Why is this template class specialization being instantiated? - c++

Inside a specialization for a templated class I call a function which should not compile, but in my mind it shouldn't matter as I pass false as a template argument and the compiler shouldn't instantiate that class in the first place:
#include <iostream>
template <typename T>
struct outerstruct
{
template <bool b>
struct innerStruct
{
};
template<>
struct innerStruct<true>
{
using type = std::underlying_type_t<T>;
};
template<>
struct innerStruct<false>
{
};
innerStruct<false> member; // std::underlying_type_t cannot be called on a T,
// even though that class shouldn't be instantiated
// because I'm passing false (or I think)
};
int main()
{
outerstruct<int> a;
}
Why is innerStruct<true> being instantiated and the compiler trying to call underlying_type if I pass the false argument?
Very confusingly for me, I can make it behave the way I want (ie., have the compiler ignore the call to std::underlying_type_t) by just adding another template argument to innerClass, here I add a U:
#include <iostream>
template <typename T>
struct outerstruct
{
template <typename ExtraArg, bool b>
struct innerStruct
{
};
template<typename ExtraArg>
struct innerStruct<ExtraArg, true>
{
using type = std::underlying_type_t<T>;
};
template<typename ExtraArg>
struct innerStruct<ExtraArg, false>
{
};
//innerStruct<bValue> member;
innerStruct<int, true> member; // Now it's working as I expected,
// I can pass int or whatever argument to ExtraArg
// and it compiles as long as I have 'false'
};
int main()
{
outerstruct<int> a;
}
What is the difference between the first and the second examples? What business does template specialization innerStruct<true> have being instantiated if in the first example I only called innerStruct<false>

Related

C++ partial specialization with multiple optional parameters [duplicate]

Is it possible to specialize particular members of a template class? Something like:
template <typename T,bool B>
struct X
{
void Specialized();
};
template <typename T>
void X<T,true>::Specialized()
{
...
}
template <typename T>
void X<T,false>::Specialized()
{
...
}
Ofcourse, this code isn't valid.
You can only specialize it explicitly by providing all template arguments. No partial specialization for member functions of class templates is allowed.
template <typename T,bool B>
struct X
{
void Specialized();
};
// works
template <>
void X<int,true>::Specialized()
{
...
}
A work around is to introduce overloaded functions, which have the benefit of still being in the same class, and so they have the same access to member variables, functions and stuffs
// "maps" a bool value to a struct type
template<bool B> struct i2t { };
template <typename T,bool B>
struct X
{
void Specialized() { SpecializedImpl(i2t<B>()); }
private:
void SpecializedImpl(i2t<true>) {
// ...
}
void SpecializedImpl(i2t<false>) {
// ...
}
};
Note that by passing along to the overloaded functions and pushing the template parameters into a function parameter, you may arbitrary "specialize" your functions, and may also templatize them as needed. Another common technique is to defer to a class template defined separately
template<typename T, bool B>
struct SpecializedImpl;
template<typename T>
struct SpecializedImpl<T, true> {
static void call() {
// ...
}
};
template<typename T>
struct SpecializedImpl<T, false> {
static void call() {
// ...
}
};
template <typename T,bool B>
struct X
{
void Specialized() { SpecializedImpl<T, B>::call(); }
};
I find that usually requires more code and i find the function overload easier to handle, while others prefer the defer to class template way. In the end it's a matter of taste. In this case, you could have put that other template inside X too as a nested template - in other cases where you explicitly specialize instead of only partially, then you can't do that, because you can place explicit specializations only at namespace scope, not into class scope.
You could also create such a SpecializedImpl template just for purpose of function overloading (it then works similar to our i2t of before), as the following variant demonstrates which leaves the first parameter variable too (so you may call it with other types - not just with the current instantiation's template parameters)
template <typename T,bool B>
struct X
{
private:
// maps a type and non-type parameter to a struct type
template<typename T, bool B>
struct SpecializedImpl { };
public:
void Specialized() { Specialized(SpecializedImpl<T, B>()); }
private:
template<typename U>
void Specialized(SpecializedImpl<U, true>) {
// ...
}
template<typename U>
void Specialized(SpecializedImpl<U, false>) {
// ...
}
};
I think sometimes, deferring to another template is better (when it comes to such cases as arrays and pointers, overloading can tricky and just forwarding to a class template has been easier for me then), and sometimes just overloading within the template is better - especially if you really forward function arguments and if you touch the classes' member variables.
This is what I came up with, not so bad :)
//The generic template is by default 'flag == false'
template <class Type, bool flag>
struct something
{
void doSomething()
{
std::cout << "something. flag == false";
}
};
template <class Type>
struct something<Type, true> : public something<Type, false>
{
void doSomething() // override original dosomething!
{
std::cout << "something. flag == true";
}
};
int main()
{
something<int, false> falseSomething;
something<int, true> trueSomething;
falseSomething.doSomething();
trueSomething.doSomething();
}

