cbaatz / damerau-levenshtein Goto Github PK

This implementation seems to be wrong.
Althought I've not dive deep enough in the code to spot where is the bug, I've isolated a simple test case.

DL("beak", "water") returns 5 instead of 4

step 1
beak water => substitution (b by w cost = 1)
step 2
weak water => permutation (e and a cost = 1)
step 3
waek water => deletion (t cost = 1)
step 4
waek waer => substitution (r by k cost = 1)

result
waek = waek => in 4 operations of cost 1

This bug cause the algorithm to lose it's metric properties, which can have a lot of implications. If you use it to build indexes it will certainly screw up your search engine. If you use it for a simple string compare, it should be good enough (even if not a real DamerauLevenshtein).