verit-core 0.2.0

//! Content hashing for schema ids.
//!
//! The schema id is a **content address**: two parties that build the same
//! logical schema must derive the same id, and an adversary must not be able
//! to cheaply find two schemas sharing one id (schema-confusion resistance,
//! and safe use as a registry key). That rules out a fast non-cryptographic
//! hash like FNV, whose structure makes collisions trivial to construct.
//!
//! We use SHA-256 truncated to 128 bits. (BLAKE3 would be a fine choice too;
//! SHA-256 is chosen here because it is simple enough to implement correctly
//! with zero dependencies — the whole crate stays dependency-free — and the
//! schema hash is computed once per schema build, never on the hot path.)
//!
//! 128 bits gives ~2^64 birthday-collision resistance; a 64-bit id would give
//! only ~2^32, which is within reach of a determined adversary — hence the
//! wider id.

/// The schema id: the first 128 bits of SHA-256 over the canonical schema
/// bytes, read big-endian.
pub fn schema_id(bytes: &[u8]) -> u128 {
    let digest = sha256(bytes);
    let mut id = [0u8; 16];
    id.copy_from_slice(&digest[..16]);
    u128::from_be_bytes(id)
}

const K: [u32; 64] = [
    0x428a2f98, 0x71374491, 0xb5c0fbcf, 0xe9b5dba5, 0x3956c25b, 0x59f111f1, 0x923f82a4, 0xab1c5ed5,
    0xd807aa98, 0x12835b01, 0x243185be, 0x550c7dc3, 0x72be5d74, 0x80deb1fe, 0x9bdc06a7, 0xc19bf174,
    0xe49b69c1, 0xefbe4786, 0x0fc19dc6, 0x240ca1cc, 0x2de92c6f, 0x4a7484aa, 0x5cb0a9dc, 0x76f988da,
    0x983e5152, 0xa831c66d, 0xb00327c8, 0xbf597fc7, 0xc6e00bf3, 0xd5a79147, 0x06ca6351, 0x14292967,
    0x27b70a85, 0x2e1b2138, 0x4d2c6dfc, 0x53380d13, 0x650a7354, 0x766a0abb, 0x81c2c92e, 0x92722c85,
    0xa2bfe8a1, 0xa81a664b, 0xc24b8b70, 0xc76c51a3, 0xd192e819, 0xd6990624, 0xf40e3585, 0x106aa070,
    0x19a4c116, 0x1e376c08, 0x2748774c, 0x34b0bcb5, 0x391c0cb3, 0x4ed8aa4a, 0x5b9cca4f, 0x682e6ff3,
    0x748f82ee, 0x78a5636f, 0x84c87814, 0x8cc70208, 0x90befffa, 0xa4506ceb, 0xbef9a3f7, 0xc67178f2,
];

const H0: [u32; 8] = [
    0x6a09e667, 0xbb67ae85, 0x3c6ef372, 0xa54ff53a, 0x510e527f, 0x9b05688c, 0x1f83d9ab, 0x5be0cd19,
];

/// A straightforward, allocation-free SHA-256 (FIPS 180-4). Not constant-time
/// — it hashes public schema bytes, not secrets.
pub fn sha256(input: &[u8]) -> [u8; 32] {
    let mut h = H0;

    // Message schedule padding: len is fine as u64 bit-count.
    let bit_len = (input.len() as u64).wrapping_mul(8);
    let mut padded = Vec::with_capacity(input.len() + 9 + 63);
    padded.extend_from_slice(input);
    padded.push(0x80);
    while padded.len() % 64 != 56 {
        padded.push(0);
    }
    padded.extend_from_slice(&bit_len.to_be_bytes());

    let mut w = [0u32; 64];
    for block in padded.chunks_exact(64) {
        for i in 0..16 {
            w[i] = u32::from_be_bytes([
                block[i * 4],
                block[i * 4 + 1],
                block[i * 4 + 2],
                block[i * 4 + 3],
            ]);
        }
        for i in 16..64 {
            let s0 = w[i - 15].rotate_right(7) ^ w[i - 15].rotate_right(18) ^ (w[i - 15] >> 3);
            let s1 = w[i - 2].rotate_right(17) ^ w[i - 2].rotate_right(19) ^ (w[i - 2] >> 10);
            w[i] = w[i - 16]
                .wrapping_add(s0)
                .wrapping_add(w[i - 7])
                .wrapping_add(s1);
        }

