[gumby]

I don’t read those extracts as implying the same result as you do. The result type (promoted type in the C++ case) will still be unsigned, yet will be in most cases (especially small ones) full of 1s where there were zeros, and won’t compute the expected modulus. Draw out the bit patterns of 1’s vs 2’s complement for (say) n=5 and n=65535 and see if I have it wrong.

[Arnavion]

Not sure what you're trying to say. The point of what I wrote, and what the article says, is that `-n` is equivalent to `2^N - n`, and thus `-n % n` is equivalent to `(2^N - n) % n`.

If 2^N - n happens to be a number that's "full of 1s where there were zeros", then that's what it's supposed to be. I don't understand why you think it "won't compute the expected modulus". That's exactly what it will do.

If N is 32 and n is 5, then -n is 4294967291 and -n % n is 1. It doesn't matter whether that was calculated on a ones complement or twos complement machine.