On 2021-06-04, David Newall <openssh@xxxxxxxxxxxxxxx> 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. _______________________________________________ openssh-unix-dev mailing list openssh-unix-dev@xxxxxxxxxxx https://lists.mindrot.org/mailman/listinfo/openssh-unix-dev
