Stephan Szabo <sszabo@megazone23.bigpanda.com> writes:
> Yeah, that can get to be a problem... In any case, you'll probably get
> other comments. Oh yeah, and you'll probably be asked for documentation
> comments if it's even considered since you're adding a visible GUC entry.
> :)
Well, it won't be --- diking out required error checks without providing
a substitute isn't my idea of a useful patch ;)
The SQL99 spec defines an improved version of this behavior which I
think is what Daniele really would like to have. Basically it says that
if column A can be proved functionally dependent on column B then you
only need to GROUP BY column B, and then you can use column A without
having to explicitly mention it in the GROUP BY list. "Functionally
dependent" means there is no possibility of A values being different in
rows with the same B value. The spec has a whole lot of verbiage about
possible ways to deduce functional dependency, but one easy one is where
column B is a primary key and column A is in its table.
If someone wants to implement the SQL99 behavior (or even just a useful
subset of it), that would be cool with me. It looks like a lot of work
though.
regards, tom lane
