Andrew Dunstan wrote:
> I haven't been following this thread due to pressure of time, so my
> apologies in advance if these comments have already been covered.
>
> I've been asked to look at and comment on the SQL API of the feature. I
> think it's basically sound, although there is one thing that's not clear
> from the regression tests: what happens if we're inserting into an object
> and the key already exists? e.g.:
>
> select jsonb_insert('{"a": {"b": "value"}}', '{a, b}', '"new_value"');
It has been covered: Petr said, and I agree with him, that this new
interface is for arrays, not objects, and so the above example should be
rejected altogether. For objects we already have jsonb_set(), so what
is the point of having this work there? It feels too much like
TIMTOWTDI, which isn't a principle we adhere to.
I think it'd be much more sensible if we were just to make this function
work on arrays only. There, the solution to Andrew's question is
trivial: if you insert a value in a position that's already occupied,
the elements to its right are "shifted" one position upwards in the
array (this is why this is an "insert" operation and not "replace" or
"set").
--
Álvaro Herrera http://www.2ndQuadrant.com/
PostgreSQL Development, 24x7 Support, Remote DBA, Training & Services