Calculating CRC16 on Vec

Question

I'm sending and receiving raw binary data via the serial port, so I have a predefined message stored in a u8 vector. I need to calculate the 16bit CRC and append that onto the end before sending it, however I keep running into issues with casting and integer overflows. This is how I have previously done the calculation in C:

void Serial::AddCRC(void *data, int len, void *checksum)
{
    uint8_t *dataPtr = (uint8_t *)data;
    uint8_t *crcPtr = (uint8_t *)checksum;
    uint16_t crc = 0x0;
    unsigned char x;
    int byteCount = 0;

    while ((len--) > 0) {
        x = (unsigned char)(crc >> 8 ^ dataPtr[byteCount++]);
        x ^= (unsigned char)(x >> 4);
        crc = (uint16_t)((crc << 8) ^ (x << 12) ^ (x << 5) ^ x);
    }
    crcPtr[0] = crc >> 8;
    crcPtr[1] = crc &0x00ff;
}

I tried to do something similar in rust, but first ran into some borrow checker issues, so I tried to simplify it and just write a function to calculate the crc and return the u16 result, without needing to mutate the vector:

#[allow(overflowing_literals)]
pub fn checksum(msg: &Vec<u8>) -> u16{
    if msg.len() == 0 {return 0}

    let crc: u16 = 0x0;
    let x: u16;
    for byte in msg.iter() {
        x = crc >> 8 ^ byte;
        x ^= x >> 4;
        crc = (crc << 8) ^ (x << 12) ^ (x << 5) ^ x;
    }
    crc
}

however I can't find a way to make this work. The code posted above fails to compile because I can't perform a bitwise xor between a u8 and a u16, however the data is treated as u8s because its raw bytes, so that can't change. I could add the mut to the vector and make it mutable then cast to a u16 but that seems like a risky way of doing it, and I shouldn't need to mutate the vector to calculate the checksum:

error[E0308]: mismatched types
  --> vips_interface/src\beacon_comms.rs:14:24
   |
14 |         x = crc >> 8 ^ byte;
   |                        ^^^^ expected `u16`, found `u8`

error[E0277]: no implementation for `u16 ^ &u8`
  --> vips_interface/src\beacon_comms.rs:14:22
   |
14 |         x = crc >> 8 ^ byte;
   |                      ^ no implementation for `u16 ^ &u8`

What is the best way of implementing a similar function in rust? The rust compiler is great for catching integer type overflows, but unfortunately that's a key part of how the CRC works, hence why I allowed the overflowing literals, but this didn't seem to fix it. I had a look at some of the crates that mentioned CRC calculations but none of them offered what I wanted, plus its a fairly simple calculation so I'd rather use this as a learning exercise.

Answer 1

I didn't look at the details, just made it compile.

First, Vec is not needed, any slice is suitable (even if it comes from a Vec).

byte is a reference to a u8 in this slice, thus *byte is a copy of this u8 (since u8 is Copy), then we can cast it into a u16.

crc is updated in the loop, so it needs the mut qualifier.

x is local to the loop, and is updated in several steps so it needs the mut qualifier too.

n.b. : in your C++ code there is (x << 12) with x declared as unsigned char. In C/C++ integer promotion takes place and this x is promoted to int before the shift. In Rust, if x is a u8, this shift will give zero in any case (and Rust complains about that). I let x be a u16 and cancel the upper byte, but I'm not certain this is what you want.

pub fn checksum(msg: &[u8]) -> u16 {
    let mut crc: u16 = 0x0;
    for byte in msg.iter() {
        let mut x = ((crc >> 8) ^ (*byte as u16)) & 255;
        x ^= x >> 4;
        crc = (crc << 8) ^ (x << 12) ^ (x << 5) ^ x;
    }
    crc
}

Answer 2

This version is vetted against a popular implementation of it in Python (Klipper project) and gives the same answer as that (the one above did not)

pub fn checksum(msg: &[u8]) -> u16 {
   let mut crc: u16 = 0xffff;
   for byte in msg.iter() {
      let mut x = (*byte as u16) ^ (crc & 0xff);
      x ^= (x & 0x0f) << 4;
      crc = ((x << 8) | (crc >> 8)) ^ (x >> 4) ^ (x << 3);
   }
   crc
}