[postgis-users] Bizarre behavior with postgis function as part of where clause!

[Regina Obe]

Have you tried adding a MATERIALIZED to your WITH?  In PostgreSQL 12 and above, CTEs are no longer always materialized.

 

e.g.

 

WITH foo AS MATERIALIZED (SELECT ST_GeometryType(geom) gtype, geom FROM tmp_test_geoms WHERE ST_GeometryType(geom) = 'ST_Point') SELECT * FROM foo WHERE ST_X(geom) > 1;

 

This article describes it a bit

 

https://paquier.xyz/postgresql-2/postgres-12-with-materialize/

 

From: postgis-users [mailto:postgis-users-bounces at lists.osgeo.org] On Behalf Of Bo Guo
Sent: Friday, September 4, 2020 11:28 PM
To: postgis-users at lists.osgeo.org
Subject: [postgis-users] Bizarre behavior with postgis function as part of where clause!

 

Hi List,

I just ran into a pretty bizarre query behavior involving simple queries involving PostGIS functions as part of where clause condition.  

I have two geometries in table tmp_test_geoms , one point and a line: 

    ST_Point    "0101000000806FE29BC53B2241800FDDE3E9C53341"
    ST_LineString    "010200000005000000E072B95C215A22412033C40C65963341401C0E074D5A22412097CB056E963341809821E6235E22412033C48C6D963341A027E909BA5F22410087C3E16E963341C0B2AEACD95F224100FCFDFE71963341"

Both Query 1 and Query 2 fails w/ message "Argument to ST_X() must have type POINT," Query 3 works however

    1. SELECT ST_GeometryType(geom) gtype, geom FROM tmp_test_geoms WHERE ST_GeometryType(geom) = 'ST_Point' AND ST_X(geom) > 1;

    2. WITH foo AS (SELECT ST_GeometryType(geom) gtype, geom FROM tmp_test_geoms WHERE ST_GeometryType(geom) = 'ST_Point') SELECT * FROM foo WHERE ST_X(geom) > 1;
    3. WITH foo AS (SELECT ST_GeometryType(geom) gtype, geom FROM tmp_test_geoms WHERE ST_GeometryType(geom) = 'ST_Point' LIMIT 10) SELECT * FROM foo WHERE ST_X(geom) > 1;

It seems that the query engine evaluates ST_X first before applying the GeometryType filter, even when the filter is in a CTE.  

It further seems that by adding LIMIT with an arbitrary value at the end of the CTE, the filter is applied first!

Wonder if there is a way to give some type of query hint to force the evaluation sequences.

Thanks for any suggestions!

Bo

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20200905/ed75de2c/attachment.html>