This code, when compiled with Visual Studio 2010
#include <cstdlib>
#include <cstdio>
int numbers[] = { 23, 24, 25, 25, 28, 20, 20 };
int main(void) {
int d = -1, x=0;
size_t count = sizeof(numbers) / sizeof(numbers[0]);
if (d <= (sizeof(numbers) / sizeof(numbers[0]))-2)
x = numbers[d+1];
if (d <= count-2)
x = numbers[d+1];
}
gives me a signed/unsigned mismatch warning on if (d <= count-2)
but not on if (d <= (sizeof(numbers) / sizeof(numbers[0]))-2)
. Why is this? I have enabled all warnings.
It is a bug that the Visual C++ compiler does not emit warning C4018 in this case. See the Microsoft Connect bug report, "warning C4018 (signed/unsigned mismatch) for sizeof operator."
The issue appears to be only indirectly related to usage of sizeof
. The compiler also fails to emit the warning in other cases where a constant value of unsigned type is involved in the comparison. For example, if you add const
-qualification to count
, C4018 is not emitted even for your second comparison.
My guess is that it treats 2
as an int so promotes
(sizeof(numbers) / sizeof(numbers[0]))-2)
to an int because of the brackets and so the comaprison is int < int
size_t
is defined as unsigned so you get a signed unsigned conflict.
Please note, as sizeof
is evaluated during compile time, and is being replaced by a constant value, it does not complain about sign mismatch, but in the other expression, you explicitly convert the value to a size_t variable which can possibly overflow when you try to compare with a signed number.