destroy(bool_constant<is_ptr<T>::value>{});
}
private:
void destroy(bool_constant<true>) {
delete data;
}
void destroy(bool_constant<false>) {} // nothing to do
};
Demo
However, this pretty much amounts to recreating the standard type traits, but probably worse. Just use the standard library when possible.
I think you had the right idea with the helper type, but I'd do it like the following example illustrates.
template <typename B, typename T>
struct BoxTraits {
static T& get_data(B *const box) { return box->data; }
// ^--- important
static T const& get_data(B const* const box) { return box->data; }
};
template <typename T>
struct BoxTraits<Box<T*>, T> {
static T& get_data(Box<T*>* const box) { return *box->data; }
static T const& get_data(Box<T*> const* const box) { return *box->data; }
};
Both versions always return T, so you can use them the same regardless of your Box's payload. You could add a type alias in Box so you don't have to pass the template arguments:
typedef Traits BoxTraits<Box, T>; // in Box class
The goal
I try to create a set of classes that removes boilerplate code for implementing extensions to a game in C++.
For that, I have a designated value class, that can hold one of the following types:
float, std::string, bool, std::vector<value>, void
For that, I would like to have a host class to which I can add one or more method instances like follows:
using namespace std::string_literals;
host h;
h.add(
method<bool, req<std::string>, req<std::string>, opt<bool>>("compare_strings"s,
[](std::string s_orig, std::string s_comp, std::optional<bool> ingore_case) -> bool {
if (ignore_case.has_value() && ignore_case.value()) {
// ... lowercase both
}
return s_orig.compare(s_comp) == 0;
}));
Note that req<T> should be a meta info that a given value is required, opt<T> a meta info that a given value is not required and may only be provided after all required parameters.
The host class now contains a method execute(std::string function, std::vector<value> values) with function and values originating from a method getting char* for method and ´char** argv+ int argcfor values. Theexecutemethod now is supposed to call the correctmethod` instances function
value host::execute(std::string function, std::vector<value> values) {
// get matching method group
std::vector<method> mthds = m_methods[function];
// get matching parameter list
for (method& mthd : mthds) {
if (mthd.can_call(mthds, values)) {
// call generic method
auto res = mthd.call_generic(values);
// pass result back to callee
// return [...]
}
}
// return error back to callee
// return [...]
}
which means that the actual method class now needs to mangle two methods properly can_call and call_generic.
The value class has corresponding template<typename T> bool is() and template<typename T> T get() methods.
What remains
I did have other attempts at this, but as those failed, I deleted them (not very smart in hindside, but needed to get the whole thing out as another person relied on the results working) and now cannot figure out another attempt then prior ... so this is what I am left with as of now:
class method_base
{
public:
template<typename T> struct in { using type = T; };
template<typename T> struct opt { using type = T; };
public:
virtual bool can_call(std::vector<sqf::value> values) = 0;
virtual sqf::value call_generic(std::vector<sqf::value> values) = 0;
};
template<typename T, typename ... TArgs>
class method : public method_base
{
func m_func;
sqf::value val
public:
using func = T(*)(TArgs...);
method(func f) : m_func(f) {}
virtual retval can_call(std::vector<sqf::value> values) override
{
}
};
Appendix
If something is unclear, confusing or you just have further questions, please do ask them. I will try my best to rephrase whatever is unclear as this will help greatly with developing further extensions in the future, possibly defining a "go to" way for how to create extensions in the community for the game in question (Arma 3 just in case somebody wondered)
I may note that this is pretty much my first deep dive into meta programming so things I present may not be possible at all. If so, I kindly would like to ask you if you may also explain why that is so and the thing I attempt is not possible.
