I am trying to achieve something like LEFT JOIN
with array unnest()
function. I want the query to return a row with a null value if the array is empty. Therefore, by using a CASE
construct, I wanted to pass fake array with single null element if the source array is empty, but it does not work as expected:
Query 1
select element
from (
select array['a']::text[] as arr --< single non-null element
) sub, unnest(
(
case when array_length(sub.arr, 1) <= 0 then (array[null])::text[]
else sub.arr
end
)
) element
-- returns 1 row with element = "a"
Query 2
select element
from (
select array[]::text[] as arr --< empty array
) sub, unnest(
(
case when array_length(sub.arr, 1) <= 0 then (array[null])::text[]
else sub.arr
end
)
) element
-- returns 0 rows (should return 1 row with element = NULL?)
Query 3
select element
from (
select array[null]::text[] as arr --< single null element
) sub, unnest(
(
case when array_length(sub.arr, 1) <= 0 then (array[null])::text[]
else sub.arr
end
)
) element
-- return single row with element = NULL
Just figured out that select array_length(array[]::text[], 1)
returns NULL
- my question is why?
To quote Tom Lane in the thread "Should array_length() Return NULL" on pgsql-hackers:
The thing is that that syntax creates an array of zero dimensions, not one that has 1 dimension and zero elements. So "0" would be incorrect.
Hard to say whether array_length()
should return NULL
or 0
for an empty array dimension ...
You could invert the logic and process the original source array only if the length is >= 1
:
WITH tbl(id, arr) AS (
VALUES
(1, '{a}'::text[])
, (2, '{NULL}'::text[])
, (3, '{}'::text[])
, (4, '{x,y,z}'::text[])
)
SELECT id, elem
FROM tbl t
, unnest (
CASE WHEN array_length(t.arr, 1) >= 1
THEN t.arr
ELSE '{null}'::text[] END
) elem;
Returns at least one row for each source row.
Or, much simpler, use an actual LEFT JOIN
(the comma in your FROM
clause is short notation for CROSS JOIN
):
SELECT id, elem
FROM tbl t
LEFT JOIN LATERAL unnest(t.arr) elem ON true;
This fills in NULL values automatically for columns of the right table if no match is found. Exactly what you are trying to achieve.
Related:
db<>fiddle here
External links referenced by this document: