[postgis-devel] RECHECK condition on geometry index scans

[Sandro Santilli]

On Fri, Feb 21, 2014 at 07:03:20AM -0800, Paul Ramsey wrote:
> Dunno how you’re still seeing it, could it be an artifact of PgSQL explain? Like, the core system says “I wonder if I need to recheck” and then after asking us (and we say “no”) it moves on, but the initial question still shows up in the explain output? We definitely say “no” in both the 2d and nd index bindings.

Spotted on a PostgreSQL 9.3.0 with a PostGIS lib "2.1.1 r12113"
and procs "2.1.0 r11822". 

The recheck seems to be only in the bitmap heap scan indeed:

  ->  Bitmap Heap Scan on tab  (cost=372.82..28794.17 rows=5441 width=4)
        Recheck Cond: ((rast)::geometry && '...'::geometry)
        Filter: _st_intersects('...'::geometry, rast, NULL::integer)
        ->  Bitmap Index Scan on tab_rast_gist_idx  (cost=0.00..372.55 rows=16322 width=0)
              Index Cond: ((rast)::geometry && '...'::geometry)

Do we have any say on that one ?

--strk;