I have this example:
#include <iostream>
#include <tuple>
#include <string>
template <typename T>
class A {
public:
A(const T &t) : m_t(t) {}
void foo() {
std::cout << m_t << std::endl;
}
private:
T m_t;
};
typedef std::tuple<std::string, std::string> Type;
std::ostream &operator<<(std::ostream &os, const Type &t) {
os << std::get<0>(t) << " " << std::get<1>(t);
return os;
}
int main() {
A<Type> a(Type{"ala", " ma kota"});
a.foo();
return 0;
}
which with clang++ (3.6) produces:
test_clang.cpp:10:19: error: call to function 'operator<<' that is neither visible in the template definition nor found by argument-dependent lookup
std::cout << m_t << std::endl;
^
test_clang.cpp:26:7: note: in instantiation of member function 'A<std::tuple<std::basic_string<char>, std::basic_string<char> > >::foo' requested here
a.foo();
^
test_clang.cpp:19:15: note: 'operator<<' should be declared prior to the call site
std::ostream &operator<<(std::ostream &os, const Type &t) {
No errors occurred during g++-4.8 with C++11 and g++-5.2.1 with C++17 builds. clang++-3.6 needs std::ostream &operator<<(std::ostream &os, const Type &t)
to be defined before A::foo<T>
.
From my perspective member m_t
depends on template argument type and used operator<<
for this type shouldn't be needed during template definition. Why clang has compilation error and g++ doesn't?
std::tuple<std::string, std::string>
Let's look at the associated namespaces of this type. [basic.lookup.argdep]/(2.2):
Its associated namespaces are the innermost enclosing namespaces of its associated classes.
That would be namespace std
or auxiliary ones, but certainly not the global namespace.
Furthermore, if
T
is a class template specialization, its associated namespaces and classes also include: the namespaces and classes associated with the types of the template arguments provided for template type parameters (excluding template template parameters); [… inapplicable rules…]
Recursively applying the above to std::string
gives namespace std
(and, again, auxiliary ones) for the associated namespaces. Certainly not the global namespace. Clearly, the same argumentation can be repeated for std::cout
, giving the same conclusion.
Thus ADL won't look in the global namespace, which is precisely where your overload is declared in.
Finally, as per [temp.dep.candidate]/1, name resolution is unsuccessful:
GCC behaves non-conforming here; see #51577.
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With