Обсуждение: Full text search randomly not working for short prefixes?

Something funny going on with my full text search.. and I have no idea what.
I have a receiver called "Ana", this is her tsv column:

'38651000000':4 'aceventura@mailinator.com':3B 'ana':1A 'novak':2A

This queries do not find her:
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('a:*'));
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('an:*'));

This does:
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('ana:*'));

Now to an even more interesting part: I have 3 people with last name
"Novak" and one with name "Nov"

This query finds all 4:
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('n:*'));

This finds NONE:
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('no:*'));

This finds all 4 again:
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('nov:*'));

..and this finds all with the last name only:
SELECT * FROM receivers r WHERE r.tsv @@ to_tsquery(unaccent('nova:*'));


These are the TSV columns of last name:
"'38651000000':4 'janez':1A 'janeznovak@mailinator.com':3B 'novak':2A"
"'38651000000':4 'aceventura@mailinator.com':3B 'ana':1A 'novak':2A"
"'38651000000':4 'novak':2A 'tine':1A 'tnovak@mailinator.com':3B"
"'21415000000':4 'alen.nova@gmailer.com':3B 'allan':1A 'novak':2A"

And the first name:
"'38651604724':6 'brez':3A 'list':4A 'nov':1A
'novreceiver101@mailinator.com':5B 'receiv':2A"


What is going on here?

cen <imbacen@gmail.com> writes:
> Something funny going on with my full text search.. and I have no idea what.

The way to debug this sort of thing is generally to look at what tsquery
you're actually getting.  I get

regression=# select to_tsquery(unaccent('a:*'));
NOTICE:  text-search query contains only stop words or doesn't contain lexemes, ignored
 to_tsquery
------------

(1 row)

regression=# select to_tsquery(unaccent('an:*'));
NOTICE:  text-search query contains only stop words or doesn't contain lexemes, ignored
 to_tsquery
------------

(1 row)

regression=# select to_tsquery(unaccent('ana:*'));
 to_tsquery
------------
 'ana':*
(1 row)

Of course, only the last is going to match 'ana'.

So you need to use a text search configuration in which a/an are
not stop words.  Or possibly you could cast the unaccent result
directly to tsquery rather than passing it through to_tsquery(),
though likely that would just have a different set of failure modes
with queries where you do wish stemming would occur.

The problem with "no" seems to be the same.

            regards, tom lane

Thanks, that makes sense. I think I'll go with the cast approach, I
don't really need stemming anywhere.


Tom Lane je 02. 12. 2016 ob 16:33 napisal:
> cen <imbacen@gmail.com> writes:
>> Something funny going on with my full text search.. and I have no idea what.
> The way to debug this sort of thing is generally to look at what tsquery
> you're actually getting.  I get
>
> regression=# select to_tsquery(unaccent('a:*'));
> NOTICE:  text-search query contains only stop words or doesn't contain lexemes, ignored
>   to_tsquery
> ------------
>
> (1 row)
>
> regression=# select to_tsquery(unaccent('an:*'));
> NOTICE:  text-search query contains only stop words or doesn't contain lexemes, ignored
>   to_tsquery
> ------------
>
> (1 row)
>
> regression=# select to_tsquery(unaccent('ana:*'));
>   to_tsquery
> ------------
>   'ana':*
> (1 row)
>
> Of course, only the last is going to match 'ana'.
>
> So you need to use a text search configuration in which a/an are
> not stop words.  Or possibly you could cast the unaccent result
> directly to tsquery rather than passing it through to_tsquery(),
> though likely that would just have a different set of failure modes
> with queries where you do wish stemming would occur.
>
> The problem with "no" seems to be the same.
>
>             regards, tom lane

cen <imbacen@gmail.com> writes:
> Something funny going on with my full text search.. and I have no idea what.

The way to debug this sort of thing is generally to look at what tsquery
you're actually getting.  I get

regression=# select to_tsquery(unaccent('a:*'));
NOTICE:  text-search query contains only stop words or doesn't contain lexemes, ignored
 to_tsquery
------------
 
(1 row)

regression=# select to_tsquery(unaccent('an:*'));
NOTICE:  text-search query contains only stop words or doesn't contain lexemes, ignored
 to_tsquery
------------
 
(1 row)

regression=# select to_tsquery(unaccent('ana:*'));
 to_tsquery
------------
 'ana':*
(1 row)

Of course, only the last is going to match 'ana'.

So you need to use a text search configuration in which a/an are
not stop words.  Or possibly you could cast the unaccent result
directly to tsquery rather than passing it through to_tsquery(),
though likely that would just have a different set of failure modes
with queries where you do wish stemming would occur.

The problem with "no" seems to be the same.