James Carlson wrote:
Eric Boutilier writes:
... The alternative is to
nuke /opt from the system and just install everything under /usr...
Hmm, Sun service would not like it (to put it mildly), but other than
that it sounds worth considering if you ask me. So James, correct me if
I'm wrong... the following is a sample future scenario that I think
you're suggesting with a /usr solution...?
Actually, I was proposing that as a bit of hyperbole. I can't imagine
someone taking it seriously because I believe that in that direction
lay both architectural and support madness.
Fink and Debian don't suffer from architectural and support madness
though...
A user/customer installing a system decides they want an unsupported
freeware (ports) stack on their box. They check opensolaris.org to see
which stacks have petitioned for and received an endorsement from the
opensolaris community. They install the stack in /usr, and everyone
(except Sun service :-/ ) is happy...
That's exactly what it would mean, and it's essentially what I think
falls naturally out of the Fink/Debian placeholder-module mechanism.
If you have two or more installable features that can satisfy the
abstracted dependencies for other features on the system, then the
implication is that you can choose among them.
If you can't actually choose among them, if it damages the system or
costs the user supportability, then why bother having the mechanism at
all? It's either an unnecessary degree of freedom, or an accident
looking for a good place to happen.
Same thing here. Debian's and Fink's success seems to contradict the
assertions in the above paragraph.
There might be some limited cases where we can claim that this is
marginally doable (would you like /usr/lib/sendmail to be sendmail or
postfix?) but it's really unclear to me whether it's the right model
in general.
A purist would argue that that pollutes /usr, but 99% of users just care
that things work and are easy to upgrade/maintain. Besides, the purist
is perfectly capable of rolling their own.
True. Determining, though, what is and is not supported, and just how
any poor implementor can ever test the pi(f(X)) supported combinations
is an exercise for the reader. :-/
Yeah, the support ramifications is what I fear most. Although Darren M.
has a good point about the pkg tools helping a lot here.
Eric
_______________________________________________
opensolaris-discuss mailing list
opensolaris-discuss@opensolaris.org
