openssh unix dev - Jun 2021 - [PATCH v2 2/2] Remove trailing semicolon after RB_GENERATE_STATIC

On Thu, 13 May 2021, Michael Forney wrote:
> This expands to a series of function definitions, so the semicolon is
> not necessary (in fact, it is not allowed in ISO C).
I went looking for that, and failed to find it.? The best I could find 
says otherwise.? ISO/IEC 9899:2017 (C17): Section 6.8.3 Expression and 
null statements specifically allows a null statement (as you'd expect 
given the section name).

It must be a new revision.? When did the null statement become 
disallowed?? Reference, please.

If the null statements are still allowed, I urge that the patch be 
reverted as it would then be mere noise in the change history, a 
distraction at best, and a source of errors at worst.

The benefit of the semi-colon (if allowed) is that it makes explicit 
that the macro is a psuedo-statement.? Also, if the macro is redefined 
to produce an expression that is not a (terminated) statement, the 
program will no longer compile.

In case you hadn't noticed, I look unfavourably on trivial, janitorial 
patches.

On 2021-06-04, David Newall <openssh at davidnewall.com>
wrote:
> On Thu, 13 May 2021, Michael Forney wrote:
>> This expands to a series of function definitions, so the semicolon is
>> not necessary (in fact, it is not allowed in ISO C).
>
> I went looking for that, and failed to find it.  The best I could find
> says otherwise.  ISO/IEC 9899:2017 (C17): Section 6.8.3 Expression and
> null statements specifically allows a null statement (as you'd expect
> given the section name).
>
> It must be a new revision.  When did the null statement become
> disallowed?  Reference, please.
I'm not sure why statements are being discussed here. The patch is
about top-level external definitions (declarations and function
definitions), not statements. As you point out, null statements are
perfectly valid. Null declarations are not.

The relevant sections of the C standard are external definitions
(https://port70.net/~nsz/c/c99/n1256.html#6.9) and declarations
(https://port70.net/~nsz/c/c99/n1256.html#6.7).

I'll reproduce these parts of the grammar:

          translation-unit:
                  external-declaration
                  translation-unit external-declaration
          external-declaration:
                 function-definition
                 declaration

          function-definition:
                 declaration-specifiers declarator declaration-listopt
compound-statement
          compound-statement:
                { block-item-listopt }

          declaration:
                 declaration-specifiers init-declarator-listopt ;
          declaration-specifiers:
                 storage-class-specifier declaration-specifiersopt
                 type-specifier declaration-specifiersopt
                 type-qualifier declaration-specifiersopt
                 function-specifier declaration-specifiersopt

As you can see, a function definition ends with a '}', not an optional
semicolon. Additionally, there is no "null declaration",
declaration-specifiers are a  required part of a declaration. This has
always been the case, from C89 to the latest C23 draft.

If you enable -Wpedantic, gcc will flag these null declarations:

$ echo ';' | gcc -Wpedantic -c -x c -
<stdin>:1:1: warning: ISO C does not allow extra ';' outside of a
function [-Wpedantic]
> If the null statements are still allowed, I urge that the patch be
> reverted as it would then be mere noise in the change history, a
> distraction at best, and a source of errors at worst.
>
> The benefit of the semi-colon (if allowed) is that it makes explicit
> that the macro is a psuedo-statement. Also, if the macro is redefined
> to produce an expression that is not a (terminated) statement, the
> program will no longer compile.
If the macro is redefined to produce an expression or statement, it
won't compile because statements and expressions aren't allowed at
top-level, only in function bodies. The best you could do if you
wanted to add a semicolon after a macro that expands to a function
definition would be to require C11 and use a dummy `_Static_assert(1,
"")` at the end.

If you care deeply about extending C with null declarations, I suggest
you send a proposal to WG14.