IPU's vs. Fresh Installs

[tony@aplawrence.com]

We've had a fair amount of discussion of this recently, but it has been
mushed into to a partially unrelated thread, so I think it might help
to make a fresh start.

First, let's correct a misunderstanding that Bela had:

The majority of my work is NOT with clients I've had a long term
relationship with. I'm much more apt to be working on systems I have
never seen in my life, systems whose provenance is unknown to me and
perhaps even to the owners.

In fact, if I am the person who originally installed and has since
administered a system, I am actually much more favored toward an IPU,
because in that case I know what modifications and configuration changes
have been made.

It's interesting that this seems to be quite contrary to what Bela
suggested (that it's easier to do IPU's on unfamiliar systems).

Another misconception is that my concern is with catastrophic failures.
Actually, that's not the case at all: those are rather unimportant.
It's the nasty little glitches, the unexpected inconsistencies that come
from preserving the wrong stuff that concerns me. Those problems are
much more difficult to root out, and in fact may not exhibit immediately.

At [url]http://aplawrence.com/Unixart/ipuvsfreshinstall.html[/url] , I've posted my
reasons for preferring fresh installs, and suggested that I would like
IPU's a lot better if they would produce a full directory hierarchy diff
of the upgraded system vs. the pre-upgrade system (or against what a
fresh install would have produced), and a diff of all files preserved
or merged against the fresh versions that otherwise would have been
installed.

It seems to me that would be a pretty simple thing to add to the IPU code,
and it would have tremendous benefits. It would be easy to fix any mods
that should not have been applied, easy to spot crapola that you don't
need or want anymore, and in fact easy to "roll back" to as close to
"fresh" as you needed to go to get it right.

Until that happens though, I'll do fresh installs because I then can
generate my own diffs and from that decide what to copy, merge, hand
configure or ignore.

--
[email]tony@aplawrence.com[/email] Unix/Linux/Mac OS X resources: [url]http://aplawrence.com[/url]
Get paid for writing about tech: [url]http://aplawrence.com/publish.html[/url]

[Bela Lubkin]

[email]tony@aplawrence.com[/email] wrote:

> We've had a fair amount of discussion of this recently, but it has been
> mushed into to a partially unrelated thread, so I think it might help
> to make a fresh start.
>
> First, let's correct a misunderstanding that Bela had:
>
> The majority of my work is NOT with clients I've had a long term
> relationship with. I'm much more apt to be working on systems I have
> never seen in my life, systems whose provenance is unknown to me and
> perhaps even to the owners.
>
> In fact, if I am the person who originally installed and has since
> administered a system, I am actually much more favored toward an IPU,
> because in that case I know what modifications and configuration changes
> have been made.
>
> It's interesting that this seems to be quite contrary to what Bela
> suggested (that it's easier to do IPU's on unfamiliar systems).

I'm not saying either is generally "easier", I'm saying that in some
cases an IPU will preserve some important detail(s) you would otherwise
have a hard time finding.

> Another misconception is that my concern is with catastrophic failures.
> Actually, that's not the case at all: those are rather unimportant.
> It's the nasty little glitches, the unexpected inconsistencies that come
> from preserving the wrong stuff that concerns me. Those problems are
> much more difficult to root out, and in fact may not exhibit immediately.

So we're concerned about the same thing. You worry that subtleties may
be carried along to a system's detriment. I worry that subtleties may
be lost, to a system's detriment.

> At [url]http://aplawrence.com/Unixart/ipuvsfreshinstall.html[/url] , I've posted my
> reasons for preferring fresh installs, and suggested that I would like
> IPU's a lot better if they would produce a full directory hierarchy diff
> of the upgraded system vs. the pre-upgrade system (or against what a
> fresh install would have produced), and a diff of all files preserved
> or merged against the fresh versions that otherwise would have been
> installed.
>
> It seems to me that would be a pretty simple thing to add to the IPU code,
> and it would have tremendous benefits. It would be easy to fix any mods
> that should not have been applied, easy to spot crapola that you don't
> need or want anymore, and in fact easy to "roll back" to as close to
> "fresh" as you needed to go to get it right.
>
> Until that happens though, I'll do fresh installs because I then can
> generate my own diffs and from that decide what to copy, merge, hand
> configure or ignore.

What you're asking for is already available to you. You do an IPU,
restore the old system into a subdirectory, and compare. You can also
compare to "original" versions relative to the new release, by following
paths like /opt/K/SCO/Unix/*/.softmgmt/var/etc/default/login. For any
"var" file in the SSO,

/var/opt/K/SCO/[product]/[version]/rest-of-link

the original unmodified version (belonging to the current version of
that SSO) is:

/opt/K/SCO/[product]/[version]/.softmgmt/var/rest-of-link

i.e. remove the leading "/var", add "/.softmgmt/var" after the version.

>Bela<