Trait KevyHash

Source

pub trait KevyHash {
    // Required method
    fn kevy_hash(&self) -> u64;
}

Expand description

Single-call hashing for kevy’s per-command hot path.

std::hash::Hasher is a state-machine API — every hash is Hasher::default() → write_* → finish, with BuildHasher indirection on top. For kevy-map’s open-addressing table the keyspace is a small handful of well-known leaf types ([u8], u32, u64, i32); we get a faster, inline- friendly hash by exposing one method on each that produces the final mixed 64-bit value in one go.

All impls must agree with feeding the value through FxHasher then calling finish — this lets us cut the trait dispatch without changing the hash function. kevy-map consumes both the full hash (for bucket index) and its top 7 bits (for the metadata byte).

Required Methods§

Source

fn kevy_hash(&self) -> u64

Compute the final mixed 64-bit hash of self in one call.

Dyn Compatibility§

This trait is dyn compatible.

In older versions of Rust, dyn compatibility was called "object safety".

Implementations on Foreign Types§

Source §

fn kevy_hash(&self) -> u64

Byte-slice hash. Uses the two-stream pipelined path ([hash_bytes_pipelined]) for ILP on the bench’s 8-64 byte keyspace, closing the prior 1 ns gap vs rustc-hash 2.x’s hash_bytes. The final fmix64 retains the anti-clustering guarantee that the no_catastrophic_clustering_on_low_entropy_keys test enforces.

Note: the result diverges from the legacy FxHasher absorb path — callers using FxHashMap<Vec<u8>, _> route through std’s Hash::hash → Hasher::write → finish (the legacy single-stream path), which intentionally stays put for cross-instance hash stability with anything that depended on the v0.polish bit pattern. The KevyHash for [u8] impl is for one-call hot paths like kevy-map::find_by_borrow, which is the only one we measure.

Source §