Remove RECHECK requirement on index

[pramsey]

​http://postgis.refractions.net/pipermail/postgis-users/2007-September/017036.html

The utility of the SRID recheck is low, and more performance is always desirable.

[pramsey]

<i>(No comment was entered for this change.)</i>

[pramsey]

Removed recheck from opclass definition, and from consistent test for 8.4+, in r3514.

[pramsey]

<i>(No comment was entered for this change.)</i>

[mcayland]

Excellent. One of things I've been thinking is that with this change (and also potential alignment changes after 1.4), how do we benchmark this to show it actually is an improvement? At the moment, I don't think there is anything in the regression tests to do this…

ATB,

Mark.

[pramsey]

Regressing for performance is practically impossible, as far as I know. If we were a huge organizations, we would maintain a set of older versions we could run a performance regression suite on, but we're not, and I cannot think of any general purpose way to build performance improvement testing into our regression suite so it could make a meaningful determination with just one copy of the software.

[pramsey]

An individual data point:

select sum(v.ndp) from vada2005 v join ed2000 e on (st_intersects(v.the_geom, e.the_geom));

— recheck = false

Time: 4101.556 ms Time: 4027.646 ms Time: 4013.932 ms Time: 4082.649 ms

— recheck = true

Time: 5312.517 ms Time: 5419.429 ms Time: 5293.803 ms Time: 5219.899 ms