Re: Warnings from latest HEAD (zsh-5.9.0.3-test-357-g599af46) build

On Mon, Jun 15, 2026 at 1:38 AM Bart Schaefer <[email protected]> wrote:
>
> Apple clang version 17.0.0 (clang-1700.6.4.2)
> Target: arm64-apple-darwin24.6.0
> Thread model: posix
>
> These are all from "54710: handle multibyte arguments in error messages"
>
> gcc -c -I. -I../Src -I../Src -I../Src/Zle -I. -I/opt/local/include
> -DHAVE_CONFIG_H -Wall -Wmissing-prototypes -ggdb  -o exec.o exec.c
> exec.c:3303:8: warning: label followed by a declaration is a C23
> extension [-Wc23-extensions]
>  3303 |                             convchar_t opt = unmeta_one(cmdopt, NULL);
>       |                             ^
> 1 warning generated.compctl.c:544:4: warning: label followed by a
> declaration is a C23 extension [-Wc23-extensions]
>   544 |                         convchar_t s = unmeta_one(p, NULL);
>       |                         ^
> 1 warning generated.complete.c:800:3: warning: label followed by a
> declaration is a C23 extension [-Wc23-extensions]
>   800 |                 convchar_t c = unmeta_one(p, NULL);
>       |                 ^
> complete.c:1168:2: warning: label followed by a declaration is a C23
> extension [-Wc23-extensions]
>  1168 |         convchar_t c = unmeta_one(&argv[0][1], NULL);
>       |         ^
> 2 warnings generated.
>
> I'm surprised glob.c doesn't produce a similar error from
>
>                 default:
>                     untokenize(--s);
> -                   zerr("unknown file attribute: %c", *s);
> +                   convchar_t attr = unmeta_one(s, NULL);
> +                   zerr("unknown file attribute: %c", attr);
>                     restore_globstate(saved);
>                     return;
>
> IIRC we've been trying to stick to the "declarations only at the start
> of a block" convention.

I think we gave up on that a while ago, the problem here is actually
the opposite. The untokenize(--s); statement being between the label
and the variable declaration makes the glob.c instance ok. Of course,
the warning was intended to protect against jumping past the
initialization of the variable, but it's insufficient exactly because
it doesn't fire in this case. It's also not a problem in these
instances because they're all in the final label so you can't jump
past them, but I'll change them to either declare convchar_t earlier
or add braces on the default case, whichever seems nicer.

-- 
Mikael Magnusson