RC4 PRNG unrolled with C++ templates

rc4.cpp

#include <cstdint>
#include <cstddef>
#include <cstdio>
#include <cstring>

struct RC4State {
  uint8_t i, j;
  uint8_t S[256];

  RC4State(uint8_t *key, size_t key_len) : i(0), j(0) {
    size_t i2;
    uint8_t j2 = 0;
    for (i2 = 0; i2 < 256; i2++) {
      S[i2] = i2;
    }
    for (i2 = 0; i2 < 256; i2++) {
      j2 += S[i2] + key[i2 % key_len];
      Swap(i2, j2);
    }
  }

  uint8_t GenerateOutput() {
    i += 1;
    j += S[i];
    Swap(i, j);
    return S[(S[i] + S[j]) % 256];
  }

private:
  void Swap(uint8_t i, uint8_t j) {
    uint8_t temp = S[i];
    S[i] = S[j];
    S[j] = temp;
  }
};

// Fill out with n successive PRNG outputs
// return the last one.
template <size_t n>
struct RC4Output {
  static uint8_t generate(RC4State& s, uint8_t *out) {
    RC4Output<n-1>::generate(s, out);
    uint8_t K = s.GenerateOutput();
    out[n-1] = K;
    return K;
  }
};

template <>
struct RC4Output<0> {
  static uint8_t generate(RC4State& s, uint8_t *out) {
    return 0;
  }
};

int main(int argc, char **argv) {
  if (argc != 2) {
    puts("Supply one argument, key to use");
    return 1;
  }
  RC4State s((unsigned char *)argv[1], strlen(argv[1]));

  uint8_t output[256];

  RC4Output<256>::generate(s, output);

  for (int i = 0; i < 256; i++) {
    printf("%02x", output[i]);
  }
  printf("\n");

  return 0;
}