C++ template specialization for pointer? - c++

I read book < C++ Templates - the Complete Guide > and learned template specialization for pointer. (maybe I misunderstand this part of the book)
(1) Here's my simple template:
#include <iostream>
template<typename T>
void Function(const T& a)
{
std::cout << "Function<T>: " << a << std::endl;
}
template<typename T>
void Function<T*>(const T* a)
{
std::cout << "Function<T*>: " << a << std::endl;
}
int main(void)
{
Function(1);
Function(1.2);
Function("hello");
Function((void*)0x25);
return 0;
}
I use ubuntu16.04 x64, g++ 5.3, the compiler report:
$ g++ main.cpp -o main.exe
main.cpp:10:29: error: non-type partial specialization ‘Function<T*>’ is not allowed
void Function<T*>(const T* a)
(2) but this code is correct:
#include <iostream>
template<typename T>
void Function(const T& a)
{
std::cout << "Function<T>: " << a << std::endl;
}
int main(void)
{
Function(1);
Function(1.2);
Function("hello");
Function((void*)0x25);
return 0;
}
result shows:
$ g++ main.cpp -o main.exe
$ ./main.exe
Function<T>: 1
Function<T>: 1.2
Function<T>: hello
Function<T>: 0x25
My question is: Is the book about pointer specialization is wrong ? Or I mis understand the meaning of this part in the book ? Or something else ?
Update about pointer specialization in class.
(3) template class with pointer specialization:
#include <iostream>
template<typename T>
struct Base {
T member;
Base(const T& a)
: member(a)
{
}
void hello()
{
std::cout << member << std::endl;
}
};
template<typename T>
struct Base<T*> {
T* member;
Base(T* a)
: member(a)
{
}
void hello()
{
std::cout << member << std::endl;
}
};
int main(void)
{
Base<int> b1(12);
Base<double> b2(2.4);
Base<char*> b3("hello");
Base<void*> b4((void*)0x25);
b1.hello();
b2.hello();
b3.hello();
b4.hello();
return 0;
}
this code is correct with one warning:
$ g++ main.cpp -o main.exe
main.cpp: In function ‘int main()’:
main.cpp:37:27: warning: deprecated conversion from string constant to ‘char*’ [-Wwrite-strings]
Base<char*> b3("hello");
^
$ ./main.exe
12
2.4
hello
0x25
(4) template class without pointer specialization:
#include <iostream>
template<typename T>
struct Base {
T member;
Base(const T& a)
: member(a)
{
}
void hello()
{
std::cout << member << std::endl;
}
};
int main(void)
{
Base<int> b1(12);
Base<double> b2(2.4);
Base<char*> b3("hello");
Base<void*> b4((void*)0x25);
b1.hello();
b2.hello();
b3.hello();
b4.hello();
return 0;
}
result is the same:
$ g++ main.cpp -o main.exe
main.cpp: In function ‘int main()’:
main.cpp:39:27: warning: deprecated conversion from string constant to ‘char*’ [-Wwrite-strings]
Base<char*> b3("hello");
^
$ ./main.exe
12
2.4
hello
0x25
Does this means pointer specialization is needless ?
Or maybe this feature behave differently on different compiler ?

as you've been already told, partial specialization of function templates are not allowed. You can use std::enable_if for this:
template <typename T, typename std::enable_if_t<!std::is_pointer<T>::value>* = 0>
void func(T val) { std::cout << val << std::endl; }
template <typename T, typename std::enable_if_t<std::is_pointer<T>::value>* = 0>
void func(T val) { func(*val); }
If you are looking for simpler syntax, wait for concepts

The error message told you what is wrong:
non-type partial specialization ‘Function<T*>’ is not allowed
You can only partially specialize types (classes). You've tried to partially specialize a function. Functions are not types; you can only fully specialize them.

Two problems:
You are not allowed to partially specialise a function.
The behaviour of (void*)0x25 is undefined. With the exception of nullptr, you are not allowed to set a pointer to memory you don't own, with the exception of one past the final element of an array and one past the address of a scalar.

