Postgres remove gaps in row numbers - sql

Question

I have a table called assignment with an index and unit_id columns. Every assignment in a unit has a unique index. This way, assignments in a unit can be reordered by swapping their indexes.
I'm trying to figure out the best way to remove potential gaps in the indexes (if a row is deleted, the indexes will go 0, 1, 3 for example). My current solution is to loop through each assignment in a unit programmatically and run an UPDATE query if its index doesn't match the loop index. Like this:
let i = 0;
for (const assignmentId of assignmentIds) {
await Assignment.query()
.patch({ index: i })
.where('id', assignmentId);
i++;
}
I'm trying to figure out how to do this with a single query using the ROW_NUMBER function like this:
UPDATE assignment SET
index = subquery.new_index - 1
FROM (
SELECT
ROW_NUMBER() OVER () as new_index
FROM assignment
WHERE assignment.unit_id = 35
ORDER BY assignment.index
) as subquery
WHERE unit_id=35;
But when I run this, it just sets all the indexes in unit 35 to 1. Why is this happening?

Answer 1

You need to provide a new index for each row by identifying the rows by old index:
update assignment
set index = subquery.new_index - 1
from (
select index as old_index, row_number() over (order by index) as new_index
from assignment
where assignment.unit_id = 35
) as subquery
where unit_id = 35
and old_index = index
and new_index <> old_index + 1; -- eliminate unnecessary updates

Answer 2

You need to join the subquery on the PK column (you neend you need to include that in the sub-query to be able to do that). The order by needs to go into the window function, not the overall query:
UPDATE assignment
SET index = subquery.new_index - 1
FROM (
SELECT pk_column,
ROW_NUMBER() OVER (partition by unit_id order by index) as new_index
FROM assignment
) as subquery
WHERE subquery.pk_column = assignment.pk_column
If you only want to do that for a single unit, you can add AND unit_id = 35 to the WHERE clause of the UPDATE statement.

Answer 3

If you don't mind the original index, you can run update with a temp sequence:
CREATE TEMP SEQUENCE new_index;
SELECT SETVAL('new_index', 1);
And run the update based on your unit:
UPDATE assignment SET index=nextval('new_index') WHERE unit_id=35;
This will update all lines on the unit_id=35 on the index column.
What's the diference from your FROM clause? The nextval() executes a function for every line, returning the next number of sequence.

Answer 4

Since you specifically indicated unit_id was not unique ("sets all the indexes in unit 35") you can use rownum() to project the expected index for each intem (as others have indicated). But I also hate using of magic numbers, the following does not depend on them.
with expectations as
(select *
from (
select unit_id
, unit_index
, row_number() over (partition by unit_id order by unit_id, unit_index) exp_index
from units
) au
where unit_index != exp_index
)
update units u
set unit_index = exp_index
from expectations e
where (u.unit_id, u.unit_index) = (e.unit_id,e.unit_index);
See fiddle here.