C++: Partially specializing template's type parameter as another template class's member-type

I have a template struct SFoo that contains a member struct SZug:
template <typename tTYPE>
struct SFoo
{
struct SZug {};
};
I have another struct SBar that takes a type parameter:
template <typename tTYPE>
struct SBar
{ /* stuff */ };
I would like to specialize SBar using SZug for the type parameter, like so:
template <typename tTYPE>
struct SBar<typename SFoo<tTYPE>::SZug>
{ /* different stuff */ };
This doesn't compile - LLVM outputs:
non-deducible template parameter 'tTYPE'
While a compiler could easily deduce this if it wished, I'm guessing it's just that the C++ spec would need to specifically cover this case.
Is there any way to achieve this?
(note: I'm currently working around it by moving SZug outside of SFoo and using a using declaration, but it's ugly.)
I am not sure I fully understood what you want to do, but you could try the following (it only requires adding a specific attributes to SZug:
template <typename tTYPE>
struct SFoo {
struct SZug {
// Add this to be able to obtain SFoo<T> from SFoo<T>::SZug
using type = tTYPE;
};
};
Then a small template to check if a type is a SFoo<T>::SZug:
template <typename tTYPE, typename Enabler = void>
struct is_SZug: public std::false_type { };
template <typename tTYPE>
struct is_SZug<tTYPE, typename std::enable_if<
std::is_same<tTYPE, typename SFoo<typename tTYPE::type>::SZug>{}
>::type>: public std::true_type { };
And a slight modification to the SBar template to enable the "specialization" if the type is a SZug:
template <typename tTYPE, typename Enabler = void>
struct SBar
{ static void g(); };
template <typename tTYPE>
struct SBar<tTYPE, typename std::enable_if<is_SZug<tTYPE>{}>::type>
{ static void f(); };
A little check:
void f () {
SBar<int>::g();
SBar<SFoo<int>::SZug>::f();
}
Note: You could also directly set SFoo<T> as the type attribute in SFoo<T>::SZug, you would simply need to change the second argument of std::is_same a little.
You can get the effect for which you're looking through the following (which prints out 0 1, BTW):
#include <type_traits>
#include <iostream>
namespace detail
{
struct SZugBase{};
}
template <typename tTYPE>
struct SFoo
{
struct SZug : public detail::SZugBase {};
};
template<typename tType, bool IsFoo>
struct SBarBase
{
int value = 0;
};
template<typename tType>
struct SBarBase<tType, true>
{
int value = 1;
};
template <typename tTYPE>
struct SBar : public SBarBase<tTYPE, std::is_convertible<tTYPE, detail::SZugBase>::value>
{ /* stuff */ };
int main()
{
SBar<int> b0;
SBar<SFoo<int>::SZug> b1;
std::cout << b0.value << " " << b1.value << std::endl;
}
Explanation
First, we give SZug a regular-class base:
namespace detail
{
struct SZugBase{};
}
template <typename tTYPE>
struct SFoo
{
struct SZug : public detail::SZugBase {};
};
Note the following:
SZugBase is not parameterized by anything, so it is easy to refer to it independently of the parameter of SFoo
SZugBase is in a detail namespace, so, by common C++ conventions, you're telling clients of your code to ignore it.
Now we give SBar two base classes, specialized on whether something is convertible to the non-template base of SZug:
template<typename tType, bool IsFoo>
struct SBarBase
{
int value = 0;
};
template<typename tType>
struct SBarBase<tType, true>
{
int value = 1;
};
Finally, we just need to make SBar a subclass of these bases (depending on the specialization):
template <typename tTYPE>
struct SBar : public SBarBase<tTYPE, std::is_convertible<tTYPE, detail::SZugBase>::value>
{ /* stuff */ };
Note that you don't specialize SBar here, you rather specialize the base classes. This effectively gives the same effect, though.