Related

what is the way to remove the first element from a std::span<T>?

when reading the document of std::span, I see there is no method to remove the first element from the std::span<T>.
Can you suggest a way to solve my issue?
The large picture of my problem(I asked in another question: How to instantiatiate a std::basic_string_view with custom class T, I got is_trivial_v<_CharT> assert error) is that I would like to have a std::basic_string_view<Token>, while the Token is not a trivial class, so I can't use std::basic_string_view, and someone suggested me to use std::span<Token> instead.
Since the basic_string_view has a method named remove_prefix which remove the first element, while I also need such kinds of function because I would like to use std::span<Token> as a parser input, so the Tokens will be matched, and consumed one by one.
Thanks.
EDIT 2023-02-04
I try to derive a class named Span from std::span, and add the remove_prefix member function, but it looks like I still have build issues:
#include <string_view>
#include <vector>
#include <span>
// derived class, add remove_prefix function to std::span
template<typename T>
class Span : public std::span<T>
{
public:
// Inheriting constructors
using std::span<T>::span;
// add a public function which is similar to std::string_view::remove_prefix
constexpr void remove_prefix(std::size_t n) {
*this = subspan(n);
}
};
struct Token
{
Token(){};
Token(const Token& other)
{
lexeme = other.lexeme;
type = other.type;
}
std::string_view lexeme;
int type;
// equal operator
bool operator==(const Token& other)const {
return (this->lexeme == other.lexeme) ;
}
};
template <typename T>
struct Viewer;
template <>
struct Viewer<Token>
{
using type = Span<Token>; // std::span or derived class
};
template <>
struct Viewer<char>
{
using type = std::string_view;
};
template <typename T> using ViewerT = typename Viewer<T>::type;
template <typename T>
class Parser
{
using v = ViewerT<T>;
};
// a simple parser demo
template <typename Base, typename T>
struct parser_base {
using v = ViewerT<T>;
constexpr auto operator[](v& output) const noexcept;
};
template<typename T>
struct char_ final : public parser_base<char_<T>, T> {
using v = ViewerT<T>;
constexpr explicit char_(const T ch) noexcept
: ch(ch)
{}
constexpr inline bool visit(v& sv) const& noexcept {
if (!sv.empty() && sv.front() == ch) {
sv.remove_prefix(1);
return true;
}
return false;
}
private:
T ch;
};
template <typename Parser, typename T>
constexpr bool parse(Span<T> &input, Parser const& parser) noexcept {
return parser.visit(input);
}
int main()
{
Token kw_class;
kw_class.lexeme = "a";
std::vector<Token> token_stream;
token_stream.push_back(kw_class);
token_stream.push_back(kw_class);
token_stream.push_back(kw_class);
Span<Token> token_stream_view{&token_stream[0], 3};
auto p = char_(kw_class);
parse(token_stream_view, p);
return 0;
}
The build error looks like below:
[ 50.0%] g++.exe -Wall -std=c++20 -fexceptions -g -c F:\code\test_crtp_twoargs\main.cpp -o obj\Debug\main.o
F:\code\test_crtp_twoargs\main.cpp: In member function 'constexpr void Span<T>::remove_prefix(std::size_t)':
F:\code\test_crtp_twoargs\main.cpp:52:17: error: there are no arguments to 'subspan' that depend on a template parameter, so a declaration of 'subspan' must be available [-fpermissive]
52 | *this = subspan(n);
| ^~~~~~~
F:\code\test_crtp_twoargs\main.cpp:52:17: note: (if you use '-fpermissive', G++ will accept your code, but allowing the use of an undeclared name is deprecated)
