Question

In a GoogleTechTalks video on Youtube, Bjarne Stroustrup talks about the upcoming C++0x standard. In the video he mentions the following example:

#include <iostream>

struct Sick
{
    Sick(double d)       { std::cout << d << "\n"; }
    explicit Sick(int i) { std::cout << i << "\n"; }
};


int main()
{
    Sick s1 = 2.1;
    Sick s2(2.1);
}

Did he mean to place the explicit keyword before Sick(double) rather than Sick(int), in order to highlight problems associated with implicit conversions in certain contexts?

Answer 1

In his discussion, Stroustrup mentions that a direct initialization, such as

Sick s2(2.1);

will consider only constructors marked explicit if there are any explicit constructors. That’s not my experience with several compilers (including GCC 4.6.1 and MSVC 16/VS 2010), and I can’t find such a requirement in the standard (though I’d be interested if someone can point it out).

However, if ints are used in the initializers, I think the example would show what Stroustrup meant to demonstrate:

#include <iostream>

struct Sick
{
    Sick(double d)       { std::cout << "double " << d << "\n"; }
    explicit Sick(int i) { std::cout << "int " << i << "\n"; }
};


int main()
{
    Sick s1 = 2;
    Sick s2(2);
}

Running the above will display:

double 2
int 2

Shows that the two apparently equivalent initializations actually select different constructors.

(Or as Truncheon mentioned in the question – and I missed – that the explicit keyword should be on the Sick(double d) constructor).