When I call a function declared in a namespace from outside this namespace, I usually need to explicitly prefix with the namespace:
namespace ns1 {
void myfunc();
}
myfunc(); // compiler complains
ns1::myfunc(); // good
However I have this situation where it seems the compiler can automatically devise what function I want to use:
namespace mymath {
struct Vector3 {
float x, y, z;
};
Vector3 vector_cross(const Vector3& lhs, const Vector3 &rhs)
{
return lhs; // Pretend this is a cross product
}
} // ns
int main(int argc, char** argv)
{
mymath::Vector3 v, w, x;
x = vector_cross(v, w); // <---- Here, I do not need to
// prefix with the namespace
return 0;
}
Q1: Is it because the compiler can automatically select the appropriate function based on the arguments type? Or something else?
After more testing, I found out that my compiler does not complain if I add another function with the same declaration in another namespace:
namespace mymath {
// same as above
} // ns
namespace math2 {
mymath::Vector3 vector_cross(const mymath::Vector3& lhs, const mymath::Vector3 &rhs)
{
return rhs; // Return rhs this time
}
} // ns
int main(int argc, char** argv)
{
mymath::Vector3 v, w, x;
x = vector_cross(v, w); // <---- Here, which one did the compiler chose?
return 0;
}
Q2: How can I de-activate this behavior?
Edit:
My compiler: gcc version 4.7.2 (Debian 4.7.2-5)
Q1: 'appropriate' is debated, but essentially yes, it selects function combining all kinds of "associated" namespaces, that works fine for many samples.
Q2: You make a few dozen people join WG21 and finally vote one of the 'fix ADL' proposals to life, this far the attempts fell flat.
Q2, pragmatic answer: ADL only applies to unqualified lookup, so you always have the option to use qualified names. (prefix :: std:: and so on...). Or stop using namespaces completely. Other than that you are subject to "Koenig uncertainty principle".