F:\code\test_crtp_twoargs\main.cpp: In instantiation of 'constexpr void Span<T>::remove_prefix(std::size_t) [with T = Token; std::size_t = long long unsigned int]':
F:\code\test_crtp_twoargs\main.cpp:113:29: required from 'constexpr bool char_<T>::visit(v&) const & [with T = Token; v = Span<Token>]'
F:\code\test_crtp_twoargs\main.cpp:125:24: required from 'constexpr bool parse(Span<T>&, const Parser&) [with Parser = char_<Token>; T = Token]'
F:\code\test_crtp_twoargs\main.cpp:141:10: required from here
F:\code\test_crtp_twoargs\main.cpp:52:24: error: 'subspan' was not declared in this scope, and no declarations were found by argument-dependent lookup at the point of instantiation [-fpermissive]
52 | *this = subspan(n);
| ~~~~~~~^~~
F:\code\test_crtp_twoargs\main.cpp:52:24: note: declarations in dependent base 'std::span<Token, 18446744073709551615>' are not found by unqualified lookup
F:\code\test_crtp_twoargs\main.cpp:52:24: note: use 'this->subspan' instead
F:\code\test_crtp_twoargs\main.cpp:52:15: error: no match for 'operator=' (operand types are 'Span<Token>' and 'std::span<Token, 18446744073709551615>')
52 | *this = subspan(n);
| ~~~~~~^~~~~~~~~~~~
F:\code\test_crtp_twoargs\main.cpp:44:7: note: candidate: 'constexpr Span<Token>& Span<Token>::operator=(const Span<Token>&)'
44 | class Span : public std::span<T>
| ^~~~
F:\code\test_crtp_twoargs\main.cpp:44:7: note: no known conversion for argument 1 from 'std::span<Token, 18446744073709551615>' to 'const Span<Token>&'
F:\code\test_crtp_twoargs\main.cpp:44:7: note: candidate: 'constexpr Span<Token>& Span<Token>::operator=(Span<Token>&&)'
F:\code\test_crtp_twoargs\main.cpp:44:7: note: no known conversion for argument 1 from 'std::span<Token, 18446744073709551615>' to 'Span<Token>&&'
Any idea on how to fix this issue?
Also, I don't know how to make a general parse function:
template <typename Parser, typename T>
constexpr bool parse(Span<T> &input, Parser const& parser) noexcept {
return parser.visit(input);
}
Currently, the first argument of the parse should be a Viewer like type?
EDIT2023-02-05
Change the function as below, the above code can build correctly. This is from Benjamin Buch's answer.
constexpr void remove_prefix(std::size_t n) {
auto& self = static_cast<std::span<T>&>(*this);
self = self.subspan(n);
}
There is still one thing remains: How to generalize the parse function to accept both input types of std::string_view and Span<Token>?
If I change the parse function to this:
template <typename Parser, typename T>
constexpr bool parse(ViewerT<T> &input, Parser const& parser) noexcept {
return parser.visit(input);
}
I got such compile error:
[ 50.0%] g++.exe -Wall -std=c++20 -fexceptions -g -c F:\code\test_crtp_twoargs\main.cpp -o obj\Debug\main.o
F:\code\test_crtp_twoargs\main.cpp: In function 'int main()':
F:\code\test_crtp_twoargs\main.cpp:143:24: error: no matching function for call to 'parse(Span<Token>&, char_<Token>&)'
143 | bool result = parse(token_stream_view, p);
| ~~~~~^~~~~~~~~~~~~~~~~~~~~~
F:\code\test_crtp_twoargs\main.cpp:125:16: note: candidate: 'template<class Parser, class T> constexpr bool parse(ViewerT<T>&, const Parser&)'
125 | constexpr bool parse(ViewerT<T> &input, Parser const& parser) noexcept {
| ^~~~~
F:\code\test_crtp_twoargs\main.cpp:125:16: note: template argument deduction/substitution failed:
F:\code\test_crtp_twoargs\main.cpp:143:24: note: couldn't deduce template parameter 'T'
