Subject: Re: postinstall fixes failed: gid sendmail uid
To: John Nemeth <jnemeth@victoria.tc.ca>
From: Hauke Fath <hauke@Espresso.Rhein-Neckar.DE>
List: current-users
Date: 11/21/2005 01:27:39
At 15:40 Uhr -0800 20.11.2005, John Nemeth wrote:
>} >     sysinst automatically runs postinstall after unpacking etc.tgz.
>} >It doesn't automatically run etcupdate.
>}
>} Ah, okay, if you upgrade with sysinst...
>}
>} Although I don't really see the point in doing that. I mean - you already
>
>     Because it is quick and simple.

And incomplete, since both by its batch approach and by the tools that
sysinst does (and does not) provide it cannot resolve conflicts
interactively.

>} have a running system, with a toolset more complete and comfy than what
>} sysinst has to offer (run etcupdate with ed(1)? ;). And if [sysinst] for
>} whatever reason stumbles half way through the update, you get to clean -
>} no, wait, you get to find out what it has changed first, and then clean up
>} after it. BTST, BTT.
>}
>} What I do to update a system is
>}
>} (1) install new kernel, reboot
>} (2) 'shutdown now', untar all the sets except etc.tgz and xetc.tgz, reboot
>} (4) untar etc.tgz into /newetc or similar
>} (5) run etcupdate -a -l -v -b /newetc
>} (6) run postinstall -s /path/to/etc.tgz {check,fix}, reboot
>
>     Are you really expecting the average enduser to do this?

When you do it for the third time, it's a ten minutes job on a decent machine.

What is the "average enduser", btw.? Especially on NetBSD? Yes, I expect
the "average *admin*" updating a system to have an idea what's going about
in /etc - or to use a Linux OS with big, friendly, lickable buttons.

>  I've
>certainly done similar things in the past, especially if I'm updating a
>system remotely; but, it isn't something I would expect of the average
>enduser.  BTW, you forgot update the bootblocks in your list.

Not on mac68k.   ;)

And it is hardly ever necessary on other systems - definitely not in the
4-8 week rhythm I update machines @ work.

>This illustrates that one reason for using sysinst is that it does
>everything that is needed to be done.

No, it does not, since it cannot merge with conflicting local changes.

And if anything breaks, esp. on the Lesser Platforms (see
http://mail-index.NetBSD.org/tech-install/2002/07/20/0000.html for what
happened to me a while back), then you get to keep the pieces and need way
more low-level knowledge to clean up after sysinst than with the procedure
outlined above.

	hauke


--
"It's never straight up and down"     (DEVO)