Adjacency List to JSON graph with Postgres - json

Question

I have the following schema for the tags table:
CREATE TABLE tags (
id integer NOT NULL,
name character varying(255) NOT NULL,
parent_id integer
);
I need to build a query to return the following structure (here represented as yaml for readability):
- name: Ciencia
parent_id:
id: 7
children:
- name: Química
parent_id: 7
id: 9
children: []
- name: Biología
parent_id: 7
id: 8
children:
- name: Botánica
parent_id: 8
id: 19
children: []
- name: Etología
parent_id: 8
id: 18
children: []
After some trial and error and looking for similar questions in SO, I've came up with this query:
WITH RECURSIVE tagtree AS (
SELECT tags.name, tags.parent_id, tags.id, json '[]' children
FROM tags
WHERE NOT EXISTS (SELECT 1 FROM tags tt WHERE tt.parent_id = tags.id)
UNION ALL
SELECT (tags).name, (tags).parent_id, (tags).id, array_to_json(array_agg(tagtree)) children FROM (
SELECT tags, tagtree
FROM tagtree
JOIN tags ON tagtree.parent_id = tags.id
) v
GROUP BY v.tags
)
SELECT array_to_json(array_agg(tagtree)) json
FROM tagtree
WHERE parent_id IS NULL
But it returns the following results when converted to yaml:
- name: Ciencia
parent_id:
id: 7
children:
- name: Química
parent_id: 7
id: 9
children: []
- name: Ciencia
parent_id:
id: 7
children:
- name: Biología
parent_id: 7
id: 8
children:
- name: Botánica
parent_id: 8
id: 19
children: []
- name: Etología
parent_id: 8
id: 18
children: []
The root node is duplicated.
I could merge the results to the expected result in my app code but I feel I am close and it could be done al from PG.
Here's an example with SQL Fiddle:
http://sqlfiddle.com/#!15/1846e/1/0
Expected output:
https://gist.github.com/maca/e7002eb10f36fcdbc51b
Actual output:
https://gist.github.com/maca/78e84fb7c05ff23f07f4

Answer 1

Here's a solution using PLV8 for your schema.
First, build a materialized path using PLSQL function and recursive CTEs.
CREATE OR REPLACE FUNCTION get_children(tag_id integer)
RETURNS json AS $$
DECLARE
result json;
BEGIN
SELECT array_to_json(array_agg(row_to_json(t))) INTO result
FROM (
WITH RECURSIVE tree AS (
SELECT id, name, ARRAY[]::INTEGER[] AS ancestors
FROM tags WHERE parent_id IS NULL
UNION ALL
SELECT tags.id, tags.name, tree.ancestors || tags.parent_id
FROM tags, tree
WHERE tags.parent_id = tree.id
) SELECT id, name, ARRAY[]::INTEGER[] AS children FROM tree WHERE $1 = tree.ancestors[array_upper(tree.ancestors,1)]
) t;
RETURN result;
END;
$$ LANGUAGE plpgsql;
Then, build the tree from the output of the above function.
CREATE OR REPLACE FUNCTION get_tree(data json) RETURNS json AS $$
var root = [];
for(var i in data) {
build_tree(data[i]['id'], data[i]['name'], data[i]['children']);
}
function build_tree(id, name, children) {
var exists = getObject(root, id);
if(exists) {
exists['children'] = children;
}
else {
root.push({'id': id, 'name': name, 'children': children});
}
}
function getObject(theObject, id) {
var result = null;
if(theObject instanceof Array) {
for(var i = 0; i < theObject.length; i++) {
result = getObject(theObject[i], id);
if (result) {
break;
}
}
}
else
{
for(var prop in theObject) {
if(prop == 'id') {
if(theObject[prop] === id) {
return theObject;
}
}
if(theObject[prop] instanceof Object || theObject[prop] instanceof Array) {
result = getObject(theObject[prop], id);
if (result) {
break;
}
}
}
}
return result;
}
return JSON.stringify(root);
$$ LANGUAGE plv8 IMMUTABLE STRICT;
This will yield the required JSON mentioned in your question. Hope that helps.
I've written a detailed post/breakdown of how this solution works here.

Answer 2

Try PL/Python and networkx.
Admittedly, using the following doesn't yield JSON in exactly the requested format, but the information seems to be all there and, if PL/Python is acceptable, this might be adapted into a complete answer.
CREATE OR REPLACE FUNCTION get_adjacency_data(
names text[],
ids integer[],
parent_ids integer[])
RETURNS jsonb AS
$BODY$
pairs = zip(ids, parent_ids)
import networkx as nx
import json
from networkx.readwrite import json_graph
name_dict = dict(zip(ids, names))
G=nx.DiGraph()
G.add_nodes_from(ids)
nx.set_node_attributes(G, 'name', name_dict)
G.add_edges_from(pairs)
return json.dumps(json_graph.adjacency_data(G))
$BODY$ LANGUAGE plpythonu;
WITH raw_data AS (
SELECT array_agg(name) AS names,
array_agg(parent_id) AS parent_ids,
array_agg(id) AS ids
FROM tags
WHERE parent_id IS NOT NULL)
SELECT get_adjacency_data(names, parent_ids, ids)
FROM raw_data;

Answer 3

i was finding same solution and may be this example could be useful for anyone
tested on Postgres 10 with table with same structure
table with columns: id, name and pid as parent_id
create or replace function get_c_tree(p_parent int8) returns setof jsonb as $$
select
case
when count(x) > 0 then jsonb_build_object('id', c.id, 'name', c.name, 'children', jsonb_agg(f.x))
else jsonb_build_object('id', c.id, 'name', c.name, 'children', null)
end
from company c left join get_c_tree(c.id) as f(x) on true
where c.pid = p_parent or (p_parent is null and c.pid is null)
group by c.id, c.name;
$$ language sql;
select jsonb_agg(get_c_tree) from get_c_tree(null::int8);