Re: tgrm index for word_similarity

On Sat, Oct 21, 2017 at 10:01:56PM -0700, Igal @ Lucee.org wrote:
>
> 1) I thought that the whole idea behind indexes on expressions is that the
> index would be used in a WHERE clause?  See
> https://www.postgresql.org/docs/10/static/indexes-expressional.html - Am I
> missing something?
>

I think the idea is a little bit different. It is about computing index
entries only once, during index creation. During scan PostgreSQL doesn't
compute such entries every time.
I am not very good at PostgreSQL's planner. But I know that PostgreSQL
uses index scan for pg_trgm only with %, <%, ~~, ~~*, ~, ~* operators.
pg_trgm's operator classes (which should be implemented for index scan) are designed in this way.

> 2) A query with `WHERE input <% name` utilizes the index, but a query
> without a WHERE clause at all does not?

Because sequential scan is cheaper here than index scan.

>
> 3) What happens if I do not create an index at all?  Does the query that I
> run in 30 - 40ms, the one that does not utilize an index, creates all of the
> tri-grams on the fly each time that it runs?  Would it be possible for me to
> create a TABLE or a VIEW with the tri-grams so that there is no need to
> create them each time the query runs?
>

As far as I know you can't do it nowadays. You can't create an trigram
column, as you can do it for FTS, you can create an tsvector column.

--
Arthur Zakirov
Postgres Professional: http://www.postgrespro.com
Russian Postgres Company