On Sun, Mar 7, 2010 at 12:08 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
> Robert Haas <robertmhaas(at)gmail(dot)com> writes:
>> On Fri, Mar 5, 2010 at 5:32 PM, Tom Lane <tgl(at)sss(dot)pgh(dot)pa(dot)us> wrote:
>>> It's arguably a bug, but since we lack consensus on whether NULL and
>>> ROW(NULL,NULL,...) are the same thing, it's difficult to make a
>>> bulletproof case either way.
>> Have we or can we somehow document why this happens?
> The reason it happens is that the assignment target is a "row" variable,
> meaning that it doesn't have concrete existence as a tuple but is just
> an alias for a list of scalar variables. So there is no way for it to
> represent an atomic NULL; setting each of the individual scalars to NULL
> is possible but the result acts more like ROW(NULL,NULL,...).
> I'm not sure about documenting that. It seems like an implementation
> detail. If we had consensus that the two cases either should or should
> not be distinguishable, we could work towards making that happen; but
> lacking such consensus I'm hesitant to touch it at all.
What seems odd to me is that NEW is apparently some other kind of
thing that is not the same kind of thing as the row variable.
In response to
pgsql-bugs by date
|Next:||From: Tom Lane||Date: 2010-03-09 15:25:53|
|Subject: Re: Bug in triggers |
|Previous:||From: Andrea Suisani||Date: 2010-03-09 12:28:05|
|Subject: duplicate key violates unique contraint on pg_type_typname_nsp_index|