A QP-Trie implementation for Rust

API documentation

A qp-trie is a fast and compact associative array.

It is similar to a crit-bit trie, with a larger fan-out per internal node to save memory and reduce lookup costs.

It supports the following operations at high speed:

• See whether a key is in the trie and retrieve an optional associated value
• Add a (key, value) pair to the trie
• Remove a key from the trie
• Find all keys matching a given prefix

This implementation uses 4 bits per index and doesn't require keys to be zero-terminated.

Example

use qptrie::Trie;

let mut trie = Trie::new();
trie.insert("key number one", 1);
trie.insert("key number two", 2);

for (k, v) in trie.prefix_iter(&"key").include_prefix() {
     println!("{} => {}", k, v);
}

trie.remove(&"key number one");

let v = trie.get(&"key number two").unwrap();

Benchmarks

~500,000 4-bytes keys accessed in random order (source), using rustc 1.15.0-dev (d9aae6362 2016-12-08):

test test::bench_btreemap_get    ... bench: 112,349,209 ns/iter (+/- 9,450,753)
test test::bench_btreemap_insert ... bench: 115,952,204 ns/iter (+/- 7,066,195)
test test::bench_hashmap_get     ... bench:  52,239,122 ns/iter (+/- 2,225,861)
test test::bench_hashmap_insert  ... bench:  60,889,965 ns/iter (+/- 27,314,557)
test test::bench_qptrie_get      ... bench:  51,843,861 ns/iter (+/- 18,878,702)
test test::bench_qptrie_insert   ... bench:  67,449,566 ns/iter (+/- 16,887,173)

qp-tries are more than twice as fast as Rust's BTreeMap, and roughly as fast as Rust's excellent HashMap implementation while being more compact and allowing range queries.