Great tip about citext! And it’s cool to use only plSQL in this case, I’m not sure about the speed difference though but might be worth a benchmark. Thanks for sharing!

That’s a really cool idea and maybe it’s possible to use a custom type for this, but it’s something I still need to explore. Here is the doc: http://www.postgresql.org/docs/current/static/sql-createtype.html

You can do the same thing in plSQL, which will probably be faster than plpgsql:

CREATE OR REPLACE FUNCTION array_sort(anyarray) RETURNS TABLE( sorted_array anyarray ) LANGUAGE sql AS $body$
SELECT ARRAY_AGG(val) AS sorted_array
FROM (
SELECT UNNEST( $1 ) AS val
ORDER BY val
) AS sorted_vals
$body$;

Yeah, it really sucks that Postgres doesn’t have the concept of a class. I suspect that normally when you do a separate model then you’d end up with a single tags table in the database and a many-many. That’s going to suck performance-wise.

What we’d really want is the ability to tell Postgres in one place “Hey, here’s how you handle tags” and then you could just instantiate that in whataver table you want.

Cool gem, didn’t know about it. That’s an interesting approach to the problem, thanks for sharing!

For the uniqueness a standard unique index over the foreign key and tag name will suffice, if it can be case sensitive. For the case insensitivity with a tag model I would use the schema_plus gem https://github.com/lomba/schema_plus to allow expressions for the index. With this gem the schema format can be kept ruby.