Subject: Re: Updating /etc...
To: der Mouse <mouse@Collatz.McRCIM.McGill.EDU>
From: Ted Lemon <mellon@fugue.com>
List: current-users
Date: 12/21/1995 12:42:53
[Okay, before you /etc/rc vs. /etc/init.d flamers respond to this,
please read it. Please don't respond with something to the effect of
``I think that /etc/init.d is a better idea than /etc/rc, or vice
versa'' We already know you think that. Whichever opinion you
hold, you're not wrong. This is a matter of personal preference.
There are advantages to both. Now, back to our regularly sponsored
programming.]
Hm. I guess I'm with Chris on the notion that this shouldn't be a
toy operating system, so we do want it to work for Large
Installations. I will grudgingly admit that /etc/rc probably isn't
right for Large Installations with heterogenous workstation
configurations.
I'm with der Mouse on the notion that the startup script that gets run
by init should be a single script - /etc/rc. One could argue that we
need a second script, /etc/rc.single, to get things into a manageable
state before coming up into single user mode, but I do not like the
Sun solution of having /etc/rc.single check and mount all the ufs
filesystems. There is never a need to be multi-user when the network
is down. As a consequence, I don't think that the run level model is
particularly useful. I do think there *is* a need for an
/etc/rc.shutdown. I think that's about it.
I would like to have a system that supports both Chris's goal and der
Mouse's. I'd also like to make it trivial for a naif to administer,
something which neither /etc/rc nor /etc/init.d accomplishes. I
presented a solution to this a couple of messages ago. Apparently it
went over like a lead balloon.
Anyway, the point of this message is that I think neither /etc/rc nor
/etc/init.d has much to recommend it, although both encompass some
very good ideas. I think that a fusion of the two, perhaps along the
lines I suggested, or perhaps less complicated than that, is the right
solution. A clever fusion of this functionality would be compatible
with a standard SysV install. I suspect that this would make
everybody happy.
This approach would be a lot of work compared to either sticking with
the /etc/rc model or switching to the SysV model. However, as
several people have pointed out, we could probably get there in
incremental steps.
The one important thing from my perspective is that I think it's a
*bad* idea to put this functionality into init. Let's keep init
simple. I think diverting down the path of making init more
SysV-like would be a mistake, even if we decide that we want similar
functionality to what SysV init provides.
_MelloN_