Consider the following code.

 class C {
     C() { }
     C(int) { }
 public:
     static C *pc;  // (1a)
     static C c1;   // (2a)
     static C c2;   // (3a)
 };

 C *C::p = new C;  // (1b)
 C C::c;           // (2b)
 C C::c(1);        // (3b)

Cfront 3.0.1 gives errors for variant 2.
G++ 2.5.8 gives errors for variants 2 and 3.

Reading the ARM r9.4:
 The initializer for a static data meber can access anything
 that a static member function of the class can.

This implies that (1) and (3) are legal.  My gut feeling is that (2)
is legal as well, but am I correct?

Secondly, would it not be a good idea to allow a private destructor to
be implicitly called for static members of a class?  In other words,
I want the code above to be legal even if `C::~C()' were private.  Any
thoughts on this one?

-- Olaf Weber

>>>>> Olaf Weber <olaf@cwi.nl> writes:

> G++ 2.5.8 gives errors for variants 2 and 3.

g++ 2.6.0 will accept all three variants.

> Reading the ARM r9.4:
>  The initializer for a static data meber can access anything
>  that a static member function of the class can.

> This implies that (1) and (3) are legal.  My gut feeling is that (2)
> is legal as well, but am I correct?

I believe that you are.

> Secondly, would it not be a good idea to allow a private destructor to
> be implicitly called for static members of a class?  In other words,
> I want the code above to be legal even if `C::~C()' were private.  Any
> thoughts on this one?

I think that this is also the intent of the passage you quote.

Jason