The Solution
I do want to express my thanks to all who answered this question again. I ended up combining pretty much parts of all solutions here and pretty much learned a lot on the way. The final implementation I ended up with looks like the following:
namespace meta
{
template <typename ArgType>
struct is_optional : std::false_type {};
template <typename T>
struct is_optional<std::optional<T>> : std::true_type {};
template <typename ArgType>
inline constexpr bool is_optional_v = is_optional<ArgType>::value;
template <typename ArgType>
struct def_value { static ArgType value() { return {}; } };
template <typename ArgType>
struct get_type { using type = ArgType; };
template <typename ArgType>
struct get_type<std::optional<ArgType>> { using type = ArgType; };
}
struct method {
std::function<bool(const std::vector<value>&)> m_can_call;
std::function<value(const std::vector<value>&)> m_call;
template <typename ... Args, std::size_t... IndexSequence>
static bool can_call_impl(const std::vector<value>& values, std::index_sequence<IndexSequence...> s) {
// values max args
return values.size() <= sizeof...(Args) &&
// for every Arg, either...
(... && (
// the value provides that argument and its the correct type, or...
(IndexSequence < values.size() && sqf::is<sqf::meta::get_type<Args>::type>(values[IndexSequence])) ||
// the value does not provide that argument and the arg is an optional
(IndexSequence >= values.size() && sqf::meta::is_optional_v<Args>)
));
}
template <typename Ret, typename ... Args, std::size_t... IndexSequence>
static value call_impl(std::function<Ret(Args...)> f, const std::vector<value>& values, std::index_sequence<IndexSequence...>) {
return {
// call the function with every type in the value set,
// padding with empty std::optionals otherwise
std::invoke(f,
(IndexSequence < values.size() ? sqf::get<sqf::meta::get_type<Args>::type>(values[IndexSequence])
: sqf::meta::def_value<Args>::value())...)
};
}
public:
template <typename Ret, typename ... Args>
method(std::function<Ret(Args...)> f) :
m_can_call([](const std::vector<value>& values) -> bool
{
return can_call_impl<Args...>(values, std::index_sequence_for<Args...>{});
}),
m_call([f](const std::vector<value>& values) -> value
{
return call_impl<Ret, Args...>(f, values, std::index_sequence_for<Args...>{});
})
{
}
bool can_call(const std::vector<value>& values) const { return m_can_call(values); }
value call_generic(const std::vector<value>& values) const { return m_call(values); }
// to handle lambda
template <typename F>
method static create(F f) { return method{ std::function{f} }; }
};
Assumming a way to check current type of value (template <typename T> bool value::isA<T>()) and a way to retrieve the value (template <typename T> /*const*/T& get(/*const*/ value&))
It seems you might do:
struct method
{
template <typename Ret, typename ... Ts>
method(std::function<Ret(Ts...)> f) : method(std::index_sequence<sizeof...(Ts)>(), f)
{}
template <typename Ret, typename ... Ts, std::size_t ... Is>
method(std::index_sequence<Is...>, std::function<Ret(Ts...)> f) :
isOk([](const std::vector<value>& values) {
return ((values.size() == sizeof...(Is)) && ... && values[Is].isA<Ts>());
}),
call([f](const std::vector<value>& values){
return f(get<Ts>(values[Is])...);
})
{}
// to handle lambda
template <typename F>
static fromCallable(F f) { return method{std::function{f}}; }
std::function<bool(const std::vector<value>&)> isOk;
std::function<value(const std::vector<value>&)> call;
};
Here's a quick example including the machinery for ret<T> and opt<T>. You haven't given any information on what value is, so I'm going to assume something like:
struct value {
// using `std::monostate` instead of `void`
std::variant<float, std::string, bool, std::vector<value>, std::monostate> data;
};
(I'm assuming c++17 for this answer.)
From there, we need our metatypes and a few traits to branch off them. I implement them using partial specialisations, but there are other ways too.
// types to determine optional vs. required
template <typename T>
struct req { using type = T; };
template <typename T>
struct opt { using type = T; };
// trait to determine if it's an optional type
template <typename ArgType>
struct is_optional : std::false_type {};
template <typename T>
struct is_optional<opt<T>> : std::true_type {};
template <typename ArgType>
inline constexpr bool is_optional_v = is_optional<ArgType>::value;
// get the "real" function parameter type
template <typename ArgType>
struct real_type;
template <typename ArgType>
using real_type_t = typename real_type<ArgType>::type;
template <typename T>
struct real_type<req<T>> { using type = T; };
template <typename T>
struct real_type<opt<T>> { using type = std::optional<T>; };
Now we implement method. I'll use a similar polymorphic relationship with method_base as you do in your partial demo; I also template on the function type passed in, to allow e.g. the functions to use const references to the type instead of the type itself.
