TITLE: the legality of null statements

(Newsgroups: comp.lang.c++.moderated, 7 Jun 99)

GLASSBOROW: Francis Glassborow <francis@robinton.demon.co.uk>

>> namespace A
>> {
>> };
>>
>> It is actually an error to have a semicolon here (a purely technical
>> language law point AFAIK)


ERDEI: ccg@freemail.c3.hu (Andras Erdei)

> My silly question: is it an error? Isn't it just dummy?
>
> I'm routinely terminating my fns with } ;


CLAMAGE: Steve Clamage <clamage@eng.sun.com>

It is an error at namespace scope, although not all compilers
enforce it.

Null statements are valid:
 x = y;
 ; // null statement
 while( ++*p = ++*q )
     ; // null statement

Null declarations are not valid, because there is no syntax
production that produces a null declaration. At namespace scope
there are no statements, only declarations and definitions.
At block scope, a declaration is really a declaration-statement,
so a null statement can appear between declarations.
 int i = 3;
 ; // error at namespace scope, OK at block scope
 int j = 4;

At class scope, there are no null declarations, but there is
a special rule for member functions. A semicolon following
an in-class member function definition is optional:
 class C {
     int f() { return 0; }   // OK
     int g() { return 1; } ; // also OK
     int x;
     ; // error
 };

The rules are the same in C for the C subset of C++.

If you think the rules are inconsistent and peculiar, I would not
disagree. :-)