Are members of a class template instantiated when the class is instantiated?

Supposedly members of a template class shouldn't be instantiated unless they are used.
However this sample seems to instantiate the do_something member and the enable_if fails (which would be expected if we'd instantiated it - but AFAIK we did not).
Am I missing something really basic here?
#include <string>
#include <boost/utility.hpp>
struct some_policy {
typedef boost::integral_constant<bool, false> condition;
};
struct other_policy {
typedef boost::integral_constant<bool, true> condition;
};
template <typename policy>
class test {
void do_something(typename boost::enable_if<typename policy::condition>::type* = 0) {}
};
int main() {
test<other_policy> p1;
test<some_policy> p2;
}
coliru
From C++11 14.7.1/1:
The implicit instantiation of a class template specialization causes the implicit
instantiation of the declarations, but not of the definitions or default arguments, of the class member functions
So the function declaration is instantiated; that fails since it depends on an invalid type.
(Unfortunately, I don't have any historic versions of the standard to hand, but I imagine this rule was similar in C++98)
Mike Seymour already answered why it doesn't work, here's how to workaround it:
#include <string>
#include <boost/utility.hpp>
struct some_policy {
typedef boost::integral_constant<bool, false> condition;
};
struct other_policy {
typedef boost::integral_constant<bool, true> condition;
};
template <typename policy>
class test {
private:
template<typename P>
void do_something_impl(typename boost::enable_if<typename P::condition>::type* = 0) {}
public:
void do_something()
{
do_something_impl<policy>();
}
};
int main() {
test<other_policy> p1;
test<some_policy> p2;
}
Quick rule of thumb: If you want to do SFINAE, you need a member function template.
SFINAE happens only on template function/method (here, it is your class which is template),
You may do in C++11 (default template parameter for function/method):
template <typename policy>
class test {
template <typename T = policy>
void do_something(typename boost::enable_if<typename T::condition>::type* = 0) {}
};
You may alternatively use specialization, something like
template <bool> struct helper_do_something {};
template <> struct helper_do_something<true>
{
void do_something() { /* Your implementation */ }
};
template <typename policy>
class test : helper_do_something<T::condition::value>
{
// do_something is inherited (and it is present only when T::condition::value is true)
};

