PostGis 3.1.4 fails to test (mvt) successfully on ppc64el in Ubuntu 22.04

[cpaelzer]

Hi, As reported in Ubuntu [1] and today on IRC we have trouble building postgis 3.1.4 on Ubuntu right now. What I see is an issue in mvt:

./regress/core/mvt .. failed (diff expected obtained: /tmp/pgis_reg/test_110_diff)
-----------------------------------------------------------------------------
--- ./regress/core/mvt_expected 2021-09-04 05:03:43.000000000 +0000
+++ /tmp/pgis_reg/test_110_out 2021-11-20 06:56:10.856917815 +0000
@@ -138,7 +138,7 @@
 #4348Clip|t
 #4348Invalid|t
 #4348Dropped|t
-#4348ReversedSmall|t
+#4348ReversedSmall|
 #4348Reversed2|t
 #4348Point|t
 #4348Crash|t
--------------------------------------------------------------------------

Earlier this year 3.1.1 built fine [5] in Ubuntu. So did 3.1.4 in Debian in January of this year [6].

I fail to see/understand what is broken here and was hoping that you could make sense of it.

I failed to recreate the issue on a ppc64el machine. With the same SW versions of build dependencies installed I got it working:

./regress/core/mvt .. ok in 279 ms

But OTOH in the launchpad build environment [2][3] it is 100% reproducible.

The Debian build already ignores test issues on many non-x86 platforms and I guess for now I'll need to add ppc64el to that list. But if mid-term there would be a real fix that would be great.

In the log [4] you'll also find the postgresql server errors, but usually - since some tests trigger errors intentionally - those logs are not very helpful.

Thanks in advance. I'd even be happy to get hints at what might affect this, is it very sensible to lib versions, memory or cores available at run time or anything else that could be different from the official Ubuntu build environment and my try to recreate? What would usually affect this bug?

[1]: ​https://bugs.launchpad.net/ubuntu/+source/postgis/+bug/1952604 [2]: ​https://launchpad.net/ubuntu/+source/postgis/3.1.4+dfsg-3build1 [3]: ​https://launchpad.net/ubuntu/+source/postgis/3.1.4+dfsg-3build1/+build/22346629 [4]: ​https://launchpadlibrarian.net/571347323/buildlog_ubuntu-jammy-ppc64el.postgis_3.1.4+dfsg-3build1_BUILDING.txt.gz [5]: ​https://launchpad.net/ubuntu/+source/postgis/3.1.1+dfsg-1 [6]: ​https://buildd.debian.org/status/fetch.php?pkg=postgis&arch=ppc64el&ver=3.1.4%2Bdfsg-3%2Bb1&stamp=1634938635&raw=0

[strk]

Referenced ticket is #4348 - it may be a good idea to rewrite the test to show the actual area when it is NOT in the tolerated range

[robe]

Is this an issue also in 3.0 or just 3.1.6 and up?

[robe]

Bas,

I forget is ppc64el emulatable with a VM or doesn't work, need to have a real ppc64el?

[Bas Couwenberg]

QEMU should be able to emulate ppc64el:

• ​https://wiki.debian.org/ppc64el/Installation#debootstrap_for_QEMU.2FKVM
• ​https://etbe.coker.com.au/2020/07/14/debian-ppc64el-emulation/

[robe]

Thanks. Will try to take a look at this after release.

[pramsey]

User patch accepted, but I do not think any core team member is going to try and track this one down, the overhead of getting an environment up and running is too high relative to the effected population of users.