Re: query plan different for "SELECT ..." and "DECLARE

Tom Lane wrote:
> David Blasby <dblasby(at)refractions(dot)net> writes:
>>The only real problem is that the user has to manual keep stats
>>up-to-date. Is there anyway to attach something to VACUUM ANALYSE?
>
> The ANALYZE code is set up with the idea that there could be multiple
> analysis methods and various kinds of stuff stored in pg_statistic.
> Right now there isn't any way to physically plug in a different analysis
> routine :-( but it seems like it'd be reasonable to add some hook of
> that kind. Perhaps CREATE TYPE could be extended to specify an analysis
> routine for ANALYZE to call for columns of that type.
>
> Please take a look at pg_statistic.h and commands/analyze.c and see if
> this would do the job for you. Obviously it's too late for 7.4 but we
> could think about a solution in 7.5.
>

It would be interesting if a PL/R function could be plugged in for both
the ANALYZE function and the selectivity function. There are quite a few
spatial data related packages available for R; one of them might fit
nicely for this application.

Joe