The implementation itself uses the common trick of delegating to helper functions with std::index_sequence and fold expressions to "iterate" through the variadic template args.
// base class for polymorphism
struct method_base {
virtual ~method_base() = default;
virtual bool can_call(const std::vector<value>& values) const = 0;
virtual value call_generic(const std::vector<value>& values) const = 0;
};
// provide a different method implementation for each set of args
// I also overload on
template<typename RetType, typename Fn, typename... Args>
struct method : method_base {
private:
Fn func;
static_assert(std::is_invocable_r_v<RetType, Fn, real_type_t<Args>...>,
"function must be callable with given args");
public:
// accept any function that looks sort of like what we expect;
// static assert above makes sure it's sensible
template <typename G>
method(G&& func) : func(std::forward<G>(func)) {}
template <std::size_t... Is>
bool can_call_impl(const std::vector<value>& values, std::index_sequence<Is...>) const {
// for every Arg, either...
return (... and (
// the value provides that argument and its the correct type, or...
(Is < values.size() and std::holds_alternative<typename Args::type>(values[Is].data))
// the value does not provide that argument and the arg is an optional
or (Is >= values.size() and is_optional_v<Args>)
));
}
bool can_call(const std::vector<value>& values) const override {
return can_call_impl(values, std::index_sequence_for<Args...>{});
}
template <std::size_t... Is>
value call_generic_impl(const std::vector<value>& values, std::index_sequence<Is...>) const {
return {
// call the function with every type in the value set,
// padding with empty std::optionals otherwise
std::invoke(func,
(Is < values.size() ? std::get<typename Args::type>(values[Is].data)
: real_type_t<Args>{})...)
};
}
value call_generic(const std::vector<value>& values) const override {
return call_generic_impl(values, std::index_sequence_for<Args...>{});
}
};
I'll also create a helper function to make methods:
template <typename RetType, typename... Args, typename Fn>
std::unique_ptr<method_base> make_method(Fn&& func) {
return std::make_unique<method<RetType, std::decay_t<Fn>, Args...>>(std::forward<Fn>(func));
}
Live example.
It's not perfect, but this should give you a general idea of how to do it.
Change your method to:
method< R(Args...) >
your tags seem useless. Detect optional with ... std::optional.
For storage, use std variant. Use some non-void type for void (I don't care what).
As a first pass we aim for perfect compatibility.
template<class...Args>
struct check_signature {
bool operator()( std::span<value const> values ) const {
if (sizeof...(Args) != values.size()) return false;
std::size_t i=0;
return (std::holds_alternative<Args>(values[i++])&&...);
}
};
this can be stored in a std::function<bool(std::span<value const>)> or just called in your class impementation.
Similar code can store the callable.
template<class F, class R, class...Args>
struct execute {
F f;
template<std::size_t...Is>
R operator()( std::index_sequence<Is...>, std::span<value const> values ) const {
if (sizeof...(Args) != values.size()) return false;
return f( std::get<Args>(values[Is])... );
}
R operator()( std::span<value const> values ) const {
return (*this)( std::make_index_sequence<sizeof...(Args)>{}, values );
}
};
some work may have to be done for the fake void.
Your method is now a aggregate.
struct method {
std::function<bool(std::span<value const>)> can_call;
std::function<value(std::span<value const>)> execute;
};
if you want it to be. The two template objects above can be stored in these two std functions.
There are probably tpyos, I just wrote this on my phone and have not tested it.
Extending this to cover optional args is a little bit of work. But nothing hard.
In both cases, you'll write a helper function that says if the argument is compatible or generates the value based on if you are past the end of the incoming vector.
