Re: Two improvements to etcupdate(8)

[Jean-Yves Migeon <jym%NetBSD.org@localhost>]

Le 2016-01-27 19:36, Edgar Fuß a écrit :
> Should this go to tech-install?
>
> I wrote two improvements (or so I think) to etcupdate(8) for which I 
> attach
> two separate patches. I must admit I did't test them too hard yet.
>
> 1. I never used automatic (-a) mode because I never used it before.
> The problem is that unless you used -a last time, you lack the 
> reference
> checksums in /var/etcupdate.
> So I added a ``bootstrap'' mode (-A) that only builds the checksums 
> from its
> -s arguments and doesn't do anything else.
> The idea is that you first etcupdate -A -s /path/to/old/etc.tgz and 
> then
> etcupdate -a -s /path/to/new/etc.tgz afterwards.

That one seems fine to me, there should be a way to "init/bootstrap" 
etcupdate if it was not.

> 2. I regularily update a whole set of machines. I usually did this by
> running etcupdate on a development machine, noted the per-file 
> decisions
> I made (on paper) and then used that list to know what to type to 
> further
> etcupdate invocations on the production machines.
> I think computers are better at recording and playback than humans, so 
> I added
> two switches, -W and -R, that record the (hand-made) 
> keep/overwrite/merge
> decisions to a file resp. use that file to follow the same decisions 
> again.
> The file doesn't record the details of a merge operations, only the 
> fact,
> but manually re-merging three files is an improvement over typing d/i/m
> for a hundred files. Or so I think.
>
> Comments/suggestions/improvements (and better names than ``bootstrap'',
> ``read'' and ``write'') welcome.

bootstrap, record, apply [...]

Does it not collide with the "postinstall fix" operation? I thought 
postinstall's purpose was to handle all kind of operations that could be 
"automatically" solved (and log those that required manual 
intervention), and use etcupdate to handle the m/d/i of specific files 
(passwd and the like?)

--
Jean-Yves Migeon