Thread

Make CREATE AGGREGATE check validity of initcond value?

Tom Lane <tgl@sss.pgh.pa.us> — 2012-10-04T02:38:00Z

In http://archives.postgresql.org/pgsql-general/2012-10/msg00138.php
we see an example where a user tried to create an aggregate whose
"initcond" (initial transition value) wasn't valid for the transition
data type.  CREATE AGGREGATE didn't complain because it just stores the
initial condition as a text string.  It seems to me that it'd be a lot
more user-friendly if it did check the value, as per the attached
proposed patch.

Does anyone have an objection to this?  I can imagine cases where the
check would reject values that would get accepted at runtime, if the
type's input function was sensitive to the phase of the moon or
something.  But it doesn't seem very probable, whereas checking the
value seems like an eminently useful thing to do.  Or maybe I'm just
overreacting to the report --- I can't recall any previous complaints
like this, so maybe entering a bogus initcond is a corner case too.

			regards, tom lane

Re: Make CREATE AGGREGATE check validity of initcond value?

Jaime Casanova <jaime@2ndquadrant.com> — 2012-10-04T16:48:42Z

El 03/10/2012 21:38, "Tom Lane" <tgl@sss.pgh.pa.us> escribió:
>
> Does anyone have an objection to this?  I can imagine cases where the
> check would reject values that would get accepted at runtime, if the
> type's input function was sensitive to the phase of the moon or
> something.  But it doesn't seem very probable, whereas checking the
> value seems like an eminently useful thing to do.  Or maybe I'm just
> overreacting to the report --- I can't recall any previous complaints
> like this, so maybe entering a bogus initcond is a corner case too.

I guess a wrong initcond value, probably is a pilot error.
So, my first reaction is +1 to make it an error.

But if you feel there a corner cases maybe at least a warning

--
Jaime Casanova
2ndQuadrant: Your PostgreSQL partner