Subject: Re: sysinst problems
To: NetBSD/sparc Discussion List <port-sparc@NetBSD.org>
From: Greywolf <greywolf@starwolf.com>
List: port-sparc
Date: 12/06/2004 15:49:21
** notice:  I am not, at this point, on a crusade for a knob into sysinst.
** My head has bruises from the wall, I am not seeing light yet (stars,
** sure), and the wall is still there.

I will also concede that Nathan has a very valid point about obscure
corner cases not receiving sufficient testing when controls get introduced.

[Thus spake Greg A. Woods ("GAW: ") 3:40pm...]

GAW: [ On Monday, December 6, 2004 at 01:14:18 (-0500), der Mouse wrote: ]
GAW: > Subject: Re: sysinst problems
GAW: >
GAW: > If it really is possible to use "upgrade" to install onto a blank disk
GAW: > that is labeled and partitioned but not otherwise set up - and in
GAW: > particular without any version of NetBSD installed on it - that should
GAW: > be made clearer.  Certainly the word "upgrade" implies the presence of
GAW: > an earlier version which is to be upgraded.
GAW:
GAW: Well, you'd have to manually newfs the partitions you wanted to use and
GAW: put a valid etc/fstab in the "a" partition first....

Assuming 'a' is pretty rash in and of itself, first off...

GAW: > So I used "install" and it took my existing label as a starting point
GAW: > for the partitioning step, which was good, but wouldn't let me use it
GAW: > unchanged, which was bad.
GAW:
GAW: No, that was a "Good Thing", a very good thing, and presumably by design.

Greg, you miss the points again.

And I have a Very Heavy Brick for the next time I see "Get over it!"
posted.  That must be some other language for "You really don't matter
because everybody else knows more than you do because they agree with
me.", a concept to which I ( and probably quite a few others! ) take
umbrage.

GAW: If you want to duplicate someone else's setup then you need to do so
GAW: 100%, no?  (in theory at least...)

Again, tangential, really, to the underlying point.

GAW: OK, now that I'm clear in exactly what you were trying to do I can
GAW: perhaps point out one more way to look at this issue.
GAW:
GAW: There are, as you've discovered, sanity checks in sysinst that define a
GAW: proper complete and non-crufty partition table.  Those checks did their
GAW: job and detected crufty junk in your existing label

["junk?  crufty?  my home this is!"]

GAW: -- they cannot tell
GAW: at the time they must be executed whether or not the /etc/fstab
GAW: contained within the "a" partition might specify filesystems on
GAW: overlapping partitions.

It is not "their" job to correct your crufty disklabel.  To warn about
it, sure.  To declare it a fatal error, NEVER.   i.e. "Uh, this is broken."
"Yes, I know.  Thank you." vs. "I AM REFUSINK TO BE WORKINK WITH YOUR
OBFEUSLY BROKEN DISKLABELINK BEKOS YOU ARE A TOTAL ZLOTNIK."  "...oh,
well, gee, that screws my install from this point forward."

Do you see the difference in the two concepts?

GAW: I.e. those sanity checks really cannot ever be safely turned off.

Bull pucky.

GAW: If
GAW: you want to be the expert and tell the kernel that the cruft is OK as
GAW: you won't cause any of it to come into play and cause problems then be
GAW: the expert and don't use sysinst.
GAW: However if you want to test sysinst and still save your old partition
GAW: table then save it somewhere else -- don't try to merge it with the new
GAW: one you're going to test and please especially don't ask that sysinst to
GAW: even have a non-default hidden option to allow you to use it to create
GAW: the merged label.  That's a very bogus requirement, for _sysinst_!

isn't there a cure for myopia?  Greg, you haven't given a very compelling
argument AGAINST overlapping disklabels.  Fine, I'll go use something else
besides sysinst [which I will admit, by the way, is a fine installer!],
but I still see NO damn reason that sysinst should totally disallow an
overlap.  Warning?  YES, by all means!  I've had stupid overlaps of the
front end of a filesystem by the previous one.  A warning at that point
would have been nice (of course, I find it much more engaging to check the
stuff by hand these days anyway, and double check just in case!).  But to
say "I'm not going to let you do this because you cannot convince me
that you know what you are doing" is just crockery.

And to sit and repeat the mantra of "Best Practices" doesn't ALWAYS FIT
every situation, and allowances do need to be made for this.

And for something better to happen to sysinst, yes, I agree there is a
more compelling change for something more meaningful -- such as being
allowed to spread the system across multiple disks [this is apparently
a hard problem].

GAW: Now of course if you use the "install to disk" option you're also going
GAW: to tell sysinst what partitions to newfs and to mount and install to,
GAW: and presumably it could later figure out that it would be OK to leave
GAW: the overlapping parititons in place but as I keep saying that's just an
GAW: accident waiting to happen.

You'd say a soldering iron that was plugged in was an accident waiting
to happen, by that same reasoning.  The difference here is that the
soldering iron isn't going to ask you if you really want to grasp it by
the hot end before allowing you to do so, but I'm sure you're going to point
out that this is a mere technicality.

GAW: However from a "Good Housekeeping"
GAW: perspective the in-use partition table should only contain the in-use
GAW: partitions, and it should never include overlapping paritions since they
GAW: cause the device driver to set up device nodes that, if accessed, would
GAW: conflict with other device nodes and at that point the system cannot
GAW: prevent an accident -- it has given you the rope and it expects you to
GAW: hang it and/or yourself.  :-)

Uh, no, it does NOT expect you to "hang yourself".

It doesn't care if you do or don't, and it goes out of the way
neither to prevent it nor to cause it.

GAW:  You can play around and experiment with
GAW: such configurations if you're willing to take the risks, but "sysinst"
GAW: isn't intended to facilitate such risks -- it's designed to mitigate
GAW: them and to reliably create a working, safe, and secure, install.

...which, BY DEFAULT, it should.  [The non-default cases and presentations
are left to the imagination...for those of you that have them.]

				--*greywolf;
--
Microsoft:
	"Just click on the START button and your journey to the Dark Side
	 will be complete!"