Make it a rust crate about levenshtein

[maxbachmann]

This would require at least partially porting rapidfuzz to rust. I was working on this a couple of years ago in https://github.com/maxbachmann/rapidfuzz-rs, but this would basically need to be completely rewritten.

The algorithms themselves should be fairly easy to port. However I am completely unfamiliar with rust and the C++ implementation uses a couple of more advanced C++ features:

• conditional compilation using if constexpr to reduce code duplication (Worst case this could be duplicated in rust)
• should work for any numeric iterables not just strings. In C++ this is defined as:

template <typename InputIt1, typename InputIt2>
double ratio(InputIt1 first1, InputIt1 last1, InputIt2 first2, InputIt2 last2, double score_cutoff = 0);

not sure how this would need to be implemented with rust generics to work both for strings and any other iterables of numeric items

This topic already came up a couple of months ago here: rapidfuzz/RapidFuzz#331, but I never heard back from @bcvanmeurs again. While I am absolutely willing to work on the algorithms themselves, I think it would be good to have someone with more rust knowledge help with the API design.

[maxbachmann]

I had a look at this this evening. I will probably work on this, but it will certainly take a while to port.

[G2GreenTea]

It’s fantastic news. I am extremely grateful for your willingness to take on this task. As a beginner, I feel limited in my abilities to contribute, but I am truly appreciative of your efforts. If there’s any way I can assist or support the project, please let me know. Thank you once again for your dedication and expertise in this endeavor!

[maxbachmann]

I assume you are working with relatively large strings.

I started working on the rust rewrite in https://github.com/maxbachmann/rapidfuzz-rs/tree/rewrite. So far I have the normal Levenshtein distance pretty much fully implemented, but there are still quite a few features missing. I will probably implement a decent chunk of them, but make a release on cargo before implementing all features.

Performance of the Levenshtein implementation looks to be on par with the C++ implementation (generates largely the same assembler when compiled with llvm). Here is an initial comparision with strsim which appears to be a widely used library for string similarities in rust:

[G2GreenTea]

Perfect! This is a huge progress. Looking forward to the release on cargo!

[maxbachmann]

I published a first version. It mostly includes the basic metrics. Still missing are:

• most of the fuzz module
• simd implementations for Many x Many comparisons of short strings (the python implementation uses this in process.cdist under the hood)
• editops / opcodes to find edit operations needed for a conversion

You can find it here: https://crates.io/crates/rapidfuzz. Let me know if you run into any issues or have suggestions for improvements for the API.