Peter Eisentraut <peter_e@gmx.net> writes:
> Instead of trying to fix this, how about making the RI triggers not use
> the column names in the first place. (Instead they should use the oid of
> the table and the attnums.)
That would probably be a better answer in the long run, but since it
implies initdb for everyone with an RI trigger, I'm disinclined to
attack it that way during beta.
regards, tom lane