Ie, std::get<Args>(values[Is])... becomes getArgFrom<Is, Args>{}(values)..., and we specialize for std optional producing nullopt if Is>values.size().
Consider the following simplified piece of code for a variant class. Most of it is for informational purposes, the question is about the conditional_invoke method.
// Possible types in variant.
enum class variant_type { empty, int32, string };
// Actual data store.
union variant_data {
std::int32_t val_int32;
std::string val_string;
inline variant_data(void) { /* Leave uninitialised */ }
inline ~variant_data(void) { /* Let variant do clean up. */ }
};
// Type traits which allow inferring which type to use (these are actually generated by a macro).
template<variant_type T> struct variant_type_traits { };
template<class T> struct variant_reverse_traits { };
template<> struct variant_type_traits<variant_type::int32> {
typedef std::int32_t type;
inline static type *get(variant_data& d) { return &d.val_int32; }
};
template<> struct variant_reverse_traits<std::int32_t> {
static const variant_type type = variant_type::int32;
inline static std::int32_t *get(variant_data& d) { return &d.val_int32; }
};
template<> struct variant_type_traits<variant_type::string> {
typedef std::string type;
inline static type *get(variant_data& d) { return &d.val_string; }
};
template<> struct variant_reverse_traits<std::string> {
static const variant_type type = variant_type::string;
inline static std::string *get(variant_data& d) { return &d.val_string; }
};
// The actual variant class.
class variant {
public:
inline variant(void) : type(variant_type::empty) { }
inline ~variant(void) {
this->conditional_invoke<destruct>();
}
template<class T> inline variant(const T value) : type(variant_type::empty) {
this->set<T>(value);
}
template<class T> void set(const T& value) {
this->conditional_invoke<destruct>();
std::cout << "Calling data constructor ..." << std::endl;
::new (variant_reverse_traits<T>::get(this->data)) T(value);
this->type = variant_reverse_traits<T>::type;
}
variant_data data;
variant_type type;
private:
template<variant_type T> struct destruct {
typedef typename variant_type_traits<T>::type type;
static void invoke(type& v) {
std::cout << "Calling data destructor ..." << std::endl;
v.~type();
}
};
template<template<variant_type> class F, class... P>
inline void conditional_invoke(P&&... params) {
this->conditional_invoke0<F, variant_type::int32, variant_type::string, P...>(std::forward<P>(params)...);
}
template<template<variant_type> class F, variant_type T, variant_type... U, class... P>
void conditional_invoke0(P&&... params) {
if (this->type == T) {
F<T>::invoke(*variant_type_traits<T>::get(this->data), std::forward<P>(params)...);
}
this->conditional_invoke0<F, U..., P...>(std::forward<P>(params)...);
}
template<template<variant_type> class F, class... P>
inline void conditional_invoke0(P&&... params) { }
};
The code works this way, i.e. it works as long as the parameter list P... for the functor is empty. If I add another functor like
template<variant_type T> struct print {
typedef typename variant_type_traits<T>::type type;
static void invoke(type& v, std::ostream& stream) {
stream << v;
}
};
and try to invoke it
friend inline std::ostream& operator <<(std::ostream& lhs, variant& rhs) {
rhs.conditional_invoke<print>(lhs);
return lhs;
}
the compiler VS 20115 complains
error C2672: 'variant::conditional_invoke0': no matching overloaded function found
or gcc respectively
error: no matching function for call to 'variant::conditional_invoke0 >&>(std::basic_ostream&)'
I guess the compiler cannot decide when U... ends and when P... starts. Is there any way to work around the issue?
