psql one to many join filter

Question

Trying to do something on the lines of:

Table Entities

id |  timestamp
---------------
1  | 2012-06-05 12:00:00

Table Attributes

entity_id | attribute
---------------------
1         |   a
1         |   b

What i'd like to be able to do:

select entity.id, entity.timestamp from entities entity join attributes on entity.id = attribute.id where entity.has.attributes a and b

The result set would look woeful now, but assume it has other columns on the entities table that will make sense. So the result set would be:

id | timestamp
--------------
1  | 2012-06-05 12:00:00

I looked at other posts on filtering one-many join relationships that end up using a count and distinct to solve very specific niche cases ( Filter a one-to-many query by requiring all of many meet criteria ) but I am hoping there's a cleaner, more generic solution to this.

Going to continue working away and I'll post back a solution if I can beat stackoverflow to it :)

Answer 1

This is what I have working so far. As pointed out, it's not pretty and I would daresay not performant, but I am going to look into full text search indices to see what they can offer.

select e.id, e.timestamp from entities e join 
    (select entity_id, string_agg(attribute, ',' order by attribute) as attr from attributes 
     group by entity_id) a 
on e.id = a.entity_id where a.attr like '%a%b%';

But this assumes that you have the order by clause in the aggregation. If you choose to leave that out, this should work:

select e.id, e.timestamp from entities e join 
    (select entity_id, string_agg(attribute, ',') as attr from attributes 
     group by entity_id) a 
on e.id = a.entity_id where a.attr like '%a%' and a.attr like '%b%';

Very brittle, will break if your search attributes have kangaroo-word-like relationships. But, this will have to do for now.