On Aug 20, 2008, at 12:43 PM, Tom Lane wrote:
We have speculated in the past about having alternative plans that
could be conditionally executed based on information not available
at planning time. This could be seen as a first experiment in that
direction. I am not thinking of a general-purpos
> Another technique that we could play with is to have the
> AlternativeSubPlans node track the actual number of calls it gets,
> and switch from the "retail" implementation to the "hashed"
> implementation if that exceeds a threshold. This'd provide some
> robustness in the face of bad estimates,
>>> Tom Lane <[EMAIL PROTECTED]> wrote:
> when we have an EXISTS that could be done both ways,
> why not just generate plans for both ways, and leave the decision
> which to use until later?
That seems good to me. The costs for the slower plan generally come
out much higher. When the run tim
"Kevin Grittner" <[EMAIL PROTECTED]> writes:
> Tom Lane <[EMAIL PROTECTED]> wrote:
>> [ complicated scheme for improving planning of EXISTS ]
> So I'd be very happy to see this work done, not because I can't find a
> workaround, but because trying to teach all the programmers tricky
> hand-optim
>>> Tom Lane <[EMAIL PROTECTED]> wrote:
> The examples that Kevin Grittner put up awhile back included several
> uses of EXISTS() in places where it couldn't be turned into a
semijoin,
> eg in the query's targetlist. I was musing a bit about whether we
could
> improve those scenarios. I would
* Tom Lane ([EMAIL PROTECTED]) wrote:
> So ... I'm wondering if this actually touches anyone's hot-button,
> or if we should just file it in the overflowing pile of Things That
> Might Be Nice To Do Someday.
What bugs me the most about having IN() be faster than EXISTS() in
certain situations is t
Tom Lane wrote:
> I don't think any of this is out of reach, but it'd be a nontrivial
> bit of implementation effort (maybe a week or three) and it also looks
> like there might be a measurable planning slowdown for any query
> involving subqueries. I'm not sure yet how much of this is just movin
The examples that Kevin Grittner put up awhile back included several
uses of EXISTS() in places where it couldn't be turned into a semijoin,
eg in the query's targetlist. I was musing a bit about whether we could
improve those scenarios. I would like to get 8.4 to the point where we
could say as
