Leo Bicknell drops some science:
> 	All programs should write out /var/run/programname.pid (or
> in some other location, OSF/1 uses /var/run).  You should be able
> to do something like "kill -HUP inetd", which kill would intrepret
> as "kill -HUP `cat /var/run/inetd.pid`".

Better yet, how about replacing "kill" with something like Jeff Forys'
"skill" (Super KILL), which, of course, can do just that - "skill -HUP inetd".
(Check out "skill" at ftp://jaguar.CS.Utah.EDU/pub/skill/skill-3.6@P3_tar.Z)

> 	Basically I think the whole run level thing is too complicated.
> 99.9% of the time your system is fully up and running (at least with
> most people), and the other 0.1% of the time you probably want
> single user mode to do maintance.  Why add all sorts of other functionality
> that will simply confuse many people, and cause extra work for others?

Amen, brother.

Just because something is "flexible"/"extensible"/etc. doesn't mean it's the
right way to go, especially if you provide a general functionality that can
handle 10-20 cases but hardly anybody uses more than a subset of 3-4 of them.
(Here's my run states: "Off", "Single User", and "Up".  If I want to kill off
 NFS, I'd say "skill -TERM nfsiod nfsd mountd".)

To put it another way: this inittab/rc?.d stuff has been around in System V
for what, oh, 12 years now or so?  If it's such a great idea, how come we're
already to 4.4BSD++ and CSRG never felt the compelling need to implement
these oh-so-wonderful features into BSD?  I remain unconvinced ...

	- Greg