[postgis-users] Postgis Simple Feature silently giving us wrong results

[Regina Obe]

I haven't looked thru your examples, but it would be good to know which
version of GEOS you are running since ST_Difference ultimately depends on
GEOS.

 

What does:

 

SELECT postgis_full_version(), version();

 

Return on your system?

 

Thanks,

Regina

 

From: postgis-users [mailto:postgis-users-bounces at lists.osgeo.org] On Behalf
Of Lars Aksel Opsahl
Sent: Tuesday, June 13, 2023 11:34 PM
To: 'PostGIS Users Discussion' <postgis-users at lists.osgeo.org>
Subject: [postgis-users] Postgis Simple Feature silently giving us wrong
results

 

Hi

 

In Postgis we now have problems where we silently get wrong result from
ST_Difference. This bug was only found because a tester did inspect the
result manually by actually checking the maps generated on the screen. 

 

In this case, we have a fixed restricted area (blue), where users can select
an area to use that are not already used by another user (red area), so we
need to remove the area taken. The result the user saw is here
https://gitlab.com/nibioopensource/resolve-overlap-and-gap/-/blob/develop/sr
c/test/pictures/difference_test_01_POSTGIS_3_3_3.png (output from
ST_Difference, the green rectangles). 

 

The ST_Difference test and data used can be found here
https://gitlab.com/nibioopensource/resolve-overlap-and-gap/-/blob/develop/sr
c/test/sql/regress/rog_sf_difference.sql .

 

If we use Postgis Topology we get a result that seems to be correct.
https://gitlab.com/nibioopensource/resolve-overlap-and-gap/-/blob/develop/sr
c/test/pictures/topo_difference_test_01_POSTGIS_3_3_3.png . 

 

The Postgis Topology test is here
https://gitlab.com/nibioopensource/resolve-overlap-and-gap/-/blob/develop/sr
c/test/sql/regress/rog_topology_difference.sql .

 

We use the same input data and run on the same Postgis version
(POSTGIS="3.3.3 2355e8e") and server for this tests. (If a I use latest
source compiled by Sandro a couple weeks ago on a server locally, this test
case works, but then another test case fails).

 

We have reported errors like before so the path that Simple Feature seems to
be heading now is where it silently generate very wrong results and then
silently corrupts your spatial database, is very problematic for us. It
would be much better for us if we got with a Topology exception that could
tell us what and where the problem was and then can decide what to do. 

 

Is that possible today?

 

The default behavior of Postgis Simple Feature (not only ST_Difference)
today makes it problematic for NIBIO to use it in systems where users can
apply for subsidies related to areas of land for instance. It's also
difficult to find new errors, because they are silently hidden and when
doing billions of operation, manual inspection is out off range. We can
maybe start to make tests to check if any the selected from from
ST_Difference intersects with area already used, but I have no idea if that
will find all problems. Another problem is producing spatial change logs
because with Postgis bugs like this, you get new changes just because of a
new Postgis version and not because data have changed.

 

It would have been better throwing an error in cases like this and then user
can decide if he wants to report this as error maybe modify the input data,
I think, actually one of the reasons why NIBIO started use Postgis was that
others system did cover up operations like Postgis are doing now.

 

Another option is to use Postgis Topology and the way that Sandro is doing
it here seem much better related to data corruption, because here we get an
error if the system are not able to handle a new input line. In Postgis
Topology we also have methods
(https://postgis.net/docs/ValidateTopology.html ) that can check the
database consistency.

 

NIBIO will have both a workshop and talk at Foss4g 2023 in Prizren. Here we
will talk about Postgis Topology related used cases
https://gitlab.com/nibioopensource/topology_update_app and
https://gitlab.com/nibioopensource/resolve-overlap-and-gap. 

 

 

Thanks.

 

Lars 

 

 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20230614/8131498b/attachment.htm>