On Fri, 26 May 2000, Peter Eisentraut wrote:
> Zeugswetter Andreas SB writes:
>
> > Which is imho not what the test is for. I think they mean ctid,
> > which again I think we should have a rowid alias for (as in Informix,
> > Oracle).
>
> Let's step back and ask: How is the behaviour of rowid (or whatever)
> defined in various existing DBMS. Then we can see if we have anything that
> matches.
This has been discussed. The outcome is, that you are only safe using rowid
if nobody else changes the row inbetween you reading it and accessing it by rowid.
This is essentially the same in all db's only the risk of rowid changing is lower
in other db's since they do inplace update, but the risk is there nevertheless.
Andreas