hashbrown

hashbrown is a Rust implementation of Google’s SwissTable hash map design, built as a drop-in replacement for the standard library’s HashMap and HashSet. Since Rust 1.36 the standard library itself has adopted this implementation internally, but the standalone crate remains essential for #[no_std] environments — embedded systems and kernels — where the standard library isn’t available.

Compared to a naive hash table, hashbrown uses SIMD-accelerated lookups to scan multiple entries in parallel, only one byte of per-entry overhead instead of eight, and allocates no memory for empty maps, delivering roughly 2x the throughput of the pre-SwissTable standard library HashMap.

What You Get

• Drop-in HashMap and HashSet types API-compatible with the Rust standard library’s collections
• SIMD-accelerated lookups that scan multiple hash table slots in parallel
• Only one byte of per-entry metadata overhead versus eight bytes in naive designs
• #[no_std] compatibility (with a global allocator) for embedded systems and kernels
• Optional serde, rayon parallel-iterator, and custom-equivalence (Equivalent trait) integrations via Cargo features

Common Use Cases

• Using a fast, low-overhead hash map in #[no_std] embedded or kernel code where std collections aren’t available
• Squeezing extra performance out of hot-path hash map lookups in latency-sensitive Rust services
• Building custom hashing-based data structures on top of hashbrown’s lower-level raw table API
• Serializing/deserializing hash maps with serde support enabled via the serde feature flag

Under The Hood

Architecture - The crate’s core is a raw SIMD-aware hash table (src/raw and table.rs, ~3,200 lines) that implements SwissTable’s group-probing scheme, with HashMap/HashSet built as thin, std-API-compatible wrappers over that raw table; foldhash is used as the fast default hasher, replacing SipHash for non-adversarial workloads. Tech Stack - Nearly pure Rust (99%+), gated behind Cargo feature flags for nightly (unstable optimizations), serde, rayon parallel iterators, raw-entry, and allocator-api2 support, with a CI matrix (ci/ directory, Cross.toml) validating multiple targets and toolchains including no_std configurations. Code Quality - Extensive integration tests (tests/set.rs, tests/serde.rs, tests/rayon.rs, tests/hasher.rs, tests/hasher_unwind.rs, tests/equivalent_trait.rs) plus dedicated benchmark suites (benches/) validate both correctness and performance; the project is one of the most heavily reviewed crates in the Rust ecosystem given its role inside std itself. API Design - The public API deliberately mirrors std::collections::HashMap/HashSet method-for-method, so migrating existing code is close to a drop-in import swap, while advanced users can drop to the raw table API for custom hashing schemes.