On 9/23/15, Landry Breuil <[email protected]> wrote: > On Tue, Sep 22, 2015 at 10:28:37PM -0700, patrick keshishian wrote: >> On 9/19/15, Landry Breuil <[email protected]> wrote: >> > On Fri, Sep 18, 2015 at 11:50:19PM -0700, patrick keshishian wrote: >> >> On 9/17/15, Landry Breuil <[email protected]> wrote: >> [...] >> >> Attached is new tar-ball that compile, but unfortunately "make test" >> >> mostly fails[1]. >> > >> > Yes, but at least it builds :) Lots of ports have failing tests in the >> > tree.. >> > >> >> Most failures are due test DB not existing followed by core dump. >> > >> > Hmm, maybe have a look at databases/postgresql MODULE and the ports >> > that >> > use it, the module provides helpers to run/create databases for >> > testing. >> > >> >> One hints of bad option. >> > >> > This one is strange.. >> > >> >> One on Python module import. >> > >> > Needs TEST_DEPENDS on databases/py-psycopg2 >> >> After adding this, the tests which were FAILing change state to >> SKIP. The error message relating to their SKIP state, indicated >> PG template0 database encoding was ASCII vs expected UTF8. >> >> Patching databases/postgresql/postgresql.port.mk[1] to accept >> an encoding to be specified in port's Makefile. Now more tests >> PASS: >> >> # TOTAL: 18 >> # PASS: 6 >> # SKIP: 1 >> # XFAIL: 0 >> # FAIL: 11 >> # XPASS: 0 >> # ERROR: 0 >> >> Those which FAIL claim "Out of memory" (see attached test-suite.log). > > What if you bump ulimit -d ? > >> New tar-ball with suggested changes from Landry: >> - Get rid of AUTO_ENV >> - MODULE=gcc4 sets up proper links in ${WRKDIR}/bin for c++, cc, >> g++, gcc. >> - TEST_DEPENDS = databases/py-psycopg2 >> >> My "hack" for ensuring test-PG database uses UTF8 encoding: >> - MODPOSTGRESQL_TEST_PGENCODING = -E UTF8 > > I'll let vadim comment on this one but i think it should just be the > default in postgresql.port.mk, i'm not sure we need a new knob for this. > Few ports use this module, and they probably are all tested with UTFcrap > anyway.
If the goal is to bring the tests results closer to all-PASS, the test DB needs to be initdb-ed with '-E UTF8" option. --patrick
