Cast JSON to HSTORE in Postgres 9.3+? - json

Question

I've read the docs and it appears that there's no discernible way to perform an ALTER TABLE ... ALTER COLUMN ... USING statement to directly convert a json type column to an hstore type. There's no function available (that I'm aware of) to perform the cast.
The next best alternative I have is to create a new column of type hstore, copy my JSON data to that new column using some external tool, drop the old json column and rename the new hstore column to the old column's name.
Is there a better way?
What I have so far is:
$ CREATE TABLE blah (unstructured_data JSON);
$ ALTER TABLE blah ALTER COLUMN unstructured_data
TYPE hstore USING CAST(unstructured_data AS hstore);
ERROR: cannot cast type json to hstore

Answer 1

Unfortunately, PostgreSQL doesn't allow all kind of expressions within the USING clause of ALTER TABLE ... SET DATA TYPE ... (f.ex. sub-queries are disallowed).
But, you can write a function to overcome this, you just need to decide what to do with advanced types (in object's values), like arrays & objects. Here is an example, which simply converts them to string:
CREATE OR REPLACE FUNCTION my_json_to_hstore(json)
RETURNS hstore
IMMUTABLE
STRICT
LANGUAGE sql
AS $func$
SELECT hstore(array_agg(key), array_agg(value))
FROM json_each_text($1)
$func$;
After that, you can use this in your ALTER TABLE, like:
ALTER TABLE blah
ALTER COLUMN unstructured_data
SET DATA TYPE hstore USING my_json_to_hstore(unstructured_data);

Answer 2

There is "trap" for repeated keys - allowed by both json and hstore input, but unfortunately resolved differently (!). Consider this example value:
json '{"double_key":"key1","foo":null,"double_key":"key2"}'
In json, 'double_key is effectively 'key2'. The manual:
Because the json type stores an exact copy of the input text, it will
preserve semantically-insignificant white space between tokens, as
well as the order of keys within JSON objects. Also, if a JSON object
within the value contains the same key more than once, all the
key/value pairs are kept. (The processing functions consider the last value as the operative one.)
Bold emphasis mine.
In hstore, however, for the same order of key/value pairs, 'double_key' might effectively be 'key1'. The manual:
Each key in an hstore is unique. If you declare an hstore with
duplicate keys, only one will be stored in the hstore and there is no guarantee as to which will be kept:
Typically, the first instance of a key, but that's an implementation details that might change.
A simple and fast option to always preserve the effective, operative value: cast to jsonb before the conversion. The manual again:
[...] jsonb does not preserve white space, does not preserve
the order of object keys, and does not keep duplicate object keys.
If duplicate keys are specified in the input, only the last value is kept.
Modifying #pozs's conversion function:
CREATE OR REPLACE FUNCTION json2hstore(json)
RETURNS hstore AS
$func$
SELECT hstore(array_agg(key), array_agg(value))
FROM jsonb_each_text($1::jsonb) -- !
$func$ LANGUAGE sql IMMUTABLE STRICT;
Requires Postgres 9.4 or later. Postgres 9.3 has the json type, but not jsonb, yet. A no-op in PL/v8 might be alternative there, like #jpmc mentioned.