143 | bool result = parse(token_stream_view, p);
| ~~~~~^~~~~~~~~~~~~~~~~~~~~~
Any ideas?
Thanks.
BTW: I have to explicitly instantiation of the parse function call like:
bool result = parse<decltype(p), Token>(token_stream_view, p);
to workaround this issue.
Call subspan with 1 as only (template) argument to get a new span, which doesn't contain the first element.
If you use a span with a static extend, you need a new variable because the data type changes by subspan.
#include <string_view>
#include <iostream>
#include <span>
int main() {
std::span<char const, 12> text_a("a test-span");
std::cout << std::string_view(text_a) << '\n';
std::span<char const, 10> text_b = text_a.subspan<2>();
std::cout << std::string_view(text_b) << '\n';
}
If you have a dynamic extend, you can assign the result to the original variable.
#include <string_view>
#include <iostream>
#include <span>
int main() {
std::span<char const> text("a test-span");
std::cout << std::string_view(text) << '\n';
text = text.subspan(2);
std::cout << std::string_view(text) << '\n';
}
The implementation of a modifying inplace subspan version is only possible for spans with a dynamic extend. It can be implemented as a free function.
#include <string_view>
#include <iostream>
#include <span>
template <typename T>
constexpr void remove_front(std::span<T>& self, std::size_t const n) noexcept {
self = self.subspan(n);
}
int main() {
std::span<char const> text("a test-span");
std::cout << std::string_view(text) << '\n';
remove_front(text, 2);
std::cout << std::string_view(text) << '\n';
}
You can use your own spans derived from std::span if you prefer the dot-call.
#include <string_view>
#include <iostream>
#include <span>
template <typename T>
struct my_span: std::span<T> {
using std::span<T>::span;
constexpr void remove_front(std::size_t const n) noexcept {
auto& self = static_cast<std::span<T>&>(*this);
self = self.subspan(n);
}
};
int main() {
my_span<char const> my_text("a test-span");
std::cout << std::string_view(my_text) << '\n';
my_text.remove_front(2);
std::cout << std::string_view(my_text) << '\n';
}
You can also write a wrapper class to call via dot syntax. This way you can additionally implement cascadable modification calls by always returning the a reference modifier class.
#include <string_view>
#include <iostream>
#include <span>
template <typename T>
class span_modifier {
public:
constexpr span_modifier(std::span<T>& span) noexcept: span_(span) {}
constexpr span_modifier& remove_front(std::size_t const n) noexcept {
span_ = span_.subspan(n);
return *this;
}
private:
std::span<T>& span_;
};
template <typename T>
constexpr span_modifier<T> modify(std::span<T>& span) noexcept {
return span;
}
int main() {
std::span<char const> text("a test-span");
std::cout << std::string_view(text) << '\n';
modify(text).remove_front(2).remove_front(5);
std::cout << std::string_view(text) << '\n';
}
Note I use the template function modify to create an object of the wrapper class, because the names of classes cannot be overloaded. Therefore class names should always be a bit more specific. The function modify can also be overloaded for other data types, which then return a different wrapper class. This results in a simple intuitive and consistent interface for modification wrappers.
You can write remove_prefix of your version,
template <typename T>
constexpr void remove_prefix(std::span<T>& sp, std::size_t n) {
sp = sp.subspan(n);
}
Demo

