addedge test failure

[pramsey]

This is against master postgis and master geos.

 POSTGIS="3.2.0dev 3.1.0rc1-104-g29e1db134" PGSQL="130" GEOS="3.10.0dev-CAPI-1.15.0" PROJ="8.0.0" LIBXML="2.9.4" LIBJSON="0.15" LIBPROTOBUF="1.3.3" WAGYU="0.5.0 (Internal)"

The regression seems to be an ordering thing

--- ./regress/addedge_expected	2019-11-28 14:49:43.000000000 -0800
+++ /tmp/pgis_reg/test_1_out	2021-03-01 13:49:19.000000000 -0800
@@ -17,8 +17,8 @@
 e6|6
 #770-1|7
 #770-2|8
-ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(8 10)
-ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(8 10)
+ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
+ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
 ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
 #770-1*|7
 1|0|0|-1|1|LINESTRING(0 0,8 0)

I find the topology tests completely opaque, so I am opening here.

[strk]

We should probably hide the detail of the error, which would mean changing the tests to catch the error and "massage" it before printing. This is assuming we have a double edge/edge intersection, with both old and new points intersecting, and GEOS versions behaving differently regarding to which one is output first (order...)

[robe]

Yap same issue on debbie:

13:59:16 -----------------------------------------------------------------------------
13:59:16 --- ./topology/test/regress/addedge_expected	2020-12-16 21:48:07.643162886 +0000
13:59:16 +++ /var/lib/jenkins/workspace/postgis/tmp/3_2_pg14w64/test_145_out	2021-03-02 18:59:16.058094181 +0000
13:59:16 @@ -17,8 +17,8 @@
13:59:16  e6|6
13:59:16  #770-1|7
13:59:16  #770-2|8
13:59:16 -ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(8 10)
13:59:16 -ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(8 10)
13:59:16 +ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
13:59:16 +ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
13:59:16  ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
13:59:16  #770-1*|7
13:59:16  1|0|0|-1|1|LINESTRING(0 0,8 0)
13:59:16 -----------------------------------------------------------------------------

[strk]

I can't currently build GEOS master so cannot reproduce locally. See ​https://trac.osgeo.org/geos/ticket/1103 (and the workaround to that bug exposes another build failure so it's not going to be an quick fix for me)

[strk]

It looks like this query changes output with new GEOS:

SELECT ST_AsText(ST_PointOnSurface('MULTIPOINT(8 10, 10 12)'));

Old GEOS returned POINT(8 10) while newer one returns POINT(10 12)

[Sandro Santilli <strk@…>]

In 289bb48/git:

Do not rely on ST_PointOnSurface ( MULTIPOINT )

The operation is not defined in that case, and different GEOS
version can return different values.

Closes #4864

[robe]

This needs to be backported to 3.1 and probably 3.0 as well.

Getting this error on 3.1 with GEOS 3.9.2dev

 ./topology/test/regress/addedge .. failed (diff expected obtained: /projects/postgis/tmp/3.1.1dev_pg13_geos3.9_gdal3.2.0w64/test_143_diff)
-----------------------------------------------------------------------------
--- ./topology/test/regress/addedge_expected    2021-03-09 22:31:24.954846700 -0500
+++ /projects/postgis/tmp/3.1.1dev_pg13_geos3.9_gdal3.2.0w64/test_143_out       2021-03-09 22:40:28.949824100 -0500
@@ -17,7 +17,7 @@
 e6|6
 #770-1|7
 #770-2|8
-ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(8 10)
+ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
 ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(8 10)
 ERROR:  Edge intersects (not on endpoints) with existing edge 7 at or near point POINT(10 12)
 #770-1*|7

[robe]

mislabeled last commit In 792a90a7/git:

[strk]

I'm assuming this was fixed by the referenced commit above, closing accordingly