Disable functions inside templated class

I'm trying to disable some functions inside a simple template class. The functions that should be removed depend on whether the template argument has certain typedefs.
The example boils down to this:
template<typename T>
struct Foo
{
typename T::Nested foo() { return typename T::Nested(); }
int bar() { return 1; }
};
struct NoNested
{
};
struct WithNested
{
typedef int Nested;
};
int main()
{
Foo<WithNested> fwn;
fwn.foo();
fwn.bar();
Foo<NoNested> fnn;
//fnn.foo();
fnn.bar();
}
However this gives me a error: no type named ‘Nested’ in ‘struct NoNested’ style error on gcc and clang++ (mind you old versions of both).
Is there an easy way to remove foo when the typedef T::Nested does not exit? (Other than template specialization of the Foo<T> class, as in the real code I have this for about 5 functions with different typedefs.. which would result in 2^5 different specialization )
EDIT:
Since there has been some asking for the motivation for wanting to do this:
I'd like to create something like acompile time FSM for use in a DSL.
I'd like to be able to do this
struct StateA;
struct StateB;
struct StateC;
struct StateA
{
typedef StateB AfterNext;
};
struct StateB
{
typedef StateA AfterPrev;
typedef StateC AfterNext;
};
struct StateC
{
typedef StateB AfterPrev;
};
template<typename T>
struct FSM
{
FSM<typename T::AfterNext> next() { return FSM<T::AfterNext>(); };
FSM<typename T::AfterPrev> prev() { return FSM<T::AfterPrev>(); };
};
So that
FSM<StateA>().next().prev().next().next();
compiles, but
FSM<StateA>().next().prev().prev();
fails.
Note that in reality there would be more transition functions than this, the transition functions would actually do something, and the FSM would store some state.
UPDATE:
I've created proper examples using the methods that have been given so far.
The answers vary in complexity, and while visitors method is the one I'd probably end up using (as it is simplest), my solution (the most complicated) is the only one that actually removes the function.
You can use class template specialization. If you have several functions, then you can move each function to a base class, and specialize each base class.
Try making function foo template itself. It will compile only when called, so you will get the error only when you will try calling it with NoNested class.
You could add a nested typedef to every class, such that compilation only fails when the function is instantiated.
struct null_type; //an incomplete type, you could use a more descriptive name for your particular problem
template<typename T>
struct Foo
{
typename T::Nested foo() { return typename T::Nested(); }
int bar() { return 1; }
};
struct NoNested
{
typedef null_type Nested;
};
struct WithNested
{
typedef int Nested;
};
int main()
{
Foo<WithNested> fwn;
fwn.foo();
fwn.bar();
Foo<NoNested> fnn;
//fnn.foo(); //attempt to use incomplete type when used
fnn.bar();
}
It is possible to choose the type T::Nested, if it exists, otherwise void, as follows.
The default choice is void:
template<class T, class = void>
struct NestedReturn
{
typedef void type;
};
A template which always returns void, whatever type you give it:
template<class T>
struct Void
{
typedef void type;
};
A specialisation for types with a Nested nested class by SFINAE. Note that typename Void<typename T::Nested>::type is always void, to match the default second parameter of void in the base template:
template<class T>
struct NestedReturn<T, typename Void<typename T::Nested>::type>
{
typedef typename T::Nested type;
};
And now we use it. Note that foo() is not actually removed when there is no T::Nested, but instantiating it causes an error.
template<typename T>
struct Foo
{
typename NestedReturn<T>::type foo() { return typename T::Nested(); }
int bar() { return 1; }
};
struct NoNested
{
};
struct WithNested
{
typedef int Nested;
};
int main()
{
Foo<WithNested> fwn;
fwn.foo();
fwn.bar();
Foo<NoNested> fnn;
//fnn.foo();
fnn.bar();
}
I suspect that using default function template parameters it would be possible to remove foo() properly using SFINAE, but that's only possible in C++11 (untested guesswork):
template<typename T>
struct Foo
{
template<class N = T::Nested>
N foo() { return N(); }
int bar() { return 1; }
};
Here's how I think I can solve it. It's inspired by user763305's comments.
It requires 2*N specialisations rather than 2^N.
template <typename T>
struct has_nested {
// Variables "yes" and "no" are guaranteed to have different sizes,
// specifically sizeof(yes) == 1 and sizeof(no) == 2.
typedef char yes[1];
typedef char no[2];
template <typename C>
static yes& test(typename C::Nested*);
template <typename>
static no& test(...);
// If the "sizeof" the result of calling test<T>(0) would be equal to the sizeof(yes),
// the first overload worked and T has a nested type named type.
static const bool value = sizeof(test<T>(0)) == sizeof(yes);
};
template<typename T>
struct FooBase
{
int bar() { return 1; }
};
template<typename T, bool>
struct FooImpl : public FooBase<T>
{
};
template<typename T>
struct FooImpl<T,true> : public FooBase<T>
{
typename T::Nested foo() { return typename T::Nested(); }
};
template<typename T>
struct Foo : public FooImpl<T, has_nested<T>::value >
{
};
struct NoNested
{
};
struct WithNested
{
typedef int Nested;
};
int main()
{
Foo<WithNested> fwn;
fwn.foo();
fwn.bar();
Foo<NoNested> fnn;
//fnn.foo();
fnn.bar();
}

