Oliver - by way of introduction, I'ved work to help Lamar a little from
time to time with the RedHat PostgreSQL packages, and volunteered to do
so again with respect to Fedora and most especially your proposal here.
So to get myself up to speed, I was going through that proposal again.
My basic reaction is still the same as when I first saw it on the
postgresql mailing lists - it's a great approach, in many ways brilliant
in it's simplicity. So I'm pretty enthusiastic.
As I see it there are 3 basic parts:
1) making the packages non-conflicting
2) being able to choose which package to run
3) migrating data from one instance to another
The file layout stuff is not too hard, it will be different on RedHat
than debian, but you've written down the parts that come to mind, they
just might have slightly different names on RedHat.
Migration is more or less dependent on how you choose which package to
run, so I won't touch on that for the moment.
As I was reading it this time, what struck me was that a lot of what you
propose to do could be done with the alternatives system. So I was
wondering if you had considered that and found a fatal flaw. The
advantage is twofold - less code to write and no new interfaces for
sysadmins to learn. But there are a few points at least, even on first
glance, where the fit is not quite perfect.
I think I may spend a little time trying to see for myself exactly how
good or bad the fit is. But if you have already do this and decided it's
a bad idea, it'd be great to get some sense of your reasoning rather
than working it out from scratch on my own. Especially of note, I am
only familiar with RedHat's impementation, so even though the concept
comes from debian, there may be differences.
--
Karl DeBisschop <kdebisschop@alert.infoplease.com>
