can't upgrade from 2.1.1 to 2.1.2 with extensions

[robe]

I don't have my 2.1.1 readily installed but I tried running this:

ALTER EXTENSION postgis UPDATE TO "2.1.2devnext";

on a 2.1.1 I had upgraded earlier to 2.1.2dev.

I got error

operator = already exists

I'll try going from 2.1.1 to 2.1.2 next once I have that setup again.

strk, pramsey can you check. Sorry for holding up the curtain call.

[strk]

Ouch, stop the release wheels pramsey !!!

[strk]

Regina, but… shouldn't that kind of upgrade be the same that happens on "make check" ? Or had you tried ./run_test.pl —extensions ?

[strk]

Confirmed !

[strk@cdb:/usr/src/postgis/postgis-2.1/raster/test/regress(svn-2.1)] ../../../regress/run_test.pl -v --upgrade --extension --raster tickets.sql 
PATH is /home/postgresql-9.3.1/bin:/home/postgresql-9.3.1/bin:/home/strk/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/bin:/usr/games:/usr/local/games://home/strk/bin://home/strk/bin
Checking for shp2pgsql ... found
Checking for pgsql2shp ... found
Checking for raster2pgsql ... found
TMPDIR is /tmp/pgis_reg
Creating database 'postgis_reg' 
Preparing db 'postgis_reg' using 'CREATE EXTENSION'
Upgrading PostGIS in 'postgis_reg' using 'ALTER EXTENSION'

Error encountered altering EXTENSION POSTGIS, see /tmp/pgis_reg/regress_log for details

[strk@cdb:/usr/src/postgis/postgis-2.1/raster/test/regress(svn-2.1)] cat /tmp/pgis_reg/regress_log 
createlang: language "plpgsql" is already installed in database "postgis_reg"
CREATE EXTENSION
ERROR:  operator = already exists

Regina: could you add this step to the buildbots ?

make check RUNTESTFLAGS=--extension

Is only supposed to work _after_ "make install"

[strk]

Regina: the —extension switch to run_test.pl currently fails due to some tests being ambiguous when run in an extension environment due to raster being also forceably loaded. Those tests could be tweaked to be immune…

[robe]

confirmed same issue going from 2.1.1

ALTER EXTENSION postgis UPDATE TO "2.1.2dev";

I have on my list to add to bots but haven't already. Though wondering why doesn't the regular upgrade test fail when we go from 2.1.2 to 2.1.2 (the regular way should fail as well unless its not running in a transaction)

[robe]

Ah I see why it fails for extensions but not for regular script. For regular there is still a rtpostgis_upgrade_21_minor.sql which is different than rtpostgis_upgrade_20_21.sql.

Since extension always uses the rtpostgis_upgrade_20_21.sql, it fails.

[strk]

I think you should change the Makefile to use 21_minor.sql for 2.1.x to 2.1.x extension upgrades. Will leave that to you as you were the original author of that Makefile, right ?

In trunk it'll be easier as the upgrade script is one for all minor upgrades.

[pramsey]

Please test the patch, please.

[pramsey]

Does

make check RUNTESTFLAGS=--extension

work for you? it does not for me, even with 2.1.1.

up=# SELECT 'contains100', ST_contains('POLYGON((0 0, 0 10, 10 10, 10 0, 0 0))', 'POINT(5 5)');
ERROR:  function st_contains(unknown, unknown) is not unique

I've confirmed this isn't a problem with the regression test, it's a problem with the extension install, by running one of the regression queries (the above one) manually on a extension-enabled database. There seem to be lots of others of these type-recognition errors for extension installs.

[strk]

Does not work for me, as stated in http://trac.osgeo.org/postgis/ticket/2684#comment:5

[pramsey]

Patch tommitted to 2.1 branch at r12371

[strk]

Still fails for me with r12371:

postgis_reg=# create extension postgis version '2.1.2dev';
CREATE EXTENSION
postgis_reg=# alter extension postgis update to "2.1.2devnext";                                                        
ERROR:  operator = already exists

[strk]

r12372 fixes that (tweaking the rule to produce the "next" version, used by run_test.pl on —extension —upgrade)

[pramsey]

Confirmed, works for me.

[robe]

Looks good. Just doing garden battery tests to confirm 2.1.1 to 2.1.2 and 2.0.4 to 2.1.2 yield same answers.

[robe]

So far did a garden test of 2.1.1 upgraded to 2.1.2 r12371, compared to the backup of test I created going straight with create extension 2.1.2 r12366.

To my shock I saw differences, but don't think they have anything to do with upgrade. I guess I'll need to retest 2.1.2 for sure. Differences are with

ST_AsGeoJSON (I assume as result of fix pramsey recently did) so not worried about it.

The one that has me a bit concerned (but not concerned enough to hold a release) are with ST_3DExtent, ST_AsLatLonText. I think these are just instability issues of some sort which I think we have related tickets for already.

To do a spot check — I'm just doing a JOIN pg_column_size(postgis_garden_log20_output.log_output) check and also stole some tricks from dustymugs and repurposed his raster_hash to create an xml hash to catch other oddities.