I said:
> The table OIDs in pg_trigger would do fine if the trigger function could
> get at them, but it can't; so we need to copy them into the trigger
> arguments. (Hmm, I suppose another option is to extend the Trigger
> data structure to include tgconstrrelid, and just ignore the table names
> in the trigger argument list.)
After further thought, this is clearly the right approach to take,
because it provides a solution path for other triggers besides the RI
ones. So we'll fix the problem at the code level. The trigger
arguments will be unchanged, but the table names therein will become
purely decorative (or documentation, if you prefer ;-)). Perhaps
someday we could eliminate them ... but not as long as pg_dump dumps
RI constraints in the form of trigger definitions.
regards, tom lane