How to update jsonb string with PostgreSQL? - json

Question

I'm using PostgreSQL 9.4.5. I'd like to update a jsonb column.
My table is structured this way:
CREATE TABLE my_table (
gid serial PRIMARY KEY,
"data" jsonb
);
JSON strings are like this:
{"files": [], "ident": {"id": 1, "country": null, "type ": "20"}}
The following SQL doesn't do the job (syntax error - SQL state = 42601):
UPDATE my_table SET "data" -> 'ident' -> 'country' = 'Belgium';
Is there a way to achieve that?

Answer 1

Ok there are two functions:
create or replace function set_jsonb_value(p_j jsonb, p_key text, p_value jsonb) returns jsonb as $$
select jsonb_object_agg(t.key, t.value) from (
select
key,
case
when jsonb_typeof(value) = 'object' then set_jsonb_value(value, p_key, p_value)
when key = p_key then p_value
else value
end as value from jsonb_each(p_j)) as t;
$$ language sql immutable;
First one just changes the value of the existing key regardless of the key path:
postgres=# select set_jsonb_value(
'{"files": [], "country": null, "ident": {"id": 1, "country": null, "type ": "20"}}',
'country',
'"foo"');
set_jsonb_value
--------------------------------------------------------------------------------------
{"files": [], "ident": {"id": 1, "type ": "20", "country": "foo"}, "country": "foo"}
(1 row)
create or replace function set_jsonb_value(p_j jsonb, p_path text[], p_value jsonb) returns jsonb as $$
select jsonb_object_agg(t.key, t.value) from (
select
key,
case
when jsonb_typeof(value) = 'object' then set_jsonb_value(value, p_path[2:1000], p_value)
when key = p_path[1] then p_value
else value
end as value from jsonb_each(p_j)
union all
select
p_path[1],
case
when array_length(p_path,1) = 1 then p_value
else set_jsonb_value('{}', p_path[2:1000], p_value) end
where not p_j ? p_path[1]) as t;
$$ language sql immutable;
Second one changes the value of the existing key using the path specified or creates it if the path does not exists:
postgres=# select set_jsonb_value(
'{"files": [], "country": null, "ident": {"id": 1, "type ": "20"}}',
'{ident,country}'::text[],
'"foo"');
set_jsonb_value
-------------------------------------------------------------------------------------
{"files": [], "ident": {"id": 1, "type ": "20", "country": "foo"}, "country": null}
(1 row)
postgres=# select set_jsonb_value(
'{"files": [], "country": null, "ident": {"id": 1, "type ": "20"}}',
'{ident,foo,bar,country}'::text[],
'"foo"');
set_jsonb_value
-------------------------------------------------------------------------------------------------------
{"files": [], "ident": {"id": 1, "foo": {"bar": {"country": "foo"}}, "type ": "20"}, "country": null}
(1 row)
Hope it will help to someone who uses the PostgreSQL < 9.5
Disclaimer: Tested on PostgreSQL 9.5

Answer 2

In PG 9.4 you are out of luck with "easy" solutions like jsonb_set() (9.5). Your only option is to unpack the JSON object, make the changes and re-build the object. That sounds very cumbersome and it is indeed: JSON is horrible to manipulate, no matter how advanced or elaborate the built-in functions.
CREATE TYPE data_ident AS (id integer, country text, "type" integer);
UPDATE my_table
SET "data" = json_build_object('files', "data"->'files', 'ident', ident.j)::jsonb
FROM (
SELECT gid, json_build_object('id', j.id, 'country', 'Belgium', 'type', j."type") AS j
FROM my_table
JOIN LATERAL jsonb_populate_record(null::data_ident, "data"->'ident') j ON true) ident
WHERE my_table.gid = ident.gid;
In the SELECT clause "data"->'ident' is unpacked into a record (for which you need to CREATE TYPE a structure). Then it is built right back into a JSON object with the new country name. In the UPDATE that "ident" object is re-joined with the "files" object and the whole thing cast to a jsonb.
A pure thing of beauty -- just so long as speed is not your thing...

Answer 3

My previous solution relied on 9.5 functionality.
I would recommend instead either going with abelisto's solutions below or using pl/perlu, plpythonu, or plv8js to write json mutators in a language that has better support for them.