You'll have to make both parameter packs deducible. That is, let the type and non-type template parameters be part of a function parameter list. For that, introduce a dummy structure:
template <variant_type...>
struct variant_type_list {};
and let the compiler deduce the variant_type... pack from a function call:
template <template <variant_type> class F
, variant_type T
, variant_type... U
, typename... P>
void conditional_invoke0(variant_type_list<T, U...> t
, P&&... params)
{
if (this->type == T)
{
F<T>::invoke(*variant_type_traits<T>::get(this->data)
, std::forward<P>(params)...);
}
this->conditional_invoke0<F>(variant_type_list<U...>{}
, std::forward<P>(params)...);
}
To break recursive calls, introduce an overload with an empty variant_type_list:
template <template <variant_type> class F, typename... P>
void conditional_invoke0(variant_type_list<>, P&&... params) {}
When calling the invoker for the first time, provide variant_types as an argument:
this->conditional_invoke0<F>(variant_type_list<variant_type::int32, variant_type::string>{}
, std::forward<P>(params)...);
DEMO
I have a component I use when implementing low-level generic types that store an object of arbitrary type (may or may not be a class type) which may be empty to take advantage of the empty base optimization:
template <typename T, unsigned Tag = 0, typename = void>
class ebo_storage {
T item;
public:
constexpr ebo_storage() = default;
template <
typename U,
typename = std::enable_if_t<
!std::is_same<ebo_storage, std::decay_t<U>>::value
>
> constexpr ebo_storage(U&& u)
noexcept(std::is_nothrow_constructible<T,U>::value) :
item(std::forward<U>(u)) {}
T& get() & noexcept { return item; }
constexpr const T& get() const& noexcept { return item; }
T&& get() && noexcept { return std::move(item); }
};
template <typename T, unsigned Tag>
class ebo_storage<
T, Tag, std::enable_if_t<std::is_class<T>::value>
> : private T {
public:
using T::T;
constexpr ebo_storage() = default;
constexpr ebo_storage(const T& t) : T(t) {}
constexpr ebo_storage(T&& t) : T(std::move(t)) {}
T& get() & noexcept { return *this; }
constexpr const T& get() const& noexcept { return *this; }
T&& get() && noexcept { return std::move(*this); }
};
template <typename T, typename U>
class compressed_pair : ebo_storage<T, 0>,
ebo_storage<U, 1> {
using first_t = ebo_storage<T, 0>;
using second_t = ebo_storage<U, 1>;
public:
T& first() { return first_t::get(); }
U& second() { return second_t::get(); }
// ...
};
template <typename, typename...> class tuple_;
template <std::size_t...Is, typename...Ts>
class tuple_<std::index_sequence<Is...>, Ts...> :
ebo_storage<Ts, Is>... {
// ...
};
template <typename...Ts>
using tuple = tuple_<std::index_sequence_for<Ts...>, Ts...>;
Lately I've been messing about with lock-free data structures and I need nodes that optionally contain a live datum. Once allocated, nodes live for the lifetime of the data structure but the contained datum is only alive while the node is active and not while the node sits in a free list. I implemented the nodes using raw storage and placement new:
template <typename T>
class raw_container {
alignas(T) unsigned char space_[sizeof(T)];
public:
T& data() noexcept {
return reinterpret_cast<T&>(space_);
}
template <typename...Args>
void construct(Args&&...args) {
::new(space_) T(std::forward<Args>(args)...);
}
void destruct() {
data().~T();
}
};
template <typename T>
struct list_node : public raw_container<T> {
std::atomic<list_node*> next_;
};
which is all fine and dandy, but wastes a pointer-sized chunk of memory per node when T is empty: one byte for raw_storage<T>::space_, and sizeof(std::atomic<list_node*>) - 1 bytes of padding for alignment. It would be nice to take advantage of EBO and allocate the unused single-byte representation of raw_container<T> atop list_node::next_.
