Consider Spaces in pg_trgm for Better Similarity

Is there a way to consider white space in tri-grams?  That would allow
for better matches of phrases.

For example, currently "one two three" and "three two one" would
generate the same tri-grams ({  o,  t, on, th, tw,ee ,hre,ne
,one,ree,thr,two,wo }), and the distance of "one two four" will be the
same for both of them.  The query:

SELECT   phrase
        ,input
        ,similarity(t1.phrase, t2.input)
        ,word_similarity(t1.phrase, t2.input)
FROM      (values('one two three'),('three two one')) t1(phrase)
        ,(values('one two four')) t2(input);

Returns:

phrase        |input        |similarity  |word_similarity |
--------------|-------------|------------|----------------|
one two three |one two four |0.444444448 |0.615384638 |
three two one |one two four |0.444444448 |0.615384638 |

But surely "one two four" is more similar to "one two three" than to
"three two one".

Any thoughts?

Igal Sapir
Lucee Core Developer
Lucee.org <http://lucee.org/>