Global anonymous unions must be declared static

[Rick C. Hodgin]

I'm using Visual Studio 2010 for a particular application, and its VC++
compiler has the requirement that this code declared with a static
global anonymous union:

// Fails:
union {
int a1;
HWND w1;
};

// Succeeds:
static union {
int a2;
HWND w2;
};

int APIENTRY WinMain(HINSTANCE hInst, HINSTANCE hPrevInst,
LPSTR cmdLine, int nShow)
{
// Code here
}

Is it a C++ requirement that global anonymous unions be declared static,
or just Microsoft's VC++ compiler requirement? If it's a C++ requirement,
why is it a requirement? What would be the difference between a global
anonymous union and a static global anonymous union? How would the
compiler see them as different?

Best regards,
Rick C. Hodgin

[David Brown]

It is a C++ thing - gcc gives the error "namespace-scope anonymous
aggregates must be static" for the first version.

Both versions are valid C. Both versions are utterly useless (in C and
C++), because they don't define or declare a variable, a typedef, or a
union tag.

I am not sure why C++ rejects the first version - since the union is not
named, and does not declare anything, it has no effect that I can see.
Thus I don't see why the first version is "worse" than the second one.

[Ian Collins]

On 10/19/16 07:51 AM, Rick C. Hodgin wrote:
> I'm using Visual Studio 2010 for a particular application, and its VC++
> compiler has the requirement that this code declared with a static
> global anonymous union:
>
> // Fails:
> union {
> int a1;
> HWND w1;
> };
>
> // Succeeds:
> static union {
> int a2;
> HWND w2;
> };
>

> Is it a C++ requirement that global anonymous unions be declared static,
> or just Microsoft's VC++ compiler requirement?

C++. See section 9.5 pp5&6 of the C++11 standard.

> If it's a C++ requirement,
> why is it a requirement? What would be the difference between a global
> anonymous union and a static global anonymous union? How would the
> compiler see them as different?

As a guess I'd say it has to be static because the members are in the
scope in which the union is declared.

--
Ian

[Ian Collins]

On 10/19/16 08:17 AM, David Brown wrote:

> Both versions are valid C. Both versions are utterly useless (in C and
> C++), because they don't define or declare a variable, a typedef, or a
> union tag.
>
> I am not sure why C++ rejects the first version - since the union is not
> named, and does not declare anything, it has no effect that I can see.
> Thus I don't see why the first version is "worse" than the second one.

It introduces its member variables into the scope in which it is
declared, so we can write:

static union
{
int a1;
int w1;
};

int main()
{
a1 = 42;
}

whether we would want to is another matter!

--
Ian