Conversion operator + conversion constructor = unintuitive behavior?

Question

I don't understand why the code below prints struct Value instead of int (which implies the conversion constructor is converting to Value instead of int). (Visual C++ 2012)

Why is this happening? Why does the compiler completely ignore the Value(int) constructor?

#include <iostream>
#include <type_info>

using namespace std;

struct Value { Value(int) { } };

struct Convertible
{
    template<class T>
    operator T() const
    { throw typeid(T).name(); }
};

int main()
{
    try { Value w((Convertible())); }
    catch (char const *s) { cerr << s << endl; }
}

Edit:

Even more bizarre is this (this time it's C++11 only, on GCC 4.7.2):

#include <iostream>
#include <typeinfo>

using namespace std;

struct Value
{
    Value(Value const &) = delete;
    Value(int) { }
};

struct Convertible
{
    template<class T>
    operator T() const
    { throw typeid(T).name(); }
};

int main()
{
    try { Value w((Convertible())); }
    catch (char const *s) { cerr << s << endl; }
}

Which gives:

source.cpp: In function 'int main()':
source.cpp:21:32: error: call of overloaded 'Value(Convertible)' is ambiguous
source.cpp:21:32: note: candidates are:
source.cpp:9:3: note: Value::Value(int)
source.cpp:8:3: note: Value::Value(const Value&) <deleted>

If the copy constructor is deleted, then why is there any ambiguity?!

Answer 1

In the first example Visual Studio is incorrect; the call is ambiguous. gcc in C++03 mode prints:

source.cpp:21:34: error: call of overloaded 'Value(Convertible)' is ambiguous
source.cpp:21:34: note: candidates are:
source.cpp:9:5: note: Value::Value(int)
source.cpp:6:8: note: Value::Value(const Value&)

Recall that a copy constructor is implicitly defaulted. The governing paragraph is 13.3.1.3 Initialization by constructor [over.match.ctor]:

When objects of class type are direct-initialized [...], overload resolution selects the constructor. For direct-initialization, the candidate functions are all the constructors of the class of the object being initialized.

In the second example, deleted functions participate equally in overload resolution; they only affect compilation once overloads have been resolved, when a program that selects a deleted function is ill-formed. The motivating example in the standard is of a class that can only be constructed from floating-point types:

struct onlydouble {
  onlydouble(std::intmax_t) = delete;
  onlydouble(double);
};