On Google BigQuery (using #standardSQL), when there is a Join between 2 tables, I need to apply a fixed prefix to all the columns of each table.
Here is the scenario, I have a structure like this:
#standardSQL
WITH user AS (
SELECT "john" as name, "smith" as surname, 1 as parent
UNION ALL
SELECT "maggie" as name, "smith" as surname, 2 as parent
),
parent AS (
SELECT 1 as id, "john" as name, "doe" as surname
UNION ALL
SELECT 2 as id, "jane" as name, "smith" as surname
)
User table
+-----+--------+---------+--------+
| Row | name | surname | parent |
+-----+--------+---------+--------+
| 1 | john | smith | 1 |
| 2 | maggie | smith | 2 |
+-----+--------+---------+--------+
Parent table
+-----+----+------+---------+
| Row | id | name | surname |
+-----+----+------+---------+
| 1 | 1 | john | doe |
| 2 | 2 | jane | smith |
+-----+----+------+---------+
A query like this
SELECT u.*, p.* FROM user u JOIN parent p ON u.parent = p.id
produces the following error
Error: Duplicate column names in the result are not supported. Found duplicate(s): name, surname
I want to avoid performing a custom aliasing of the table like this
SELECT
u.name as user_name,
u.surname as user_surname,
p.name as parent_name,
p.surname as parent_surname
FROM user u JOIN parent p ON u.parent = p.id
+-----+-----------+--------------+-------------+----------------+
| Row | user_name | user_surname | parent_name | parent_surname |
+-----+-----------+--------------+-------------+----------------+
| 1 | john | smith | john | doe |
| 2 | maggie | smith | jane | smith |
+-----+-----------+--------------+-------------+----------------+
If the table will change on fields, I'll need every time to edit the statement (or the statements) in order to apply the new fields with the given prefix. So this approach using fixed column names is not a suitable way
Is there a way, a query operator, in order to obtain the table as mentioned up there, automatic applying a prefix? Something like:
SELECT u.* AS user_*, p.* AS parent_*
FROM user u JOIN parent p ON u.parent = p.id
The only option I can think of so far is as below
#standardSQL
WITH user AS (
SELECT "john" AS name, "smith" AS surname, 1 AS parent UNION ALL
SELECT "maggie" AS name, "smith" AS surname, 2 AS parent
), parent AS (
SELECT 1 AS id, "john" AS name, "doe" AS surname UNION ALL
SELECT 2 AS id, "jane" AS name, "smith" AS surname
)
SELECT user, parent
FROM user
JOIN parent
ON user.parent = parent.id
with result as
Row user.name user.surname user.parent parent.id parent.name parent.surname
1 john smith 1 1 john doe
2 maggie smith 2 2 jane smith
It is not exactly what you expect, but the closest to it, as it wraps each row from respective joined tables into respective STRUCTs - for example:
{
"user": {"name": "john", "surname": "smith","parent": "1"},
"parent": {"id": "1","name": "john","surname": "doe"}
}