Am Freitag, 17. März 2006 15:25 schrieb Tom Lane:
> I will grant you that the contrib script ought to default to assuming
> installation paths under /usr/local instead of where the RPMs put
> things, but beyond that I'm not sure what's unportable in the current
> RPM init scripts.
Assuming that /etc/rc.d exists. (fails on Debian)
Assuming that /etc/rc.d/init.d/functions exist. (fails on Debian and SuSE)
(Later on I see a number of functions I don't recognize that probably come
from here.)
Assuming that a command named "typeset" exists. (fails on Debian and SuSE)
Assuming that a file named /etc/sysconfig/network exists. (fails on Debian and
SuSE)
Location of data and log directories are handled differently on other systems.
Various bashisms. Yes, I've seen people use the contrib script on systems
where bash != sh.
Assuming that /var/lock/subsys exists. (fails on Debian)
The one thing that really bugs me about this init script is that it tries to
run initdb if it doesn't find the data directory. This is really trouble on
the type of HA systems that people (should) use. I don't want that sort of
thing in the PostgreSQL source distribution.
So, if there is a problem with the start script in contrib, let's fix it, but
I think it's obvious that the Red Hat RPM start script should remain just
that.
--
Peter Eisentraut
http://developer.postgresql.org/~petere/