Question

Let’s say we have these tables:

CREATE TABLE A (
    id SERIAL NOT NULL PRIMARY KEY
);
CREATE TABLE B (
    id SERIAL NOT NULL PRIMARY KEY
);
CREATE TABLE Parent (
    id SERIAL NOT NULL PRIMARY KEY,
    aId INTEGER NOT NULL REFERENCES A (id),
    bId INTEGER NOT NULL REFERENCES B (id),
    UNIQUE(aId, bId)
);
CREATE TABLE Child (
    parentId INTEGER NOT NULL REFERENCES Parent (id),
    createdOn TIMESTAMP NOT NULL
);

Is it possible to create a unique constraint on Child such that for all rows in Child at most one references a Parent having some value of aId? Stated another way can I created a unique constraint so that the join of the above tables will have no duplicate aId? I'm thinking not--the grammars of every database I could find seem tied to one table per constraint--but that might be a lack of imagination on my part. (De-normalizing to include aId on Child is one solution, of course.)

Answer 1

You could try the following. You have to create a redundant UNIQUE constraint on (id, aId) in Parent (SQL is pretty dumb isn’t it?!).

CREATE TABLE Child
(parentId INTEGER NOT NULL,
 aId INTEGER NOT NULL UNIQUE,
FOREIGN KEY (parentId,aId) REFERENCES Parent (id,aId),
createdOn TIMESTAMP NOT NULL);

Possibly a much better solution would be to drop parentId from the Child table altogether, add bId instead and just reference the Parent table based on (aId, bId):

CREATE TABLE Child
(aId INTEGER NOT NULL UNIQUE,
 bId INTEGER NOT NULL,
FOREIGN KEY (aId,bId) REFERENCES Parent (aId,bId),
createdOn TIMESTAMP NOT NULL);

Is there any reason why you can’t do that?