Gurjeet Singh <singh.gurjeet@gmail.com> writes:
> The other candidate to look for possible breakage would be pgAdmin. As Amit
> says, there might be code out in the wild that does look at this column, so
> not worth breaking them for this small gain.
Well, I already did it the other way ;-). It's probably not that big a
deal either way, since in practice it's a sure thing that 9.3 will have
many bigger and more-painful-to-deal-with catalog changes than this one.
But the fact that both pg_dump and psql go through pg_get_constraintdef
rather than looking directly at the catalogs, and always have (where
"always" = "since the pg_constraint catalog was invented, in 7.3")
makes me think that other clients probably do likewise. It's rather
painful to get the list of FK column names out of pg_constraint's
array representation in pure SQL.
regards, tom lane