ST_ASMVT regress on PostgreSQL 10

[robe]

Bjorn,

I guess you can't easily check this until we have PostgreSQL 10 at least being able to compile PostGIS, as a result of #3271.

But here it goes, Debbie's last run in 10.0. MVT is just one of other problems with have with 10.0.

 mvt .. failed (diff expected obtained: /var/lib/jenkins/workspace/postgis/regress_pgdev/tmp/2_4_pg10.0w64/test_119_diff)
-----------------------------------------------------------------------------
--- mvt_expected	2017-02-06 16:48:10.479047203 +0000
+++ /var/lib/jenkins/workspace/postgis/regress_pgdev/tmp/2_4_pg10.0w64/test_119_out	2017-02-23 20:05:58.050464234 +0000
@@ -14,7 +14,7 @@
 TA4|GjAKBHRlc3QSDBICAAAYASIECTLePxIMEgIAARgBIgQJMt4/GgJjMSICIAEiAiACeAI=
 TA5|GiwKBHRlc3QSDhIEAAABARgBIgQJMt4/GgJjMRoCYzIiAiABIgYKBGFiY2R4Ag==
 TU1
-ERROR:  could not determine polymorphic type because input has type "unknown"
+ERROR:  could not determine polymorphic type because input has type unknown
 TU2
 ERROR:  pgis_asmvt_transfn: parameter row cannot be other than a rowtype
 TU3
-----------------------------------------------------------------------------

You might have even fixed this already. Can't tell till we fix up 10.0.

[Björn Harrtell]

This looks like an error message from PostgreSQL that has slightly changed in version 10.

[strk]

Ok this is tricky. In the past we used two _expected files and copied the appropriate one based on version (was done for GEOS). Annoying because then you have to remember to update both when adding new tests, but effective.

[robe]

alternatively you could change the test to not error. I'm sure there is a perfectly good reason for triggering this error.

[Björn Harrtell]

Yep it's a valid test for an expected error. I'll use the solution with two expected files if not too messy.

[robe]

warning, I don't think we've ever done that on PostGIS. We usually just strip out the NOTICE content for example. I have no idea what strk is talking about about having done that with GEOS, though he's probably right. You are a very brave soul

[Björn Harrtell]

Not that brave I thought it was for GEOS stuff in PostGIS. I'd rather find a solution without expanding on the regression test methodology. Can you point me to an example where you "strip out the NOTICE content"?

[Björn Harrtell]

In 15320:

Disable test that is too PG version specific
Fixes #3722

[Björn Harrtell]

I opted to comment out this test as the error messages are intentionally changed in PG 10. Would need PL/pgSQL test runner to fix this to test against real error codes.

Relevant change in PostgreSQL is ​https://github.com/postgres/postgres/commit/9a34123bc315e55b33038464422ef1cd2b67dab2.