Stephan Szabo <sszabo(at)megazone23(dot)bigpanda(dot)com> writes:
> ... If the table doesn't exist or there
> aren't enough args to intuit a table name, it currently
> elog errors. Is that reasonable behavior, or should
> it be taking the constraint and letting it fail at
> runtime?
I was inclined to think it should take the constraint and let the
failure occur at runtime. A NOTICE would be okay but not an ERROR.
(We can tweak the RI triggers to make the runtime failure message be
more helpful than "Relation 0 not found".) ISTM the point of having
this hack is to allow old dump files to be loaded, and an ERROR would
get in the way of that.
regards, tom lane