Re: c++/10086: static const int unresolved in ? : construct

[Gabriel Dos Reis <gdr at integrable-solutions dot net>]

"Giovanni Bajo" <giovannibajo at libero dot it> writes:

| http://gcc.gnu.org/cgi-bin/gnatsweb.pl?cmd=view%20audit-trail&database=gcc&p
| r=10086

I think arguments can be made for both sides (bug or not a bug) in
this particular case.  Certainly understand what is going on involves
understanding how the compiler is treating internal the expression 

    v ? A : B

Certainly, it is not an integrable constant expression and it is an
lvalue.  Wihch means the compiler is internally taking the address of
both Base::A and Base::B.  Hence the link error.

| I'm not sure what it is going on. I know that the standard says that you
| still have to define an initialized static const member (even if there is an
| open defect report about this, I believe). But then, why this should work:
| 
| struct Foo
| {
|    static const int A = 0;
| };
| 
| int main(void)
| {
|    return Foo::A;
| }
| 
| This one compiles and links correctly on G++ 3.2, but it should not. 

In the above, Foo::A is used in a such way that its definition isn't
needed, i.e. its address isn't taken.

I agree with you that all that is confusing.  A reason why I've always
thought of that as a misfeature -- it doesn't scale easily.

-- Gaby