Bruce Momjian <pgman@candle.pha.pa.us> writes:
> Good question. Several people thought attno was the way to go and it
> seemed more natural because almost everything else goes by attno and not
> name.
Since the primary key for pg_attribute is reloid+attnum, it would seem
that that's an appropriate representation for RI trigger links too.
But Hiroshi's got a point: we have a whole set of issues with tracking
object dependencies, and it probably makes sense to think about how we
are going to deal with those issues before we fool around with changing
the representation of RI triggers.
regards, tom lane