C++ function header matching: how does matching work when const and templates are both involved?

I had a templated function that I wished to call. This is (a trimmed-down version of) the header:
template <typename Item>
void print (shared_ptr<const MyContainer<Item>> stuff, ostream& out)
which I tried to call with a line like this:
print (make_shared<MyContainer<int>>(42), cerr);
But the compiler complained that there was no match. What confuses me is that the const mismatch is not a problem, because if I redeclare my function to omit the template it works:
void print (shared_ptr<const MyContainer<int>> stuff, ostream& out) //matches!
On the other hand, if I omit constness, the templated version does work:
template <typename Item>
void print (shared_ptr<MyContainer<Item>> stuff, ostream& out) //also matches!
But I should be able to write a function over const things and pass it a non-const value (which the function will then just not modify), right? Indeed, if I go back to non-managed pointers, the corresponding old way to write the header would have been
template <typename Item>
void print (const MyContainer<Item>* stuff, ostream& out)
and indeed then a call to
print (new MyContainer<int>(42), cerr); //yet another match!
once again just fine.
So, what is it about this particular cocktail of shared_ptr, templates, and const that causes the compiler to be unable to find the matching function? (Running g++ 8.2.1, and clang++ 7.0.1 seems to produce the same result.)
Concerning const-ness of pointee, std::shared_ptr behaves a bit different than raw-pointers.
A std::shared_ptr<T> is not the same as a std::shared_ptr<const T>. It's even not that compatible to allow an implicit conversion. (The error message in Daniels answer says this quite literally.)
It doesn't work for the same reason like in the following (counter) example:
template <typename T>
struct ContainerT {
T a;
ContainerT(T a): a(a) { }
ContainerT(const ContainerT&) = default;
ContainerT& operator=(const ContainerT&) = default;
};
int main()
{
ContainerT<int> a(42);
ContainerT<const int> b(a);
return 0;
}
Output:
g++ (GCC) 8.2.0
Copyright (C) 2018 Free Software Foundation, Inc.
This is free software; see the source for copying conditions. There is NO
warranty; not even for MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.
main.cpp: In function 'int main()':
main.cpp:15:28: error: no matching function for call to 'ContainerT<const int>::ContainerT(ContainerT<int>&)'
ContainerT<const int> b(a);
^
main.cpp:8:3: note: candidate: 'constexpr ContainerT<T>::ContainerT(const ContainerT<T>&) [with T = const int]'
ContainerT(const ContainerT&) = default;
^~~~~~~~~~
main.cpp:8:3: note: no known conversion for argument 1 from 'ContainerT<int>' to 'const ContainerT<const int>&'
main.cpp:7:3: note: candidate: 'ContainerT<T>::ContainerT(T) [with T = const int]'
ContainerT(T a): a(a) { }
^~~~~~~~~~
main.cpp:7:3: note: no known conversion for argument 1 from 'ContainerT<int>' to 'int'
Live Demo on coliru
In the case of std::shared_ptr, there is a way to circumvent this issue
→ a std::const_pointer_cast can be used:
#include <iostream>
#include <memory>
template <typename T>
struct ContainerT {
T a;
ContainerT(T a): a(a) { }
};
template <typename T>
void print(std::shared_ptr<const ContainerT<T>> ref, std::ostream &out)
{
out << "print: '" << ref->a << "'\n";
}
int main()
{
print(std::make_shared<const ContainerT<int>>(42), std::cout);
print(std::const_pointer_cast<const ContainerT<int>>(std::make_shared<ContainerT<int>>(42)), std::cout);
return 0;
}
Output:
print: '42'
print: '42'
Live Demo on coliru
For convenience, the const-cast might be done in another function template:
#include <iostream>
#include <memory>
template <typename T>
struct ContainerT {
T a;
ContainerT(T a): a(a) { }
};
template <typename T>
void print(std::shared_ptr<const ContainerT<T>> ref, std::ostream &out)
{
out << "print const: '" << ref->a << "'\n";
}
template <typename T>
void print(std::shared_ptr<ContainerT<T>> ref, std::ostream &out)
{
out << "print non-const: ";
print(std::const_pointer_cast<const ContainerT<T>>(ref), out);
}
int main()
{
print(std::make_shared<const ContainerT<int>>(42), std::cout);
print(std::make_shared<ContainerT<int>>(42), std::cout);
return 0;
}
Output:
print const: '42'
print non-const: print const: '42'
Live Demo on coliru
Here is a simplified code:
template <typename T>
void f(std::shared_ptr<const std::vector<T>>) { }
void g(std::shared_ptr<const std::vector<int>>) { }
int main() {
f(std::make_shared<std::vector<int>>()); // ERROR
g(std::make_shared<std::vector<int>>()); // OK
}
To understand what happens, read the error message, e.g., the one printed by g++:
...
note: template argument deduction/substitution failed:
note: types 'const std::vector<T>' and 'std::vector<int>' have incompatible cv-qualifiers
It tells you that the problem is with template argument deduction/substitution. The C++ rules seemingly do not allow this type of deduction. (If I have some time, I will try to find a relevant part of the Standard).
However, you can skip the template argument deduction by providing an explicit template argument:
f<int>(std::make_shared<std::vector<int>>()); // OK

c++ template recursion double linked list error gcc (but clang++ accepts)

