Martijn van Oosterhout <kleptog@svana.org> writes:
> On Tue, Dec 26, 2006 at 12:08:04PM -0500, Kevin Murphy wrote:
>> When columns are required in a GROUP BY clause even though some of them
>> are fully dependent on others, is there any plan for making it possible
>> to do the GROUP BY only on the necessary columns?
> Recent SQL standards require it, so it's likely to be implemented at
> some point.
This optimization is in the nature of depending on a constraint (ie, a
PRIMARY KEY) for plan correctness, so I'm hesitant to think about doing
it before we have plan invalidation implemented. Actually it's even
worse than that: it's in the nature of depending on a constraint for
*query* correctness. If you ALTER DROP PRIMARY KEY then your query is
now illegal, and we really oughta notice that ...
regards, tom lane