Bruno Haible wrote:
> Paul Eggert wrote:
>> portable programs should not pass a NULL format to 'error', so this is not
>> a gnulib bug.
>
> Yes. But why does it not yield a GCC warning?
>
> ======================== foo.c =========================
> #include <errno.h>
> #include <stdlib.h>
> extern void error (int __status, int __errnum, __const char *__format, ...)
> __attribute__ ((__format__ (__printf__, 3, 4)))
> #ifdef DECLARE_NONNULL
> __attribute__ ((__nonnull__(3)))
> #endif
> ;
>
> void foo ()
> {
> error (EXIT_FAILURE, EPERM, NULL);
> }
> =====================================================
>
> $ gcc -Wformat=2 -c foo.c
> $ gcc -Wformat=2 -c foo.c -DDECLARE_NONNULL
> foo.c: In function 'foo':
> foo.c:12: warning: null argument where non-null required (argument 3)
>
> Should the declaration in glibc's and gnulib's error.h be changed to contain a
> __attribute__ ((__nonnull__(3))) ?
>
> Or should gcc be changed so that __attribute__ ((__format__ (__printf__, m,
> n)))
> implies __attribute__ ((__nonnull__(m))) ?
Good point.
IMHO the latter (enhancing gcc) makes more sense.
