Enforce Uniqueness of Related Entities - sql-server

Question

In a relational database (SQL), I have a parent entity that can have 0..n related child entities. The parent entity is uniquely identified in part by its collection of related child entities, such that I should not be able to have two similar parents with the same collection of children.
So I could have Parent 1 with Child 1 and Child 2, and Parent 2 with Child 2 and Child 3, but I cannot have another parent with Child 2 and Child 3.
Ideally, I would like to enforce this uniqueness using a database constraint. I've considered storing a hash of all child records with the parent, but was wondering if there was an easier / more standard way of accomplishing this.
Any ideas?

Answer 1

This kind of constraint is tricky because SQL has no relational equality operator, i.e. no simple way of evaluting A=B where A and B are sets of rows. Standard SQL does support nested tables but unfortunately SQL Server does not.
One possible answer is a predicate like the following, which checks for any identical families in a table:
NOT EXISTS (
SELECT 1
FROM family f, family g
WHERE f.child = g.child
AND f.parent <> g.parent
GROUP BY f.parent, g.parent
HAVING COUNT(*) = (SELECT COUNT(*) FROM family WHERE parent = f.parent)
AND COUNT(*) = (SELECT COUNT(*) FROM family WHERE parent = g.parent)
)
Notice that this query doesn't attempt to deal with childless families. In set-theoretic terms two empty sets are necessarily identical. If you want to allow for childless families then you would have to decide whether two childless families should be deemed identical or not.
SQL is not a truly relational language and it falls well short of what a relational language ought to be capable of. Tutorial D is an example of a real relational language that does support relational equality and relation-valued attributes. In Tutorial D you can in principle represent each family as a value of a single attribute in a relvar. That family attribute can also be a key and therefore duplicate families would not be allowed.

Answer 2

Thanks for the help from those who suggested using a trigger. This is roughly what I have and seems to be working.
CREATE TRIGGER [dbo].[trig_Parent_Child_Uniqueness]
ON [dbo].[Parent_Child]
AFTER INSERT, UPDATE
AS
BEGIN
IF EXISTS (
SELECT 1
FROM Parent p1
--Compare each pair of parents
JOIN Parent p2 ON p1.ParentId <> p2.ParentId
WHERE NOT EXISTS (
--Find any children that are different
SELECT 1
FROM (
SELECT ChildId FROM Parent_Child c1
WHERE c1.ParentId = p1.ParentId
) as c1
FULL OUTER JOIN (
SELECT ChildId FROM Parent_Child c2
WHERE c2.ParentId = p2.ParentId
) as c2 ON c2.ChildId = c1.ChildId
WHERE c1.ChildId IS NULL OR c2.ChildId IS NULL
)
) ROLLBACK;
END;
EDIT: Or a better solution, adapted from #sqlvogel
CREATE TRIGGER [dbo].[trig_Parent_Child_Uniqueness]
ON [dbo].[Parent_Child]
AFTER INSERT, UPDATE
AS
BEGIN
IF EXISTS (
SELECT 1
FROM Parent_Child p1
FULL JOIN Parent_Child p2 ON p1.ParentId <> p2.ParentId
AND p1.ChildId = p2.ChildId
GROUP BY p1.ParentId
HAVING COUNT(p1.ParentId) = COUNT(*)
AND COUNT(p2.ParentId) = COUNT(*)
) ROLLBACK;
END;

Answer 3

This is a bit yucky as it includes triggers and cursors :(
It includes a column in the parent table which is based upon the children
Set up:
CREATE TAble Parent
(
Id INT Primary Key,
Name VARCHAR(50),
ChildItems VARCHAR(200) NOT NULL UNIQUE
)
CREATE TABLE Child
(
Id INT Primary Key,
Name VARCHAR(50)
)
CREATE TABLE ParentChild
(
Id INT Identity Primary Key,
ParentId INT,
ChildId Int
)
Triggers
-- This gives the unique colmn a default based upon the id of the parent
CREATE TRIGGER trg_Parent ON Parent
INSTEAD OF Insert
AS
SET NOCOUNT ON
INSERT INTO Parent (Id, Name, ChildItems)
SELECT Id, Name, '/' + CAST(Id As Varchar(10)) + '/'
FROM Inserted
GO
-- This updates the parent with a path based upon child items
-- If a the exact same child items exist for another parent then this fails
-- because of the unique index
CREATE Trigger trg_ParentChild ON ParentChild
AFTER Insert, Update
AS
DECLARE #ParentId INT = 0
DECLARE #ChildItems VARCHAR(8000) = ''
DECLARE parentCursor CURSOR FOR
SELECT DISTINCT ParentId
FROM Inserted
OPEN parentCursor
FETCH NEXT FROM parentCursor INTO #ParentId
WHILE ##FETCH_STATUS = 0
BEGIN
SELECT #ChildItems = COALESCE(#ChildItems + '/ ', '') + CAST(ChildID As Varchar(10))
FROM ParentChild
WHERE ParentId = #ParentId
ORDER BY ChildId
UPDATE Parent
SET ChildItems = #ChildITems
WHERE Id = #ParentId
FETCH NEXT FROM parentCursor INTO #ParentId
SET #ChildItems = ''
END
CLOSE parentCursor
DEALLOCATE parentCursor
GO
Data Setup
INSERT INTO Parent (Id, Name)
VALUES (1, 'Parent1'), (2,'Parent2'), (3, 'Parent3')
INSERT INTO Child (Id, Name)
VALUES (1,'Child1'), (2,'Child2'), (3,'Child3'), (4,'Child4')
Now insert some data
-- This one succeeds
INSERT INTO ParentChild (ParentId, ChildId)
VALUES (1,1),(1,2),(2,2),(2,3)
-- This one Fails
INSERT INTO ParentChild (ParentId, ChildId) VALUES (3,1),(3,2)