Jaro-Winkler returns unexpected values for two nil inputs

[mhuerster]

Is this intended behavior?

user=> (require '[clj-fuzzy.jaro-winkler])
nil
user=> (clj-fuzzy.jaro-winkler/jaro-winkler nil nil)
0.4

[Yomguithereal]

Hello @mhuerster. I guess it's not. Do you think it should throw or do you think it should return 1? Broadly speaking, should nil be considered as an empty string or should we refuse to take it as a valid argument?

[mhuerster]

Thanks for responding, @Yomguithereal. I think either of those is reasonable. As a user, I'd expect consistency with the other functions exposed in clj-fuzzy.metrics, most (though not all) of which seem to treat nil as the empty string:

(use '[clj-fuzzy.metrics])
(levenshtein nil nil) ; => 0
(levenshtein "" "")   ; => 0

(dice nil nil) ; => 1.0
(dice "" "")   ; => 1.0

(sorensen nil nil) ; => 1.0
(sorensen "" "")   ; => 1.0

(mra-comparison nil nil) ; => NullPointerException
(mra-comparison "" "")   ; => {:minimum 5, :similarity 6, :codex ["" ""], :match true}

(jaccard nil nil) ; => ArithmeticException Divide by zero
(jaccard "" "")   ; => ArithmeticException Divide by zero

(hamming nil nil)  ; => 0
(hamming "" "")    ; => 0

(jaro nil nil) ; => 0
(jaro "" "")   ; => 0

(tversky nil nil) ; => ArithmeticException Divide by zero
(tversky "" "")   ; => ArithmeticException Divide by zero

[Yomguithereal]

Consistency would be good indeed. I don't have much time right now to dedicate to this issue but I'll gladly accept a pull request.

[mhuerster]

Great, thanks! I'll take a stab at it if I have some free time this/next week.