This example code:
#include <string>
#include <iostream>
template <int i>
struct Wrapper
{
static const std::string _str;
typedef const Wrapper<i+1> NextType_t; // template recursion
static NextType_t _nxt;
typedef const Wrapper<i-1> PrevType_t; // template recursion
static PrevType_t _prev;
};
template<int i>
const std::string Wrapper<i>::_str = std::to_string(i);
template<int i>
typename Wrapper<i>::NextType_t Wrapper<i>::_nxt;
template<int i>
typename Wrapper<i>::PrevType_t Wrapper<i>::_prev;
// recursion termination - lower bound
template <>
struct Wrapper<-1>
{
static const std::string _str;
typedef const Wrapper<0> NextType_t;
static NextType_t _nxt;
typedef const Wrapper<-1> PrevType_t;
static PrevType_t _prev;
};
const std::string Wrapper<-1>::_str = std::to_string(-1);
typename Wrapper<-1>::NextType_t Wrapper<-1>::_nxt;
typename Wrapper<-1>::PrevType_t Wrapper<-1>::_prev;
// recursion termination - upper bound
template <>
struct Wrapper<UPPER_LIMIT>
{
static const std::string _str;
typedef const Wrapper<-1> NextType_t;
static NextType_t _nxt;
typedef const Wrapper<UPPER_LIMIT-1> PrevType_t;
static PrevType_t _prev;
};
const std::string Wrapper<UPPER_LIMIT>::_str = std::to_string(UPPER_LIMIT);
typename Wrapper<UPPER_LIMIT>::NextType_t Wrapper<UPPER_LIMIT>::_nxt;
typename Wrapper<UPPER_LIMIT>::PrevType_t Wrapper<UPPER_LIMIT>::_prev;
int
main(
int argc,
char **)
{
Wrapper<0> wrapperFirst;
Wrapper<UPPER_LIMIT> wrapperLast;
// here's the list
std::cout << wrapperFirst._str << std::endl;
std::cout << wrapperFirst._nxt._str << std::endl;
std::cout << wrapperFirst._nxt._nxt._str << std::endl;
// [...]
// and the final element
std::cout << wrapperLast._str << std::endl;
std::cout << wrapperLast._prev._str << std::endl;
std::cout << wrapperLast._prev._prev._str << std::endl;
// [...]
// and the tailing NIL
std::cout << Wrapper<UPPER_LIMIT>::NextType_t::_str << std::endl;
return 0;
}
fails for gcc:
> g++ -std=c++11 -DUPPER_LIMIT=100 -ftemplate-depth=500 -o test main.cpp
main.cpp: In instantiation of ‘struct Wrapper<499>’:
main.cpp:24:33: recursively required from ‘struct Wrapper<1>’
main.cpp:24:33: required from ‘struct Wrapper<0>’
main.cpp:43:47: required from here
main.cpp:24:33: fatal error: template instantiation depth exceeds maximum of 500 (use -ftemplate-depth= to increase the maximum)
typename Wrapper<i>::NextType_t Wrapper<i>::_nxt;
^~~~~~~~~~
compilation terminated.
but succeeds for clang:
> clang++ -std=c++11 -DUPPER_LIMIT=100 -ftemplate-depth=500 -o test main.cpp
> ./test
0
1
2
100
99
98
-1
Is there something wrong in the code? I wonder that gcc wants to go beyond the UPPER_LIMIT, because there is a terminating specialization for that.
Wrapper<-1> is instantiated on this line
typename Wrapper<-1>::NextType_t Wrapper<-1>::_nxt;
This causes Wrapper<0> to be instantiated, which causes Wrapper<1> to be instantiated, etc. At that point in the code, the specialization for Wrapper<UPPER_LIMIT> has not been defined, so this causes an infinite recursion.
Moving the definition of the Wrapper<UPPER_LIMIT> specialization above the Wrapper<-1>::_nxt definition solves the problem.
Apparently Clang defers instantiation, so that this problem doesn't arise.

friend method from a template specializated struct: g++ and clang++ different behaviour

