32-bit failure and rounding issues on ST_LineSubstring for geography

[robe]

See test ​https://trac.osgeo.org/postgis/browser/git/regress/core/geography_interpolate.sql?rev=dfd5ff50b393dd98d44bf6f92f681121efd51697

The test failing is:

SELECT round(MAX(ST_Length(ST_LineSubstring(g, 0.5, 0.7)))::numeric, 6) FROM tbl_geog_linestring3D;

Errors on gitlab and berrie 32-bit cis (though bessie32 is fine)

{{{
06:20:37  regress/core/geography_interpolate .. failed (diff expected obtained: /tmp/pgis_reg/test_115_diff)
06:20:38 -----------------------------------------------------------------------------
06:20:38 --- ./regress/core/geography_interpolate_expected	2023-07-29 04:11:12.380004361 -0600
06:20:38 +++ /tmp/pgis_reg/test_115_out	2023-07-29 04:20:38.108382789 -0600
06:20:38 @@ -27,6 +27,6 @@
06:20:38  4180892.366508
06:20:38  297
06:20:38  1
06:20:38 -5520157.913043
06:20:38 +inf
06:20:38  4790369.580943
06:20:38  3215502.296386
06:20:38 -----------------------------------------------------------------------------
}}}

 gitlab 64-bit is also failing, but that looks to be a rounding difference

{{{
--- /builds/postgis/postgis/regress/core/geography_interpolate_expected	2023-07-29 10:09:11.814967894 +0000
+++ /tmp/pgis_reg/test_115_out	2023-07-29 10:14:05.882460007 +0000
@@ -27,6 +27,6 @@
 4180892.366508
 297
 1
-5520157.913043
+5520157.912684
 4790369.580943
 3215502.296386
-------------------
}}}

[pramsey]

What are you even testing? I do not see a core/geography_interpolate in my master branch. All the linear referencing tests I see are in core/geography

[robe]

Yah as mentioned on matrix, I took those tests shortly after I added them to prevent the 32-bit bots and gitlab trixie 64-bit from failing on them. I can put them back after I flip the 3.4 switch. Just didn't want to hold up release for 32-bit issues.