how to remove duplicate results - sql

Question

given the following schema:
CREATE TABLE IF NOT EXISTS companies (
id serial,
name text NOT NULL,
PRIMARY KEY (id)
);
CREATE TABLE IF NOT EXISTS cars (
id serial,
make text NOT NULL,
year integer NOT NULL,
company_id INTEGER REFERENCES companies(id),
PRIMARY KEY (id)
);
INSERT INTO companies (id, name) VALUES
(1, 'toyota'),
(2, 'chevy');
INSERT INTO cars (make, year, company_id) VALUES
('silverado', 1995, 2),
('malibu', 1999, 2),
('tacoma', 2017, 1),
('custom truck', 2010, null),
('van custom', 2005, null);
how do i select the rows for cars, only showing the newest car for a given company?
e.g.
select make, companies.name as model, year from cars
left join companies
on companies.id = cars.company_id
order by make;
outputs
make | model | year
--------------+--------+------
custom truck | | 2010
malibu | chevy | 1999
silverado | chevy | 1995
tacoma | toyota | 2017
van custom | | 2005
but i only want to show the newest "chevy", e.g.
make | model | year
--------------+--------+------
custom truck | | 2010
malibu | chevy | 1999
tacoma | toyota | 2017
van custom | | 2005
and still be able to sort by "make", and to show cars without a null company_id.
fiddle link:
https://www.db-fiddle.com/f/5Vh1sFXvEvnbnUJsCYhCHf/0

Answer 1

SQL can be done based on Set Math (discrete math). So, you want the set of all cars minus the set of cars whose years a less than the maximum year for a given company id.
The set of all cars:
select * from cars
The set of all cars whose year is less than the maximum year for a given company id:
select a.id from cars a, cars b where a.company_id = b.company_id and a.year < b.year
One set minus the other:
select * from cars where id not in (select a.id from cars a, cars b where a.company_id = b.company_id and a.year < b.year)
Result which includes the null company_ids because they are excluded from the id comparison:
make | model | year
--------------+--------+------
custom truck | | 2010
malibu | chevy | 1999
tacoma | toyota | 2017
van custom | | 2005

Answer 2

With the help of common table expressions and row_number function, we can get the desired output and below is the query that gives the desired output.
WITH temp AS
(SELECT
make
, companies.name AS model
, year
, row_number() over(PARTITION BY coalesce(companies.name, make) ORDER BY year desc) as rnk
FROM
cars
left join
companies
ON
companies.id = cars.company_id
)
SELECT
make
, model
, year
FROM
temp
WHERE
rnk = 1
;

Answer 3

In Postgres, this is best done using distinct on:
select distinct on (co.id) ca.*, co.name as model
from cars ca left join
companies co
on ca.company_id = co.id
order by co.id, ca.year desc;
DISTINCT ON is very handy Postgres syntax. It keeps one row for each combination in parentheses. The specific row is determined by the ORDER BY clause.
However, you have a twist, because co.id can be null. In that case, you seem to want to keep all the cars with no company.
So:
select distinct on (co.id, case when co.id is null then ca.id end) ca.*, co.name
from cars ca left join
companies co
on ca.company_id = co.id
order by co.id, case when co.id is null then ca.id end, ca.year desc;
Or perhaps more simply using union all:
-- get the ones with a company
select distinct on (co.id) ca.*, co.name
from cars ca join
companies co
on ca.company_id = co.id
union all
-- get the ones with no company
select ca.*, null
from cars ca
where ca.company_id is null
order by year desc;
In other databases, this would typically be done using row_number():
select ca.*
from (select ca.*, co.name as model,
row_number() over (partition by co.id,
case when co.id is null then ca.id end
order by year desc
) as seqnum
from cars ca left join
companies co
on ca.company_id = co.id
) ca
where seqnum = 1