The project I'm working on is an application that lets you design data entry forms, and automagically generates a schema in an underlying PostgreSQL database to persist them as well as the browsing and editing UI.
The use case I've encountered this with is a store back-office database, but the app itself intends to be somewhat universal. The administrator creates the following entry forms with the given fields:
- Customers
- name (text box)
- Items
- name (text box)
- stock (number field)
- Order
- customer (combo box selecting a customer)
- order lines (a grid showing order lines)
- OrderLine
- item (combo box selecting an item)
- count (number field)
When all this is done, the resulting database schema will be equivalent to this:
create table Customers(id serial primary key,
name varchar);
create table Items(id serial primary key,
name varchar,
stock integer);
create table Orders(id serial primary key);
create table OrderLines(id serial primary key,
count integer);
create table Links(id serial primary key,
fk1 integer references Customers.id,
fk2 integer references Items.id,
fk3 integer references Orders.id,
fk4 integer references OrderLines.id);
Links being a special table that stores all the relationships between entities; every row has (usually) two of the foreign keys set to a value, and the rest set to NULL. Whenever a new entry form is added to the application instance, a new foreign key referencing the table for this form is added to Links.
So, suppose our shop stocks some widgets, gizmos, and thingeys. A customer named Adam orders two widgets and three gizmos, and Betty orders four gizmos and five thingeys. The database will contain the following data:
Customers
/----+-------\
| ID | NAME |
| 1 | Adam |
| 2 | Betty |
\----+-------/
Items
/----+---------+-------\
| ID | NAME | STOCK |
| 1 | widget | 123 |
| 2 | gizmo | 456 |
| 3 | thingey | 789 |
\----+---------+-------/
Orders
/----\
| ID |
| 1 |
| 2 |
\----/
OrderLines
/----+-------\
| ID | COUNT |
| 1 | 2 |
| 2 | 3 |
| 3 | 4 |
| 4 | 5 |
\----+-------/
Links
/----+------+------+------+------\
| ID | FK1 | FK2 | FK3 | FK4 |
| 1 | 1 | NULL | 1 | NULL |
| 2 | 2 | NULL | 2 | NULL |
| 3 | NULL | NULL | 1 | 1 |
| 4 | NULL | NULL | 1 | 2 |
| 5 | NULL | NULL | 2 | 3 |
| 6 | NULL | NULL | 2 | 4 |
| 7 | NULL | 1 | NULL | 1 |
| 8 | NULL | 2 | NULL | 2 |
| 9 | NULL | 2 | NULL | 3 |
| 10 | NULL | 3 | NULL | 4 |
\----+------+------+------+------/
(The tables also contain a bunch of timestamps for auditing and soft deletion but I don't think they're relevant here, they just make writing the SQL by the administrator that much messier. The management app is also used to implement a bunch of different use cases, but they're generally primarily data entry, master-detail views, and either scalar fields or selection boxes.)
When I've had to write a join through this thing I'd grumbled about it to my coworker, who replied "well using separate tables for each relationship is one way to do it, this is another..." Leaving aside the obvious-to-me ugliness of the above and the practical issues, I also have a nagging feeling this has to be a violation of some normal form, but it's been a while since college and I'm struggling to figure out which of the criteria apply here.
Is there something stronger "well that's just your opinion" I can use when critiquing this design?
