Thread: BUG #18654: From fuzzystrmatch, levenshtein function with costs parameters produce incorrect results

The following bug has been logged on the website:

Bug reference:      18654
Logged by:          bjdev
Email address:      bjdev.gthb@laposte.net
PostgreSQL version: 15.4
Operating system:   Ubuntu 22.04.5 LTS
Description:

Hi,

The extension fuzzystrmatch propose an implementation of levenshtein
function.
There is one version with costs parameters
levenshtein(text source, text target, int ins_cost, int del_cost, int
sub_cost) returns int

But if we use this function with parameters other than 1 (the default) the
result is incorrect

SELECT levenshtein('horses','shorse',1,1,1) => 2 (correct)

SELECT levenshtein('horses','shorse',100,10,1) => 101 (INCORRECT)
The correct result is 6 (all the letter have to be substitute and it's not
possible to have a lower score with others operations)
Here, it's easy to verify manually but you can check that using python
implementation 
from Levenshtein import distance
distance("horses","shorse",weights=(100,10,1))
# => 6

SELECT levenshtein('horses','shorse',1,10,100) => 12 (INCORRECT)
The correct result is 11 (insert "s" first (+1) and remove last "s"(+10)
Here, it's easy to verify manually but you can check that using python
implementation 
from Levenshtein import distance
distance("horses","shorse",weights=(1,10,100))
# => 11

SELECT levenshtein('horses','shorse',1,10,1) => 2 (INCORRECT)
The correct result is 6
you can check that using python implementation 
from Levenshtein import distance
distance("horses","shorse",weights=(1,10,1))
# => 6

The use of cost parameters of the levenshtein function is therefore not
possible, which is a shame.

Regards

PG Bug reporting form <noreply@postgresql.org> writes:
> But if we use this function with parameters other than 1 (the default) the
> result is incorrect

> SELECT levenshtein('horses','shorse',1,1,1) => 2 (correct)

> SELECT levenshtein('horses','shorse',100,10,1) => 101 (INCORRECT)
> The correct result is 6 (all the letter have to be substitute and it's not
> possible to have a lower score with others operations)

Hmm, works for me:

u8=# create extension fuzzystrmatch;
CREATE EXTENSION
u8=# SELECT levenshtein('horses','shorse',1,1,1);
 levenshtein 
-------------
           2
(1 row)

u8=# SELECT levenshtein('horses','shorse',100,10,1);
 levenshtein 
-------------
           6
(1 row)

I confess bafflement about why you're getting wrong answers.
You seem to be using a slightly out of date Postgres, but
none of this code has changed meaningfully since about 2016.
Maybe you hit a compiler bug?  Where did you get this copy
of Postgres from --- or if you built it yourself, what build
options did you use?

            regards, tom lane