Merge hstore data from multiple rows - postgresql

Question

Imagine I have a table with this definition:
CREATE TABLE test (
values HSTORE NOT NULL
);
Imagine I insert a few records and end up with the following:
values
-----------------------------
"a"=>"string1","b"=>"string2"
"b"=>"string2","c"=>"string3"
Is there any way I can make an aggregate query that will give me a new hstore with the merged keys (and values) for all rows.
Pseudo-query:
SELECT hstore_sum(values) AS value_sum FROM test;
Desired result:
value_sum
--------------------------------------------
"a"=>"string1","b"=>"string2","c"=>"string3"
I am aware of potential conflicts with different values for each key, but in my case the order / priority of which value is picked is not important (it does not even have to be deterministic, as they will be the same for the same key).
Is this possible out of the box or do you have to use some specific homemade SQL functions or other to do it?

Answer 1

You can do a lot of things, f.ex:
My first thought was to use the each() function, and aggregate keys and values separately, like:
SELECT hstore(array_agg(key), array_agg(value))
FROM test,
LATERAL each(hs);
But this performs the worst.
You can use the hstore_to_array() function too, to build a key-value altering array, like (#JakubKania):
SELECT hstore(array_agg(altering_pairs))
FROM test,
LATERAL unnest(hstore_to_array(hs)) altering_pairs;
But this isn't perfect yet.
You can rely the hstore values' representation, and build up a string, which will contain all your pairs:
SELECT hstore(string_agg(nullif(hs::text, ''), ','))
FROM test;
This is quite fast. However, if you want, you can use a custom aggregate function (which can use the built-in hstore concatenation):
CREATE AGGREGATE hstore_sum (hstore) (
SFUNC = hs_concat(hstore, hstore),
STYPE = hstore
);
-- i used the internal function (hs_concat) for the concat (||) operator,
-- if you do not want to rely on this function,
-- you could easily write an equivalent in a custom SQL function
SELECT hstore_sum(hs)
FROM test;
SQLFiddle

Answer 2

There is no in built function for that but hstore offers a few functions that allow to transform it to something else, for example an array. So we cast it to array, merge the arrays and create hstore from final array:
SELECT hstore(array_agg(x)) FROM
(SELECT unnest(hstore_to_array(hs)) AS x
FROM test)
as q;
http://sqlfiddle.com/#!15/cb11a/1
P.S. Some other combination (like going with JSON) might be more efficient.

Answer 3

This is what I wrote which works in production now. I avoid excessive conversion between types e.g. hstore and array. I also don't use hs_concat as sfunc directly as it will pruduce NULL if any of the hashes it is aggregating is NULL.
CREATE OR REPLACE FUNCTION public.agg_hstore_sum_sfunc(state hstore, val hstore)
RETURNS hstore AS $$
BEGIN
IF val IS NOT NULL THEN
IF state IS NULL THEN
state := val;
ELSE
state := state || val;
END IF;
END IF;
RETURN state;
END;
$$ LANGUAGE 'plpgsql';
CREATE AGGREGATE public.sum(hstore) (
SFUNC = public.agg_hstore_sum_sfunc,
STYPE = hstore
);