New regresion test for SET/RESET commnad

Zdenek Kotala <Zdenek.Kotala@Sun.COM> writes:

I created new regression test for check SET and RESET commands and
configuration subsystem.

Does this actually accomplish anything? The checks that retail SETs
and RESETs work seem redundant with numerous existing tests. The SELECT
from pg_settings is a seriously bad idea: it adds 100K of bloat to the
regression files, tests nothing of great interest, and will cause
regression test failures under a large variety of scenarios (eg, initdb
chose less-than-max shared memory settings, or we're doing "make
installcheck" against an installed server with even one nondefault
parameter setting). Not to mention the extra maintenance effort of
updating this expected file anytime anyone changes guc.c at all,
even to the extent of fixing grammar in a GUC-variable description.

regards, tom lane

Tom Lane wrote:

Zdenek Kotala <Zdenek.Kotala@Sun.COM> writes:

I created new regression test for check SET and RESET commands and
configuration subsystem.

Does this actually accomplish anything? The checks that retail SETs
and RESETs work seem redundant with numerous existing tests.

I think no. I have not found any test specialized to GUC subsystem. Some
tests use SET/RESET command but only in same situation. I want to test
"complete" GUC subsystem. My patch is first step.

The SELECT
from pg_settings is a seriously bad idea: it adds 100K of bloat to the
regression files, tests nothing of great interest, and will cause
regression test failures under a large variety of scenarios

Yes, you have right. My idea was check if GUC subsystem setups all
settings correctly, but this test should bring more problems than
advantages. I probably take subset ( one for each datatype) of relative
stable settings, and perform checks on this set. However, this test
require be first of all regression tests.

Zdenek