How to re-key a hierarchy in a table? - sql

Question

I have a two tables
cars: contains hierarchy data about cars
+-----+-------------+-----------+
| id | description | parent_id |
+-----+-------------+-----------+
| 1 | All cars | 1 |
| 30 | Toyota | 1 |
| 34 | Yaris | 30 |
| 65 | Yaris | 30 |
| 87 | Avensis | 30 |
| 45 | Avensis | 30 |
| 143 | Skoda | 1 |
| 199 | Octavia | 143 |
| 12 | Yeti | 143 |
+-----+-------------+-----------+
car_mapping: contains mapping data where duplicate cars (with different ids) are mapped to one id.
+--------+----------+--------+
| car_id | car_name | map_id |
+--------+----------+--------+
| 34 | Yaris | 1 |
| 65 | Yaris | 1 |
| 87 | Avensis | 2 |
| 45 | Avensis | 2 |
| 199 | Octavia | 3 |
| 12 | Yeti | 4 |
| 30 | Toyota | 5 |
| 143 | Skoda | 6 |
| 1 | All cars | 0 |
+--------+----------+--------+
Now, the idea is to create a third table, cars_new, based on cars and car_mapping which removes duplicates and re-keys the hierarchy in the cars table based on the map_id field in the car_mapping table. Here is the resulting cars_new:
+--------+----------+---------------+
| map_id | car_name | parent_map_id |
+--------+----------+---------------+
| 0 | All | 0 |
| 1 | Yaris | 5 |
| 2 | Avensis | 5 |
| 3 | Octavia | 6 |
| 4 | Yeti | 6 |
| 5 | Toyota | 0 |
| 6 | Skoda | 0 |
+--------+----------+---------------+
Here is the SQL Fiddle for this question. Any ideas how to re-key this hiearchy?

Answer 1

select distinct cm.map_id, cm.car_name, cm2.map_id parent_map_id
from cars c, car_mapping cm, car_mapping cm2
where c.id = cm.car_id
and c.parent_id = cm2.car_id(+)
order by cm.map_id;
PS: in your car_mapping table, you need one extra line (first one below) to get exactly the result you want:
+--------+----------+--------+
| car_id | car_name | map_id |
+--------+----------+--------+
| 1 | All | 0 |
| 34 | Yaris | 1 |
| 65 | Yaris | 1 |
Etc..

Answer 2

Based on #Majid LAISSI's accepted answer, this seems to work both in Oracle and SQL Server:
select distinct cm.map_id, cm.car_name, cm2.map_id as parent_map_id
from cars c
left outer join car_mapping cm on c.id = cm.car_id
left outer join car_mapping cm2 on c.parent_id = cm2.car_id
order by cm.map_id;

Answer 3

You don't have a hierarchy, and you're better off not creating one. Observe that your "cars" table doesn't describe cars; it merely assigns a string to a number (and another number to that number). Right from the get-go, "all cars" isn't a car, and "Toyota" is a car manufacturer, not a car.
The solution -- which would help with your uniqueness issue and simplify your queries -- is to use one table for each distinct thing:
manufactures { mfg_id, name } -- e.g. GM, Ford
makes { make_id, name, mfg_id } -- e.g. Chevrolet, Lincoln; links to manufactures
models { name, make_id } -- e.g. Yaris, etc.; links to makes.
Be sure to make "name" unique in each of the tables to prevent spurious IDs from being created.
This will let you assign new attributes to these things as they arise, such as the years they were made or how many were sold, or how many doors each model comes in. It will also let you prevent "relations" of Ford to GM or, say, making Yaris the parent of "all cars".
(BTW, I suggest you eschew "map" or "mapping" in a table name, because it doesn't say anything. Every table relates the elements in the row to each other. Every table maps the key to its values. The good news is that your car_mapping table disappears in the new design.)
As for how to convert the existing cars table, it will be a nuisance. Assuming cars_mapping is right, you'll be able to insert into each table, joining to it and taking the min(id) while grouping by name. You'll need three such queries, followed by some careful eyeballing to check for, er, misalignment.