My best attempt at creating a raw_ebo_storage performs "manual" EBO:
template <typename T, typename = void>
struct alignas(T) raw_ebo_storage_base {
unsigned char space_[sizeof(T)];
};
template <typename T>
struct alignas(T) raw_ebo_storage_base<
T, std::enable_if_t<std::is_empty<T>::value>
> {};
template <typename T>
class raw_ebo_storage : private raw_ebo_storage_base<T> {
public:
static_assert(std::is_standard_layout<raw_ebo_storage_base<T>>::value, "");
static_assert(alignof(raw_ebo_storage_base<T>) % alignof(T) == 0, "");
T& data() noexcept {
return *static_cast<T*>(static_cast<void*>(
static_cast<raw_ebo_storage_base<T>*>(this)
));
}
};
which has the desired effects:
template <typename T>
struct alignas(T) empty {};
static_assert(std::is_empty<raw_ebo_storage<empty<char>>>::value, "Good!");
static_assert(std::is_empty<raw_ebo_storage<empty<double>>>::value, "Good!");
template <typename T>
struct foo : raw_ebo_storage<empty<T>> { T c; };
static_assert(sizeof(foo<char>) == 1, "Good!");
static_assert(sizeof(foo<double>) == sizeof(double), "Good!");
but also some undesirable effects, I assume due to violation of strict aliasing (3.10/10) although the meaning of "access the stored value of an object" is debatable for an empty type:
struct bar : raw_ebo_storage<empty<char>> { empty<char> e; };
static_assert(sizeof(bar) == 2, "NOT good: bar::e and bar::raw_ebo_storage::data() "
"are distinct objects of the same type with the "
"same address.");
This solution also potential for undefined behavior upon construction. At some point the program must construct the containee object within the raw storage with placement new:
struct A : raw_ebo_storage<empty<char>> { int i; };
static_assert(sizeof(A) == sizeof(int), "");
A a;
a.value = 42;
::new(&a.get()) empty<char>{};
static_assert(sizeof(empty<char>) > 0, "");
Recall that despite being empty, a complete object necessarily has non-zero size. In other words, an empty complete object has a value representation that consists of one or more padding bytes. new constructs complete objects, so a conforming implementation could set those padding bytes to arbitrary values at construction instead of leaving memory untouched as would be the case for constructing an empty base subobject. This would of course be catastrophic if those padding bytes overlay other live objects.
So the question is, is it possible to create a standard-compliant container class that uses raw storage/delayed initialization for the contained object and takes advantage of EBO to avoid wasting memory space for the representation of the contained object?
I think you gave the answer yourself in your various observations:
You want raw memory and placement new. This requires to have at least one byte available, even if you want to construct an empty object via placement new.
You want zero bytes overhead for storing any empty objects.
These requirements are self-contradicting. The answer therefore is No, that is not possible.
You could change your requirements a bit more, though, by requiring the zero byte overhead only for empty, trivial types.
You could define a new class trait, e.g.
template <typename T>
struct constructor_and_destructor_are_empty : std::false_type
{
};
Then you specialize
template <typename T, typename = void>
class raw_container;
template <typename T>
class raw_container<
T,
std::enable_if_t<
std::is_empty<T>::value and
std::is_trivial<T>::value>>
{
public:
T& data() noexcept
{
return reinterpret_cast<T&>(*this);
}
void construct()
{
// do nothing
}
void destruct()
{
// do nothing
}
};
template <typename T>
struct list_node : public raw_container<T>
{
std::atomic<list_node*> next_;
};
Then use it like this:
using node = list_node<empty<char>>;
static_assert(sizeof(node) == sizeof(std::atomic<node*>), "Good");
Of course, you still have
struct bar : raw_container<empty<char>> { empty<char> e; };
static_assert(sizeof(bar) == 1, "Yes, two objects sharing an address");
But that is normal for EBO:
struct ebo1 : empty<char>, empty<usigned char> {};
static_assert(sizeof(ebo1) == 1, "Two object in one place");
struct ebo2 : empty<char> { char c; };
static_assert(sizeof(ebo2) == 1, "Two object in one place");
But as long as you always use construct and destruct and no placement new on &data(), you're golden.
I want to automatically choose the right pointer-to-member among overloaded ones based on the "type" of the member, by removing specializations that accept unconcerned members (via enable_if).
