ticki / eudex Goto Github PK

Do I just module table length as usually? Or there is something else to be done?

did you mean pronounciation or pronunciation in the README.md?

How does eudex handle sounds such as ch, sh, th and ng in english?
Or y in front as a consonant vs in the middle as a vowel?
Or c having a k sound or a s sound in certain situations?

FWIW I ported Eudex to Java at https://github.com/jprante/elasticsearch-analysis-phonetic-eudex/blob/master/src/main/java/org/xbib/elasticsearch/index/analysis/phonetic/Eudex.java with tests at https://github.com/jprante/elasticsearch-analysis-phonetic-eudex/blob/master/src/test/java/org/xbib/elasticsearch/index/analysis/phonetic/EudexTests.java and wrapped an Elasticsearch analysis plugin around it for experimenting.

Hi! This looks really cool -- I'm curious if there is already a C or C++ implementation available? Thanks!

line 250 let mut n = 1u8;
Are you sure this must be u8? Because I suspect it prevents words longer than 16 characters to hash properly.

I am having a hard time trying to convert Eudex to JavaScript, because there is no such thing as 64 bit integer. You can sort of "emulate" it with two numbers, which I am currently trying to do, but I am just curious, why 64 bit? Can this be adapted to 32 bit? In hash tables you normally do not need this huge amount of bits anyway.

As there has been confusion on reddit regarding the interpretation of the hash value it might be a good idea to newtype the u64 the hasher returns right now.

The newtype could have methods like distance() and weighted_distance() (or however you'd like to call them) and of course a From/Into impl for conversions to u64.

Edit: and of course similar()

if entry <= b'z' {

I am new to Rust so I may get it wrong, but it feels like you are comparing against ordinal('z'), is this the right thing to do? Because you already entry = (string[b] | 32).wrapping_sub(b'a')

I do not know what wrapping_sub does, I refer to Java port, so it feels you need to compare to 26 instead.