tickets test fails on non-amd64 architectures

[Bas Couwenberg]

The Debian package build for 3.0.3 failed on some architectures due to the tickets test failing:

arm64 (​full buildlog)

 tickets .. failed (diff expected obtained: /tmp/pgis_reg/test_102_diff)
-----------------------------------------------------------------------------
--- tickets_expected	2020-11-21 00:48:34.000000000 +0000
+++ /tmp/pgis_reg/test_102_out	2020-11-21 07:10:15.648695071 +0000
@@ -433,5 +433,5 @@
 #4670-3|LINESTRING(0 0,1 1,3 3,2 2,4 4)
 #4670-4|LINESTRING(0 0,1 1,3 3,4 4,2 2)
 ERROR:  LWGEOM_addpoint: Invalid offset
-#4689|0
+#4689|5.92515031139973e-11
 #4727|0
-----------------------------------------------------------------------------

ppc64el (​full buildlog)

 tickets .. failed (diff expected obtained: /tmp/pgis_reg/test_102_diff)
-----------------------------------------------------------------------------
--- tickets_expected	2020-11-21 00:48:34.000000000 +0000
+++ /tmp/pgis_reg/test_102_out	2020-11-21 07:10:15.648695071 +0000
@@ -433,5 +433,5 @@
 #4670-3|LINESTRING(0 0,1 1,3 3,2 2,4 4)
 #4670-4|LINESTRING(0 0,1 1,3 3,4 4,2 2)
 ERROR:  LWGEOM_addpoint: Invalid offset
-#4689|0
+#4689|5.92515031139973e-11
 #4727|0
-----------------------------------------------------------------------------

[pramsey]

Does building with '-ffp-contract=off' make it go away? We've seen differences between calculations with small tolerances in GEOS on arm64 because ARM is being slightly clever about floating point math.

[Bas Couwenberg]

We have since moved to 3.1.0 whose tests succeed on these architectures.

Unless @myon intends to provide 3.0.x for these archs on apt.postgresql.org I don't think we should worry about it.

[pramsey]

Aha. Interesting and odd. I also could not replicate in 3.1/master, and annoyingly am blocked on building 3.0 due to various old libraries on this AWS instance. Close it out if it's not something you see as important.