Robert Haas <robertmhaas@gmail.com> writes:
> Well, I think if that's the best we can do, you original proposal of
> ditching the column constraint syntax altogether might be for the
> best. I wasn't too excited about that before, but I think having two
> different syntaxes is going to be even worse. In some ways, it's
> actually sort of sensible, because the referring side isn't really the
> column itself; it's some value extracted therefrom. You can imagine
> other variants of that as well, such as the recently-suggested
> FOREIGN KEY ((somecol).member_name) REFERENCES othertab (doohicky)
> Now, what would the column-constraint version of that look like? Is
> it even sensible to think that there SHOULD be a column-constraint
> version of that? I'm not convinced it is sensible, so maybe decreeing
> that the table constraint version must be used to handle all
> non-trivial cases is more sensible than I initially thought.
I could easily go with that ...
regards, tom lane