String distance and similarity metrics for Rust.

String distance and similarity — seven public metrics across four algorithms.

use verbora_distance::{levenshtein, jaro_winkler, dice_coefficient, hamming};

assert_eq!(levenshtein("kitten", "sitting", &Default::default()), 3.0);
assert_eq!(dice_coefficient("abc", "abc"), 1.0);
assert_eq!(hamming("karolin", "kathrin", false), 3);
assert_eq!(jaro_winkler("abc", "abc", &Default::default()), 1.0);

Conventions differ between metrics

The metrics deliberately do not share a single direction convention, and this crate does not "fix" that, since doing so would change every caller's results:

The [verbora_core::StringMetric] implementations below record which direction each one uses, so generic code can adapt without any metric changing its output.

Unicode

Every metric here indexes text by UTF-16 code unit — because that is observable in the results. See [units] for the mechanism and for the ASCII fast path that keeps it free on ordinary input.

Batch computation (feature = parallel)

Every metric above is a pure, stateless free function, so scoring many independent pairs is embarrassingly parallel with zero coordination cost. With the parallel feature enabled, [par_levenshtein_batch], [par_damerau_levenshtein_batch], [par_jaro_winkler_batch], [par_dice_coefficient_batch] and [par_hamming_batch] fan a batch of pairs out across a rayon thread pool. Each is exactly pairs.par_iter().map(<the sequential function>).collect() — see the individual function docs for cost trade-offs and when a plain sequential loop is the better choice (usually: for small batches or short strings).