64k.c

#include "stdint.h"
#include "stdio.h"

void main(int argc, void *argv[]) {
  uint64_t i;
  for (i = 0; i < 2^32; i++) {
    uint32_t x = (uint32_t)i;
    uint32_t l = (-x-1) & 0xffff;
    uint32_t r = ~x & 0xffff;
    if (l != r) {
      printf("Differs for %x: (%x, %x)\n", x, l, r);
    }
  }
}

2^32 is 2 XOR 32 (i.e., 34), not UINT32_MAX (i.e., 4294967295). Then again, your relational operator takes precedence over the bit twiddling. As to your original question: two's complement guarantees that -n - 1 equals ~n. Even K&R got away with assuming that representation.