Stephan Szabo <sszabo(at)megazone(dot)bigpanda(dot)com> writes:
> While not exactly what the spec invisions, I believe we can come
> closer to the correct behavior by treating all of the referential actions
> as non-deferrable while allowing deferment of NO ACTION and the check
> itself.
Since no one seems to have objected, I've applied this patch.
regards, tom lane