I'm sorry to keep bringing this up, but I keep thinking that PostgreSQL is
a PITA to configure. It isn't that I don't know how, or can't read the
documents, it's just that there is too many things that can go wrong or
leave me with the "How does that work?" feeling where I need to test
before I implement.
OK, here is what I think we need to do, and I am willing to take the point
and do all of it:
If postmaster is started with -D or with $PGDATA set to the location of a
postgresql cluster directory, it will work as it has always worked.
If the -D or the $PGDATA values do not point to a database cluster
directory, it must point to either a postgresql.conf file, or a directory
which contains "postgresql.conf." If it points to a directory, then that
directory will be the assumed the default for pg_hba.conf and
pg_ident.conf.
The test for it being a cluster directory or a configuration directory
will be the presence of "PG_VERSION." (any better suggestion would be
welcome
Within the postgresql.conf file, there will be about 4 additional
configuration parameters:
include='/somepath/coreopts.conf'
hba_conf='/somepath/pg_hba.conf'
ident_conf='somepath/pg_ident.conf'
pgdata='/somepath/to/cluster'
(Unfortunately we lose the ability to use a standard configuration across
multiple databases, but I think this is a good compromise.)
Additionally, rather than depend on environment variables for security, we
can put acceptable database locations in the postgresql.conf file in the
form of:
altdb='/somealtpath'
and we can use createdb as:
CREATE DATABASE somedb with LOCATION 'altdb';
What do you all think?