        let [mut a, mut b, mut c, mut d, mut e, mut f, mut g, mut hh] = h;
        for i in 0..64 {
            let s1 = e.rotate_right(6) ^ e.rotate_right(11) ^ e.rotate_right(25);
            let ch = (e & f) ^ ((!e) & g);
            let t1 = hh
                .wrapping_add(s1)
                .wrapping_add(ch)
                .wrapping_add(K[i])
                .wrapping_add(w[i]);
            let s0 = a.rotate_right(2) ^ a.rotate_right(13) ^ a.rotate_right(22);
            let maj = (a & b) ^ (a & c) ^ (b & c);
            let t2 = s0.wrapping_add(maj);
            hh = g;
            g = f;
            f = e;
            e = d.wrapping_add(t1);
            d = c;
            c = b;
            b = a;
            a = t1.wrapping_add(t2);
        }
        h[0] = h[0].wrapping_add(a);
        h[1] = h[1].wrapping_add(b);
        h[2] = h[2].wrapping_add(c);
        h[3] = h[3].wrapping_add(d);
        h[4] = h[4].wrapping_add(e);
        h[5] = h[5].wrapping_add(f);
        h[6] = h[6].wrapping_add(g);
        h[7] = h[7].wrapping_add(hh);
    }

    let mut out = [0u8; 32];
    for (i, word) in h.iter().enumerate() {
        out[i * 4..i * 4 + 4].copy_from_slice(&word.to_be_bytes());
    }
    out
}

/// CRC-32 (IEEE 802.3): reflected, polynomial `0xEDB8_8320`, init and final
/// XOR `0xFFFF_FFFF`. Used by the `.verit` file footer to prove a commit was
/// not torn by a crash (File Format Specification §7).
///
/// Bitwise rather than table-driven on purpose: it only ever covers the 40-byte
/// footer prefix, so a 256-entry table would cost more than it saves, and the
/// loop is trivially portable — every implementation has to reproduce it.
pub fn crc32(bytes: &[u8]) -> u32 {
    let mut crc: u32 = 0xFFFF_FFFF;
    for &byte in bytes {
        crc ^= byte as u32;
        for _ in 0..8 {
            // Branchless: mask is all-ones when the low bit is set, else zero.
            let mask = (crc & 1).wrapping_neg();
            crc = (crc >> 1) ^ (0xEDB8_8320 & mask);
        }
    }
    !crc
}

#[cfg(test)]
mod tests {
    use super::{crc32, sha256};

    fn hex(d: &[u8]) -> String {
        d.iter().map(|b| format!("{b:02x}")).collect()
    }

    #[test]
    fn nist_vectors() {
        // FIPS 180-4 / standard test vectors.
        assert_eq!(
            hex(&sha256(b"")),
            "e3b0c44298fc1c149afbf4c8996fb92427ae41e4649b934ca495991b7852b855"
        );
        assert_eq!(
            hex(&sha256(b"abc")),
            "ba7816bf8f01cfea414140de5dae2223b00361a396177a9cb410ff61f20015ad"
        );
        assert_eq!(
            hex(&sha256(
                b"abcdbcdecdefdefgefghfghighijhijkijkljklmklmnlmnomnopnopq"
            )),
            "248d6a61d20638b8e5c026930c3e6039a33ce45964ff2167f6ecedd419db06c1"
        );
    }

    #[test]
    fn multi_block_and_length_boundaries() {
        // 64 bytes: exactly one block of input forces a second padding block.
        let a64 = vec![b'a'; 64];
        assert_eq!(
            hex(&sha256(&a64)),
            "ffe054fe7ae0cb6dc65c3af9b61d5209f439851db43d0ba5997337df154668eb"
        );
        // 1,000,000 'a' — the classic long vector.
        let million = vec![b'a'; 1_000_000];
        assert_eq!(
            hex(&sha256(&million)),
            "cdc76e5c9914fb9281a1c7e284d73e67f1809a48a497200e046d39ccc7112cd0"
        );
    }

    #[test]
    fn crc32_known_vectors() {
        // The canonical IEEE CRC-32 check values every implementation agrees on.
        assert_eq!(crc32(b""), 0x0000_0000);
        assert_eq!(crc32(b"a"), 0xE8B7_BE43);
        assert_eq!(crc32(b"123456789"), 0xCBF4_3926);
        assert_eq!(
            crc32(b"The quick brown fox jumps over the lazy dog"),
            0x414F_A339
        );
    }

    #[test]
    fn crc32_detects_single_bit_flips() {
        let base = [0xA5u8; 40];
        let good = crc32(&base);
        for byte in 0..base.len() {
            for bit in 0..8 {
                let mut torn = base;
                torn[byte] ^= 1 << bit;
                assert_ne!(crc32(&torn), good, "bit {bit} of byte {byte} undetected");
            }
        }
    }
}