On Fri, 21 Apr 2006, Andrus wrote:
> > ... and there doesn't appear (for
> > non-match partial constraints) seem to be a special case for the
> > referenced row coming back into existance as far as I can tell either.
>
> > Or, if you're willing to patch, I think a first order approximation of
> > what you want might be to remove the special cases in trigger.c
> > (afterTriggerSetState) and tablecmds.c (createForeignKeyTriggers), but I
> > haven't tested that.
>
> Thank you.
> So I must create and maintain special version of PostgreSQL ?
If the standard does say it should do what you want, it'll get changed for
a later version, but probably not backpatched, so this would be a short
term solution. The hardest part about changing it is making sure there
aren't any new holes in the constraint.
If the standard doesn't match what you want, then it's a bit more
involved. Following the standard would still require you to maintain
a special version for the rules you want or changing the expectation. Or,
alternatively, you could make a case why the standard does say what you
want (or allows what you want or is simply wrong) if there's a
disagreement.