Ed Wilts wrote:
> > Basically, all I'm suggesting is that we treat /etc/passwd and similar
> > files like XF86Config... create it, never install it. After that, specific
> > packages can add or remove as needed, including setup.
>
> I disagree. I regularly go through and search for .rpmnew files and then
> compare what's in them with my own versions. Sometimes there are new
> defaults or options that weren't there when I did my own customizations. I
> then decide to either toss my file and start with the .rpmnew version, or
> apply some changes to my own.
Not thinking like a newbie. :o) I do the same as you, but I'm not a
newbie.
What I'm suggesting applies mostly to system critical files that can
easily be manipulated through existing tools and are expected to change.
Look at the case that I keep bringing up... It is pretty simple for the
pre-install script of the postgres RPM to run useradd and post-uninstall
script to run userdel (and trap errors should the user exist/not exist).
Why would you create users to run software that isn't installed? The
same applies to other such files or RPMs.
For configuration files which are not expected to change (such
XftConfig), by all means, install them, even if they do end up with a
.rpmnew extension. If the user has changed the file, then they can
inspect for differences and make the change. Realizing, of course, that
people who are newbies won't typically change these files and people who
are experienced know what to do when a new file comes in.
Just my opinion.
John