On Wed, Jun 3, 2015 at 7:02 PM, Peter Geoghegan <pg@heroku.com> wrote:
> Consider this case:
>
> postgres=# select '{"c":5, "a":6, "b":7}'::jsonb - 1;
> ?column?
> ------------------
> {"a": 6, "c": 5}
> (1 row)
>
> Clearly anyone expecting the value "a" to be removed here would be in
> for a surprise. Moreover, it is inconsistent with the established
> behavior of the corresponding array-wise subscript operator:
>
> postgres=# select '{"c":5, "a":6, "b":7}'::jsonb -> 1;
> ?column?
> ----------
> [null]
> (1 row)
For similar reasons, I think that this inconsistency is unacceptable:
postgres=# select '["a", "b", "c"]'::jsonb - -1; ?column?
------------["a", "b"]
(1 row)
postgres=# select '["a", "b", "c"]'::jsonb -> -1;?column?
----------[null]
(1 row)
jsonb now supports Python-style negative subscripting to index
backward. I think that this a fine idea. However, I also think it's a
big POLA violation that this was not done for the ordinary array
subscripting operator ("operator jsonb -> integer") at the same time
as "operator jsonb - integer" was added. Although doing this will
require a compatibility note in the 9.5 release notes, it's extremely
unlikely to destabilize anybody's app, and makes a lot of sense.
--
Peter Geoghegan