I have the following code:
class test;
enum Type
{
INT_1,
FLOAT_1,
UINT_1,
CHAR_1,
BOOL_1,
INT_2,
FLOAT_2,
UINT_2,
CHAR_2,
BOOL_2
};
template<typename T, Type Et, typename func> struct SetterOk { static const bool value = false; };
template<typename T> struct SetterOk<T,INT_1,void (T::*)(int)> { static const bool value = true; };
template<typename T> struct SetterOk<T,FLOAT_1,void (T::*)(float)> { static const bool value = true; };
template<typename T> struct SetterOk<T,UINT_1,void (T::*)(unsigned int)> { static const bool value = true; };
template<typename T> struct SetterOk<T,CHAR_1,void (T::*)(char)> { static const bool value = true; };
template<typename T> struct SetterOk<T,BOOL_1,void (T::*)(bool)> { static const bool value = true; };
template<typename T> struct SetterOk<T,INT_2,void (T::*)(int,int)> { static const bool value = true; };
template<typename T> struct SetterOk<T,FLOAT_2,void (T::*)(float,float)> { static const bool value = true; };
template<typename T> struct SetterOk<T,UINT_2,void (T::*)(unsigned int, unsigned int)> { static const bool value = true; };
template<typename T> struct SetterOk<T,CHAR_2,void (T::*)(char,char)> { static const bool value = true; };
template<typename T> struct SetterOk<T,BOOL_2,void (T::*)(bool,bool)> { static const bool value = true; };
template <bool, class T = void> struct enable_if {};
template <class T> struct enable_if<true, T> { typedef T type; };
template<typename T, Type Et>
struct Helper
{
template<typename U>
static void func(U method, typename enable_if<SetterOk<T,Et,U>::value>::type* dummy = 0)
{
}
};
class test
{
public:
void init()
{
Helper<test,INT_2>::func(&test::set);
}
void set2(int);
void set(int);
void set(int,int);
void set(float,float);
};
int main()
{
test t;
t.init();
return 0;
}
I'm expecting it to choose the right function between all possible. The problem is that the compiler says "cannot deduce template argument as function argument is ambiguous".
It seems I don't know how to use enable_if, because if so the compiler would only allow the specialization if the specified function has the right type...
Note that I want to have C++03 solutions (if possible) - my code must compile on some old compilers.
Thanks in advance
You can never refer to an overloaded function without disambiguating it (means: static_casting it to the correct type). When you instantiate Helper::func the type of the function argument cannot be known without ever disambiguating it.
The reason it doesn't compile is quite simply that there are several different overloaded functions and it doesn't know which one you mean. Granted, only one of these (void set(int,int)) would actually compile, given the specialization Helper<test,INT_2>. However, this is not enough for the compiler to go on.
One way of getting this to compile would be to explicitly cast &test::set to the appropriate type:
Helper<test,INT_2>::func(static_cast<void (test::*)(int,int)>(&test::set));
Another way would be to use explicit template specialization:
Helper<test,INT_2>::func<void (test::*)(int,int)>((&test::set));
Either way, you need to let the compiler know which of the set functions you are trying to refer to.
EDIT:
As I understand it, you want to be able to deduce, from the use of a Type, which function type should be used. The following alternative achieves this:
template<typename T, Type Et> struct SetterOK{};
template<typename T> struct SetterOK<T,INT_1> {typedef void (T::*setter_type)(int);};
template<typename T> struct SetterOK<T,FLOAT_1> {typedef void (T::*setter_type) (float);};
// ...
template<typename T> struct SetterOK<T,INT_2> {typedef void (T::*setter_type)(int,int);};
// ....
template<typename T, Type Et>
struct Helper
{
template<typename U>
static void func(U method)
{
}
};
class test
{
public:
void init()
{
Helper<test,INT_2>::func<SetterOK<test,INT_2>::setter_type >(&test::set);
}
void set2(int);
void set(int);
void set(int,int);
void set(float,float);
};
int main()
{
test t;
t.init();
return 0;
}
ADDITIONAL EDIT:
A thought just occurred to me. In this special case which you've done, where U is SetterOK::setter_type, things can be simplified further by completely removing the template arguments for func:
static void func(typename SetterOK<T,Et>::setter_type method)
{
}
This would make the init method a simpler:
void init()
{
Helper<test,INT_2>::func(&test::set);
}