Question

I’m having an Idiot Day today. I’m sure this is relatively simple, but my brain just isn’t giving me the answer.

I have a table whose rows are types of object. Looks something like this:

id    name    foo    bar    house_id
1     Cat     12     4      1
2     Cat     9      4      2
3     Dog     8      23     1
4     Bird    9      54     1
5     Bird    78     2      2
6     Bird    29     32     3

This isn’t how I’d choose to implement it, but it’s what I’m working with. Objects (cats, dogs and birds, in real life they’re actual business things) have been added to the table on an ad-hoc basis. When house_id 1 needs cats in it, a record for cats gets put in. When house_id 3 gets dogs, a record gets put in for dogs.

I now need to update this table so every type of object (Cat, Dog, Bird) has a record for a given house_id. I want to do this by inserting the result from a select query that returns a single record for each type, with the earliest values for ‘foo’ and ‘bar’ from a row of that type, if and only if there is no existent record for that type with the given house_id.

So for the above example data, where the given house_id = 3, the select query would return the following:

name    foo    bar    house_id
Cat     12     4      3
Dog     8      23     3

which I can then insert straight into the table.

Basically, return the first row of each distinct name if there are no rows of that name with a given house_id.

Suggestions welcome. DB engine is postgres if that helps.

Answer 1

SET search_path= 'tmp';

DROP TABLE dogcat CASCADE;
CREATE TABLE dogcat
        ( id serial NOT NULL
        , zname    varchar
        , foo    INTEGER
        , bar    INTEGER
        , house_id INTEGER NOT NULL
        , PRIMARY KEY (zname,house_id)
        );
INSERT INTO dogcat(zname,foo,bar,house_id) VALUES
  ('Cat',12,4,1)
 ,('Cat',9,4,2)
 ,('Dog',8,23,1)
 ,('Bird',9,54,1)
 ,('Bird',78,2,2)
 ,('Bird',29,32,3)
        ;
-- Carthesian product of the {name,house_id} domains
WITH cart AS (
        WITH beast AS (
                SELECT distinct zname AS zname
                FROM dogcat
                )
        , house AS (
                SELECT distinct house_id AS house_id
                FROM dogcat
                )
        SELECT beast.zname AS zname
        ,house.house_id AS house_id
        FROM beast , house
        )
INSERT INTO dogcat(zname,house_id, foo,bar)
SELECT ca.zname, ca.house_id
        ,fb.foo, fb.bar
FROM cart ca
     -- find the animal with the lowes id
JOIN dogcat fb ON fb.zname = ca.zname AND NOT EXISTS
        ( SELECT * FROM dogcat nx
        WHERE nx.zname = fb.zname
        AND nx.id < fb.id
        )
WHERE NOT EXISTS (
        SELECT * FROM dogcat dc
        WHERE dc.zname = ca.zname
        AND dc.house_id = ca.house_id
        )
        ;

SELECT * FROM dogcat;

Result:

SET
DROP TABLE
NOTICE:  CREATE TABLE will create implicit sequence "dogcat_id_seq" for serial column "dogcat.id"
NOTICE:  CREATE TABLE / PRIMARY KEY will create implicit index "dogcat_pkey" for table "dogcat"
CREATE TABLE
INSERT 0 6
INSERT 0 3
 id | zname | foo | bar | house_id 
----+-------+-----+-----+----------
  1 | Cat   |  12 |   4 |        1
  2 | Cat   |   9 |   4 |        2
  3 | Dog   |   8 |  23 |        1
  4 | Bird  |   9 |  54 |        1
  5 | Bird  |  78 |   2 |        2
  6 | Bird  |  29 |  32 |        3
  7 | Cat   |  12 |   4 |        3
  8 | Dog   |   8 |  23 |        2
  9 | Dog   |   8 |  23 |        3
(9 rows)