On Mon, May 16, 2005 at 09:59:04AM -0600, Todd C. Miller wrote:
> In message <20050516151748.GA1337@moule.localdomain>
> so spake Alexandre Ratchov (alex-contact):
>
> > int ret;
> > char foo[100];
> >
> > ...
> > ret = snprintf(foo, sizeof(foo), "%s", input);
> > if (ret < 0 || (size_t)ret >= sizeof(foo))
> >
> >
> > -1 means that snprintf failed and negative number of chars is meaningless.
> > This works if sizeof(foo) > INT_MAX because res is converted to size_t.
> > Have i missed something?
>
> While this does look OK I really don't see the need for the cast
> to size_t. In practice sizeof(foo) will not be > INT_MAX unless
> you are creating 2GB data structures on the stack and the stack is
> not big enough to hold such a data structure anyway.

I agree that cast is not necessary since even lint doesn't complain
about it. It doesn't hurt to have the cast either, since that cast
_will_ happen implicitly anyway. In the above program it doesn't matter
since a positive number has a 0 for the most significant bit and remains
the same when cast to unsigned.

If sizeof returns size_t, and size_t is typedefed as unsigned int,
according to the implicit expression casting rules,
int _will_ be cast into unsigned when compared to an unsigned number.

For example this program

#include <stdio.h>
#include <limits.h>

int
main(int argc, char *argv[])
{
int x = INT_MIN;
size_t y = INT_MAX;

printf("INT_MIN = %d\nINT_MAX = %d\n", INT_MIN, INT_MAX);
if (x > y)
printf("\nCast: INT_MIN > INT_MAX unsigned\n");

return 0;
}

produces

$ make tryme
cc -O2 -o tryme tryme.c
$ ./tryme
INT_MIN = -2147483648
INT_MAX = 2147483647

Cast: INT_MIN > INT_MAX unsigned

Thus, with or without explicit cast to size_t, the cast will happen.

Regards,

Zvezdan Petkovic