On Wed, 2011-05-04 at 10:49 -0400, Tom Lane wrote:
> Well, it'd require a very large amount of
> type-specific/operator-specific knowledge, and it's not clear what
> would drive the planner towards doing useful rearrangements rather
> than counterproductive ones, and the number of real-world queries
> where it'd actually help doesn't seem to be that large. I've seen one
> or two complaints about that sort of thing, but it's way way down the
> list of serious problems.
I did suspect that the answer would be that the difficulty out ways the
benefit. But in terms of driving the planner don't we always want to be
looking to move all the constants to one side of the expression since
the planner seems to like those?
--
Michael Graham <mgraham(at)bloxx(dot)com>