Crate fuzzies 

Fuzzies

Fuzzy search crate for Rust.

More information about this crate can be found in the crate documentation

---

Installation

cargo add fuzzies

---

Example

This library allows you to build a compact, memory-mapped FST from a file and perform fast, fuzzy searches with configurable Levenshtein distances.

use fuzzies::Dictionary;

fn main() -> Result<(), Box<dyn std::error::Error>> {
    // Prepare your raw text file (must be sorted lexicographically)
    // Fuzzies provides a handy in-place sorter for convenience:
    Dictionary::sort("words.txt")?;

    // Build the immutable binary FST from the sorted text file
    Dictionary::build("words.txt", "words.fst")?;

    // Load the dictionary (memory-mapped from disk)
    let dict = Dictionary::open("words.fst")?;

    // Check for exact matches instantly
    if dict.contains("banana") {
        println!("Exact match found!");
    }

    // Perform a fuzzy search with a max typo distance of 2 and limit of 5 results
    let results = dict.search("banaan")
        .distance(2)
        .transposition(true) // Handles adjacent swaps (e.g., "teh" -> "the")
        .prefix(false)       // Set to true for prefix fuzzy lookups
        .limit(5)
        .execute()?;
    
    for result in results {
        println!("Found: {} (Distance: {}, Exact: {})", result.key, result.distance, result.is_exact);
    }

    // Batch search (multithreaded, defaults to a distance of 1)
    let queries = vec!["aple", "baxana", "cherri"];
    let batch_results = dict.batch_search(&queries);

    for (query, result) in queries.iter().zip(batch_results) {
        match result {
            Ok(matches) => println!("Query '{}' found {} matches", query, matches.len()),
            Err(e) => eprintln!("Error searching for '{}': {}", query, e),
        }
    }

    Ok(())
}

If you prefer shipping a single executable without relying on an external .fst file on disk, bake the dataset directly into your application:

static DICT_DATA: &[u8] = include_bytes!("../assets/words.fst");
let dict = Dictionary::from_embedded(DICT_DATA)?;

---

Performance

The following benchmarks were gathered using Criterion to evaluate lookup speeds for single and parallel batch searches. You can re-run these benchmarks on your hardware using cargo bench.

Single Search

Dictionary Single Search/apple          6.8904 µs/iter (+/- 0.0174 µs)
Dictionary Single Search/baxana         8.1007 µs/iter (+/- 0.0321 µs)
Dictionary Single Search/missingword   12.1830 µs/iter (+/- 0.0285 µs)

Batch Search

Rayon Parallel Batch/100 queries      406.79 µs/iter (+/- 1.60 µs)
Rayon Parallel Batch/500 queries     1.9530 ms/iter (+/- 0.0051 ms)
Rayon Parallel Batch/1000 queries    3.9583 ms/iter (+/- 0.0141 ms)

[!NOTE] Benchmarks were executed on an Intel Core i5-10300H (4 cores, 8 threads, Battery set to High Performance mode). Performance may scale significantly higher on more modern or high-end CPUs.

---

License

This project is licensed under the MIT license.

Structs

Dictionary

Memory-mapped FST dictionary for fuzzy string lookups.

FstDfaWrapper

Adapts a Levenshtein DFA to the fst::Automaton trait ecosystem.

SearchBuilder

Query builder for configuring fuzzy searches.

SearchResult

A matched item from a fuzzy search.

Enums

DictionaryError

DictionarySource

Represents the underlying storage strategy for the dictionary data.