Am Sonntag, den 30.10.2005, 07:24 -0800 schrieb David Fetter:
...
> > In most cases, just DELETE and then INSERT should work perfectly.
> > (UPDATE and MERGE would cause dead tuples in the same way so in the
> > end they are only syntactical sugar)
> >
> > Another way is a rule for insert which turns it into an update in
> > case the desired tuple is already existent.
>
> That has a race condition in it. What happens if something deletes
> the tuple between the attempted INSERT and the UPDATE?
How so in a transaction?