array default- vs. list-initialization

Question

I have the following code, a simplification of an array-like structure:

template<typename T, size_t N>
struct X
{
    T a[N];

    template<typename... A>
    explicit X(A&&... a) : a{std::forward<A>(a)...} { }  // ERROR (2)
};

int main ()
{
    X<int,3> x;           // OK
    X<X<int,3>,2> y{x,x}; // OK
    X<X<int,3>,2> z;      // ERROR (1)
}

This is compiling fine in clang 3.3 and gcc 4.8.1, both with -std=c++11. I am trying to upgrade gcc, so I now try 4.9.0. In this case, the 3rd example (ERROR (1)) instantiates X's constructor (ERROR (2)), at which point the compiler reports

error: converting to 'X<int, 3ul>' from initializer list would use explicit
   constructor 'X<T, N>::X(A&&...) [with A = {}; T = int; long unsigned int N = 3ul]

This last example attempts to default-initialize array z and so also its included arrays; however, if I get this right, here gcc essentially says that the included arrays are being list-initialized by {}, which is not allowed since the constructor is explicit.

The error is gone if I add another default constructor of either of the following forms:

explicit X() {}
explicit X() : a() {}

but not

explicit X() : a{} {}

This workaround is not difficult, but any idea who's wrong and who's right, just so I know what I'm doing and why?

The simplest fix is to add X() : a{} {} (without the explicit). I think this is related to gcc.gnu.org/PR60416 and open-std.org/jtc1/sc22/wg21/docs/cwg_active.html#1518 — Jonathan Wakely
– Jonathan Wakely, Commented Mar 27, 2014 at 19:38
I think this is a GCC 4.9 regression, which I've reported to bugzilla, but as issue 1518 shows, the standard isn't really clear what should happen. — Jonathan Wakely
– Jonathan Wakely, Commented Mar 27, 2014 at 19:49
@JonathanWakely The real simplest fix is to use std::array and get the same functionality without reinventing the wheel :-P — IdeaHat
– IdeaHat, Commented Mar 27, 2014 at 19:54
@MadScienceDreams No. using std::array<T,N> a; instead of T a[N]; does not change anything. The problem remains. — iavr
– iavr, Commented Mar 27, 2014 at 20:49
Follow the link from PR60416 to gcc.gnu.org/PR60417 and you'll see the original testcase was fixed, but I've added an new testcase as a comment. It's the same issue. — Jonathan Wakely
– Jonathan Wakely, Commented Mar 27, 2014 at 22:09

Jonathan Wakely · Accepted Answer · 2014-03-28 11:47:19Z

3

This is a GCC bug, PR 60417.

An older change for PR 54835 was intended to implement the C++ committee's proposed direction for fixing core issue 1518. Unfortunately that change breaks some valid C++03 programs, as shown by the first example in PR 60417. A fix was committed for PR 60417, but it only handles some cases. Specifically, it doesn't fix the case where list-initializing an array of types with explicit constructors, as in this question.

answered Mar 28, 2014 at 11:47

Jonathan Wakely

172k28 gold badges360 silver badges540 bronze badges

Sign up to request clarification or add additional context in comments.

Collectives™ on Stack Overflow

array default- vs. list-initialization

1 Answer 1

Comments

Your Answer

Hot Network Questions

Collectives™ on Stack Overflow

1 Answer 1

Comments

Your Answer

Sign up or log in

Post as a guest

Related