transform columns to rows - sql

Question

I have a table table1 like below
+----+------+------+------+------+------+
| id | loc | val1 | val2 | val3 | val4 |
+----+------+------+------+------+------+
| 1 | loc1 | 10 | 190 | null | 20 |
| 2 | loc2 | 20 | null | 10 | 10 |
+----+------+------+------+------+------+
need to combine the val1 to val4 into a new column val with a row for each so that the output is like below.
NOTE: - I data I have has val1 to val30 -> ie. 30 columns per row that need to be converted into rows.
+----+------+--------+
| id | loc | val |
+----+------+--------+
| 1 | loc1 | 10 |
| 1 | loc1 | 190 |
| 1 | loc1 | null |
| 1 | loc1 | 20 |
| 2 | loc2 | 20 |
| 2 | loc2 | null |
| 2 | loc2 | 10 |
| 2 | loc2 | 10 |
+----+------+--------+

Answer 1

You can use lateral join for transform columns to rows :
SELECT a.id,a.loc,t.vals
FROM table1 a,
unnest(ARRAY[a.val1,a.val2,a.val3,a.val4]) t(vals);
If you want to this with a dynamic added columns:
CREATE OR REPLACE FUNCTION columns_to_rows(
out id integer,
out loc text,
out vals integer
)
RETURNS SETOF record AS
$body$
DECLARE
columns_to_rows text;
BEGIN
SELECT string_agg('a.'||attname, ',') into columns_to_rows
FROM pg_attribute
WHERE attrelid = 'your_table'::regclass AND --table name
attnum > 0 and --get just the visible columns
attname <> all (array [ 'id', 'loc' ]) AND --exclude some columns
NOT attisdropped ; --column is not dropped
RETURN QUERY
EXECUTE format('SELECT a.id,a.loc,t.vals
FROM your_table a,
unnest(ARRAY[%s]) t(vals)',columns_to_rows);
end;
$body$
LANGUAGE 'plpgsql'
Look at this link for more detail: Columns to rows

Answer 2

You could use a cross join with generate_series for this:
select
id,
loc,
case x.i
when 1 then val1
when 2 then val2
. . .
end as val
from t
cross join generate_series(1, 4) x (i)
It uses the table only once and can be easily extended to accommodate more columns.
Demo
Note: In the accepted answer, first approach reads the table many times (as many times as column to be unpivoted) and second approach is wrong as there is no UNPIVOT in postgresql.

Answer 3

I'm sure there's a classier approach than this.
SELECT * FROM (
select id, loc, val1 as val from #t a
UNION ALL
select id, loc, val2 as val from #t a
UNION ALL
select id, loc, val3 as val from #t a
UNION ALL
select id, loc, val4 as val from #t a
) x
order by ID
Here's my attempt with unpivot but cant get the nulls, perhaps perform a join for the nulls? Anyway i'll still try
SELECT *
FROM (
SELECT * FROM #t
) main
UNPIVOT (
new_val
FOR val IN (val1, val2, val3, val4)
) unpiv

Answer 4

It will not work in postgress as needed by user. Saw when it was mentioned in comments.
I am finding a way to handle "NULL"
select p.id,p.loc,CASE WHEN p.val=0 THEN NULL ELSE p.val END AS val
from
(
SELECT id,loc,ISNULL(val1,0) AS val1,ISNULL(val2,0) AS val2,ISNULL(val3,0) AS val3,ISNULL(val4,0) AS val4
FROM Table1
)T
unpivot
(
val
for locval in(val1,val2,val3,val4)
)p
Test
EDIT:
Best Solution from my Side:
select a.id,a.loc,ex.val
from (select 'val1' as [over] union all select 'val2' union all select 'val3'
union all select 'val1' ) pmu
cross join (select id,loc from Table1) as a
left join
Table1 pt
unpivot
(
[val]
for [over] in (val1, val2, val3, val4)
) ex
on pmu.[over] = ex.[over] and
a.id = ex.id
Test