Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:
> Right now, referential actions get deferred along with normal
> checks and run against the state of the database at that time.
> I think this violates SQL92 11.8 General Rules 4-6 and have some
> reasoning and proposed ideas towards making it more complient
> although I don't actually have an implementation in mind for
> the most correct version. :(
I'm not convinced. 11.8 GR 1 refers to clause 10.6 as specifying
when the referential constraint is to be checked. 10.6 says that
immediate-mode constraints are checked "on completion" of each SQL
statement. (It doesn't say anything about deferred-mode constraints,
but I suppose those are checked at end of transaction.)
I think the intended meaning is that the actions caused by the
constraint are taken when the constraint is checked, which is
either end of statement or end of transaction. Which is what
we're doing now.
regards, tom lane