Template specialization of particular members?

Is it possible to specialize particular members of a template class? Something like:
template <typename T,bool B>
struct X
{
void Specialized();
};
template <typename T>
void X<T,true>::Specialized()
{
...
}
template <typename T>
void X<T,false>::Specialized()
{
...
}
Ofcourse, this code isn't valid.
You can only specialize it explicitly by providing all template arguments. No partial specialization for member functions of class templates is allowed.
template <typename T,bool B>
struct X
{
void Specialized();
};
// works
template <>
void X<int,true>::Specialized()
{
...
}
A work around is to introduce overloaded functions, which have the benefit of still being in the same class, and so they have the same access to member variables, functions and stuffs
// "maps" a bool value to a struct type
template<bool B> struct i2t { };
template <typename T,bool B>
struct X
{
void Specialized() { SpecializedImpl(i2t<B>()); }
private:
void SpecializedImpl(i2t<true>) {
// ...
}
void SpecializedImpl(i2t<false>) {
// ...
}
};
Note that by passing along to the overloaded functions and pushing the template parameters into a function parameter, you may arbitrary "specialize" your functions, and may also templatize them as needed. Another common technique is to defer to a class template defined separately
template<typename T, bool B>
struct SpecializedImpl;
template<typename T>
struct SpecializedImpl<T, true> {
static void call() {
// ...
}
};
template<typename T>
struct SpecializedImpl<T, false> {
static void call() {
// ...
}
};
template <typename T,bool B>
struct X
{
void Specialized() { SpecializedImpl<T, B>::call(); }
};
I find that usually requires more code and i find the function overload easier to handle, while others prefer the defer to class template way. In the end it's a matter of taste. In this case, you could have put that other template inside X too as a nested template - in other cases where you explicitly specialize instead of only partially, then you can't do that, because you can place explicit specializations only at namespace scope, not into class scope.
You could also create such a SpecializedImpl template just for purpose of function overloading (it then works similar to our i2t of before), as the following variant demonstrates which leaves the first parameter variable too (so you may call it with other types - not just with the current instantiation's template parameters)
template <typename T,bool B>
struct X
{
private:
// maps a type and non-type parameter to a struct type
template<typename T, bool B>
struct SpecializedImpl { };
public:
void Specialized() { Specialized(SpecializedImpl<T, B>()); }
private:
template<typename U>
void Specialized(SpecializedImpl<U, true>) {
// ...
}
template<typename U>
void Specialized(SpecializedImpl<U, false>) {
// ...
}
};
I think sometimes, deferring to another template is better (when it comes to such cases as arrays and pointers, overloading can tricky and just forwarding to a class template has been easier for me then), and sometimes just overloading within the template is better - especially if you really forward function arguments and if you touch the classes' member variables.
This is what I came up with, not so bad :)
//The generic template is by default 'flag == false'
template <class Type, bool flag>
struct something
{
void doSomething()
{
std::cout << "something. flag == false";
}
};
template <class Type>
struct something<Type, true> : public something<Type, false>
{
void doSomething() // override original dosomething!
{
std::cout << "something. flag == true";
}
};
int main()
{
something<int, false> falseSomething;
something<int, true> trueSomething;
falseSomething.doSomething();
trueSomething.doSomething();
}