Trying to propose a solution to another question, I bumped (again) my nose against code that compile and work with clang++ (3.5) but give compilation error with g++ (4.9.2)
The following is a simplified minimal (I hope) example
#include <iostream>
class foo;
template <std::size_t>
struct bar;
template <>
struct bar<0U>
{ static void baz (foo const & f); };
class foo
{
int i = 42;
template <std::size_t I>
friend void bar<I>::baz (foo const &);
};
void bar<0U>::baz (foo const & f)
{ std::cout << f.i << std::endl; }
int main()
{
foo f;
bar<0U>::baz(f);
}
The error from my g++ is
test_114-98,11,14,clang.cpp:18:41: error: member ‘void bar<<anonymous> >::baz(const foo&)’ declared as friend before type ‘bar<<anonymous> >’ defined
friend void bar<I>::baz (foo const &);
^
test_114-98,11,14,clang.cpp: In static member function ‘static void bar<0ul>::baz(const foo&)’:
test_114-98,11,14,clang.cpp:15:13: error: ‘int foo::i’ is private
int i = 42;
^
test_114-98,11,14,clang.cpp:22:19: error: within this context
{ std::cout << f.i << std::endl; }
^
My question, as usual, is: who's right and who's wrong.

g++ and clang++ different behaviour with SFINAE and SFINAE failure

A couple of questions for C++11 experts.
I'm fighting with SFINAE and I came across a strange case in which g++ (4.9.2), and clang++ (3.5.0) behave differently.
I have prepared the following sample code. I'm sorry but I'm unable to do it significantly more concise.
#include <string>
#include <iostream>
#include <typeinfo>
#include <type_traits>
template <typename X>
class foo
{
private:
template <typename R>
using enableIfIsInt
= typename std::enable_if<std::is_same<X, int>::value, R>::type;
public:
foo ()
{ }
template <typename R = void>
enableIfIsInt<R> bar ()
{ std::cout << "bar: is int\n"; }
void bar ()
{
std::cout << "bar: isn't int; is [" << typeid(X).name() << "]{"
<< typeid(enableIfIsInt<void>).name() << "}\n";
}
};
int main ()
{
foo<long> fl;
foo<int> fi;
fl.bar();
fi.bar();
return 0;
}
My idea was to create a template foo<X> class that (via SFINAE) can define a method in one or in another way depending on the X template argument.
The program compile well with g++ 4.9.2 but clang++ 3.5.0 give the following error
test.cpp:13:36: error: no type named 'type' in
'std::__1::enable_if<false, void>'; 'enable_if' cannot be used to disable
this declaration
= typename std::enable_if<std::is_same<X, int>::value, R>::type;
^~~~~~~~~~~~~~~~~~~~~~~~~~~
test.cpp:26:23: note: in instantiation of template type
alias 'enableIfIsInt' requested here
<< typeid(enableIfIsInt<void>).name() << "}\n";
^
test.cpp:36:7: note: in instantiation of member function
'foo<long>::bar' requested here
fl.bar();
^
1 error generated.
I suppose that is right clang++ but my first question to C++11 experts is: who right? g++ or clang++?
About the g++ produced program output, it's the following
bar: isn't int; is [i]{v}
so g++ seems to ignore the fl.bar(); instruction.
Now a little change: i modify the second version of foo<X>::bar() in this way
void bar ()
{ std::cout << "bar: isn't int; is [" << typeid(X).name() << "]\n"; }
deleting the std::enable_if inside the function abomination. Now both g++ and clang++ are compiling without problems and the output, for both compiled versions of the program, is
bar: isn't int; is [l]
bar: isn't int; is [i]
So, my second question is: what I'm doing wrong? Why, in the int case, I don't obtain the "is int" version of foo<X>::bar()?
Be patient with me if I'm doing some foolish: I'm trying to learn C++11.
And sorry for my bad English.
clang's error isn't coming from the substitution failure. It's coming from here:
void bar ()
{
std::cout << "bar: isn't int; is [" << typeid(X).name() << "]{"
<< typeid(enableIfIsInt<void>).name() << "}\n"; // <==
}
enableIfIsInt<void> isn't in the immediate context, that's a hard failure for X is not int. You simply can't use that expression in that context.
Once you remove that - the non-template bar() is always called. That's because both functions are equivalent matches and non-templates are preferred to templates in overload resolution.
So the real solution is to use tag-dispatching:
void bar() { bar(std::is_same<X, int>{}); }
void bar(std::true_type ) {
std::cout << "bar: is int\n";
}
void bar(std::false_type ) {
std::cout << "bar: isn't int; is [" << typeid(X).name() << "]\n";
}
with which both compilers happily yield:
bar: isn't int; is [l]
bar: is int