On 01/12/2014 11:20 PM, Peter Geoghegan wrote:
On Sun, Jan 12, 2014 at 8:12 AM, Andreas Karlsson <andr...@proxel.se> wrote:
On 01/11/2014 11:42 PM, Peter Geoghegan wrote:
I recently suggested that rather than RETURNING REJECTS, we could have
a REJECTING clause, which would see a DML statement project strictly
the complement of what RETURNING projects in the same context. So
perhaps you could also see what RETURNING would not have projected
because a before row trigger returned NULL (i.e. when a before trigger
indicates to not proceed with insertion). That is certainly more
general, and so is perhaps preferable. It's also less verbose, and it
seems less likely to matter that we'll need to make REJECTING a fully
reserved keyword, as compared to REJECTS. (RETURNING is already a
fully reserved keyword not described by the standard, so this makes a
certain amount of sense to me). If nothing else, REJECTING is more
terse than RETURNING REJECTS.
I do not entirely understand what you are proposing here. Any example how
this would look compared to your RETURNING REJECTS proposal?
It's very similar - REJECTING is a total generalization of what I
already have. The difference is only that REJECTING is accepted in all
contexts that RETURNING is, and not just with INSERT...ON DUPLICATE
KEY LOCK FOR UPDATE. So you could potentially have REJECTING project
the slot proposed for insertion on an UPDATE where RETURNING would
not. If for example a BEFORE ROW trigger fired, and returned NULL,
perhaps it'd then be possible to project the slot as it was before
being passed to the trigger. Perhaps there is no real demand for that,
but, as I said, from a usability perspective it may be easier to
reason about a feature that projects strictly the complement of what
RETURNING would project in the same context.
So simply this?
WITH rej AS (
INSERT INTO foo (a, b, c)
VALUES (44, 1078, 'insert'), (55, 1088, 'insert')
UPDATE foo SET c = 'update' FROM rej WHERE foo.a = rej.a;
Another question: have you given any thought on the case where you want
to use both the successfully inserted tuples and the rejected and use in
the CTE? Is that even something anyone would want? Would perhaps MERGE
be more suited for that?
Sent via pgsql-hackers mailing list (email@example.com)
To make changes to your subscription: