I have this function:
std::string SWindows::PRecv(int rc, std::string* ip);
In this function, in some case I return NULL and the code compile (IDE : VS 2012). For me, NULL it's not a std::string object so we can't write that.
So why it's correct to return NULL when the return of the function is a std::string ?
The value NULL is equivalent to 0 in C++ and 0 is a perfectly valid char const* constant. The string class has a non-explicit constructor taking char const*, i.e. the compiler will compile the code but it won't work: that is, although the compiler excepts the code, it is undefined behavior.
Looking at the definition in C++11 I don't see a deleted overload taking a std::nullptr_t, i.e., it will also compile when converting from a nullptr. I would have hoped that there would be a deleted overload taking a std::nullptr_t which would allow the compiler to catch the situation, at least, when passing nullptr.
string(0) is okay, when it's actually UB.
void const * deleted overload would make string(nullptr) ambiguous, which may accomplish the same objective.What you're doing is constructing a std::string from a char const* equal to 0 (or NULL).
This is actually prohibited by the standard:
[C++11: 21.4.2/9]Requires:sshall not be a null pointer
However, GCC is extra-kind to you and does checks to make it a no-op. It is not required to and you shouldn't ever rely on this.
