Re: BUG #5652: Optimizer does wrong thing with partitioned tables

On Fri, 2010-09-10 at 08:10 -0700, Chris Travers wrote:
> Just adding my voice to the "fix it" camp.  Is there any reason the
> table scans in this sort of thing cannot be independently planned?

I don't think it's about independent planning. For instance, AVG clearly
can't be planned this way, there are particular properties of MAX that
allow the optimization:

  1. MAX(x) can be rewritten as: ORDER BY x DESC LIMIT 1
  2. The MAX of set S is the MAX of the MAXes of each partition of S

The optimizer knows about the former, but not the latter.

Regards,
    Jeff Davis