On Tue, 20 Jan 2004, Harald Fuchs wrote:
> In article <24300.1074614549@sss.pgh.pa.us>,
> Tom Lane <tgl@sss.pgh.pa.us> writes:
>
> > Richard Huxton <dev@archonet.com> writes:
> >> On Tuesday 20 January 2004 00:01, Neil Conway wrote:
> >>> Yeah, I didn't get around to implementing that. If anyone wants this
> >>> feature, I'd encourage them to step up to the plate -- I'm not sure
> >>> when I'll get the opportunity/motivation to implement this myself.
>
> >> I didn't think they'd be meaningful for a statement-level trigger. Surely
> >> OLD/NEW are by definition row-level details.
>
> > According to the complainants, OLD/NEW are commonly available as
> > recordsets (tables) inside a statement trigger.
>
> Yes.
>
> > I'm not very clear on
> > how that works myself --- in particular, one would think it important to
> > be able to work with corresponding pairs of OLD and NEW rows, which
> > would be painful with a table-like abstraction.
>
> Why? If the underlying table has a primary key, finding corresponding
> pairs is trivial; if there isn't, it's impossible.
I don't think that's sufficient unless you can guarantee that the primary
key values never change for any reason that causes the trigger to try to
correspond them.