11 releases

0.1.10 Oct 5, 2023
0.1.9 Oct 5, 2023
0.1.8 Jun 6, 2022
0.1.7 Sep 2, 2021
0.1.1 Jun 6, 2020

#103 in Data structures

Download history 17114/week @ 2024-01-07 20473/week @ 2024-01-14 19674/week @ 2024-01-21 21631/week @ 2024-01-28 21920/week @ 2024-02-04 25260/week @ 2024-02-11 27012/week @ 2024-02-18 34737/week @ 2024-02-25 32097/week @ 2024-03-03 29986/week @ 2024-03-10 30998/week @ 2024-03-17 32320/week @ 2024-03-24 31483/week @ 2024-03-31 34647/week @ 2024-04-07 38372/week @ 2024-04-14 40486/week @ 2024-04-21

146,700 downloads per month
Used in 70 crates (2 directly)

Apache-2.0

175KB
857 lines

esaxx-rs

This code implements a fast suffix tree / suffix array.

This code is taken from sentencepiece and to be used by hugging face.

Small wrapper around sentencepiece's esaxx suffix array C++ library. Usage

let string = "abracadabra";
let suffix = esaxx_rs::suffix(string).unwrap();
let chars: Vec<_> = string.chars().collect();
let mut iter = suffix.iter();
assert_eq!(iter.next().unwrap(), (&chars[..4], 2)); // abra
assert_eq!(iter.next(), Some((&chars[..1], 5))); // a
assert_eq!(iter.next(), Some((&chars[1..4], 2))); // bra
assert_eq!(iter.next(), Some((&chars[2..4], 2))); // ra
assert_eq!(iter.next(), Some((&chars[..0], 11))); // ''
assert_eq!(iter.next(), None);

lib.rs:

Small wrapper around sentencepiece's esaxx suffix array C++ library. Usage

#[cfg(feature="cpp")]
{
let string = "abracadabra";
let suffix = esaxx_rs::suffix(string).unwrap();
let chars: Vec<_> = string.chars().collect();
let mut iter = suffix.iter();
assert_eq!(iter.next().unwrap(), (&chars[..4], 2)); // abra
assert_eq!(iter.next(), Some((&chars[..1], 5))); // a
assert_eq!(iter.next(), Some((&chars[1..4], 2))); // bra
assert_eq!(iter.next(), Some((&chars[2..4], 2))); // ra
assert_eq!(iter.next(), Some((&chars[..0], 11))); // ''
assert_eq!(iter.next(), None);
}

The previous version uses unsafe optimized c++ code. There exists another implementation a bit slower (~2x slower) that uses safe rust. It's a bit slower because it uses usize (mostly 64bit) instead of i32 (32bit). But it does seems to fix a few OOB issues in the cpp version (which never seemed to cause real problems in tests but still.)

let string = "abracadabra";
let suffix = esaxx_rs::suffix_rs(string).unwrap();
let chars: Vec<_> = string.chars().collect();
let mut iter = suffix.iter();
assert_eq!(iter.next().unwrap(), (&chars[..4], 2)); // abra
assert_eq!(iter.next(), Some((&chars[..1], 5))); // a
assert_eq!(iter.next(), Some((&chars[1..4], 2))); // bra
assert_eq!(iter.next(), Some((&chars[2..4], 2))); // ra
assert_eq!(iter.next(), Some((&chars[..0], 11))); // ''
assert_eq!(iter.next(), None);

No runtime deps