It's never too late to learn something new. And sometimes the compiler is smarter than what you think. In this case, I've discovered this behavior a couple of weeks ago while I was working on a C function that manipulates the bits of an unsigned integer variable using the right-shift operator (>>).

The function worked pretty well, until someone wanted to use it with signed integers, too. The issue here is that the sign bit (the leftmost one) is replaced with zero when right-shifting. The solution has been absurdly easy: cast to signed int.

This is an example with 8 bits:

The right shift produces different results depending on the signedness of the variable

Too Good To Be True?

This is wonderful but there is a little thing to consider when you plan to use it. According to the C standard:

1196 If E1 has a signed type and a negative value, the resulting value is implementation-defined.

This means that every compiler can behave in its own way and, worst of all, for different architectures the same compiler can act differently. So, what's the solution?

There isn't just one, of course. The following function should work pretty fine in all cases, but I'm sure you can find at least another algorithm to do the same thing.

int right_shift_with_sign(int var, unsigned int amount)
{
        int mask = (var < 0) ? -1 >> amount : -1;
        int result = var >> amount;
        return (result | ~mask);
}

[Code not fully tested - use it at your own risk]

If you need way to detect how your compiler behave, you can use this short snippet:

if ((-1 >> 2) == -1) {
        /* right shift preserves the sign */
} else {
        /* right shift DOES NOT preserve the sign */
}