Thread: BUG #16022: to_json on arrays with unusual lower bound is not intuitive

The following bug has been logged on the website:

Bug reference:      16022
Logged by:          A Bergmaier
Email address:      ab@principiamentis.com
PostgreSQL version: 11.5
Operating system:   Ubuntu
Description:

Hi!
I would expect the array indices of a JSON array to match with the
subscripts of a postgres array when converting them back and forth.
However, Postgres lets the JSON array begin at the `lower_bound` of the
array, not at subscript 1 (= index 0).
For example, `to_json('[-2:5]={-2,-1,0,1,2,3,4,5}'::int[])` results in
`[-2,-1,0,1,2,3,4,5]` where I would have expected `[1,2,3,4,5]` (or an
error, since JSON arrays must not have negative indices).
And more annoyingly, `to_json('[3:8]={3,4,5,6,7,8}'::int[])` results in
`[3,4,5,6,7,8]` where I would have needed `[null,null,3,4,5,6,7,8]`.
(See also some demo
https://dbfiddle.uk/?rdbms=postgres_11&fiddle=2d359608815f7059f3120fa17da1bdef)

Just for reference, I'm on PostgreSQL 11.4 (Debian 11.4-1.pgdg90+1) on
x86_64-pc-linux-gnu, compiled by gcc (Debian 6.3.0-18+deb9u1) 6.3.0
20170516, 64-bit, but I doubt it matters.

I can circumvent this behaviour by using a function like
CREATE FUNCTION array_fillup(arr anyarray) RETURNS anyarray
AS $$
BEGIN
   IF array_lower(arr, 1) > 1 THEN
       arr[1] = NULL;
   END IF;
   RETURN arr;
END
$$ LANGUAGE plpgsql STRICT;
before passing my sparse array into `to_json`, but I would like to know
whether there's a better solution or whether the issue could be fixed at the
core. (Presumable, breaking backwards-compatibility is a problem?)

Kind regards,
 Andreas Bergmaier

PG Bug reporting form <noreply@postgresql.org> writes:
> I would expect the array indices of a JSON array to match with the
> subscripts of a postgres array when converting them back and forth.
> However, Postgres lets the JSON array begin at the `lower_bound` of the
> array, not at subscript 1 (= index 0).
> For example, `to_json('[-2:5]={-2,-1,0,1,2,3,4,5}'::int[])` results in
> `[-2,-1,0,1,2,3,4,5]` where I would have expected `[1,2,3,4,5]` (or an
> error, since JSON arrays must not have negative indices).

I can see no reason whatever for either of those definitions to be
better than the established one.  If you want some other conversion
rule, write your own function that behaves the way you want.

            regards, tom lane