Re: Foreign keys for non-default datatypes, redux

On Sat, 10 Feb 2007, Tom Lane wrote:

> Stephan Szabo <sszabo(at)megazone(dot)bigpanda(dot)com> writes:
> > One side question is what should we do about the
> > places in the current system where it checks for the key sets being empty?
>
> I don't see that this affects that either way. I can't quite imagine
> what the semantics would be, though --- there's no such thing as a
> unique constraint with no columns, so how can there be an RI constraint
> with none?

Well, the code currently has checks with comments based on SQL3
text AFAICT.
/* ----------
* SQL3 11.9 <referential constraint definition>
* General rules 2) a):
* If Rf and Rt are empty (no columns to compare given)
* constraint is true if 0 < (SELECT COUNT(*) FROM T)
*
* Note: The special case that no columns are given cannot
* occur up to now in Postgres, it's just there for
* future enhancements.
* ----------
*/
The reason I was wondering is that it uses tgnargs == 4 as the check, and
if we change the meanings of tgnargs, we'd need to change the check.
Personally, I think we should probably just pull out the special case for
now, but thought it should be brought up.