Re: ToDo: KNN Search should to support DISTINCT clasuse?

Tom Lane wrote:
> "Kevin Grittner" <kgrittn(at)mail(dot)com> writes:
> > Pavel Stehule wrote:
> >> 2012/10/22 Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us>:
> >>> Perhaps it would be close enough to what you want to use DISTINCT ON:
> >>> contrib_regression=# explain select distinct on( t <-> 'foo') *,t <-> 'foo' from test_trgm order by t <-> 'foo' limit 10;
>
> >> good tip - it's working
>
> > If two or more values happen to be at exactly the same distance,
> > wouldn't you just get one of them?
>
> Yeah, that is a hazard. I'm not sure whether <->'s results are
> sufficiently quantized to make that a big problem in practice.

It doesn't seem too far-fetched for trigram queries:

test=# select nm, nm <-> 'anders' from (values ('anderson'),('andersen'),('andersly')) x(nm);
   nm    | ?column?
----------+----------
anderson |      0.4
andersen |      0.4
andersly |      0.4
(3 rows)

test=# select distinct on (nm <-> 'anders') nm, nm <-> 'anders' from (values ('anderson'),('andersen'),('andersly')) x(nm) order by nm <-> 'anders' limit 3;
   nm    | ?column?
----------+----------
anderson |      0.4
(1 row)

-Kevin