rpms/gnumeric/FC-4 gnumeric.spec,1.3,1.4

[Michael Schwendt]

On Mon, 24 Oct 2005 10:33:25 +0100, Paul Howarth wrote:

> >>No, indeed not. What you call a purely theoretical dependency is "one
> >>package requiring the root directory of something else". No more, no
> >>less.
> > 
> > 
> > Then have gnumeric require this dir. I don't care.
> > 
> > What you can't do is:
> > 1. state this dir may be required by packages that need to stuff things
> > inside
> > 2. ergo, ownership can not be shared
> > 3. but since in gnumeric case you find it too much hassle, do not
> > actually do 1. and keep 2.. But 2. is only a good rule if you do 1. If
> > you don't do 1., I don't see why other packages would, and so 2. has no
> > base at all.
> 
> Couldn't this be resolved by splitting out the files that go in 
> /usr/share/mc into a separate gnumeric-mc package, and make that 
> dependent on mc?

Would that be worth it? To create tiny packages only to make sure that
directory ownership is complete? It would not guarantee that /usr/share/mc
is removed when gnumeric, mc and other packages are removed. My initial
comment said "exaggeration", I stick to that. Making "gnumeric" require
"mc", although it is optional, would be exaggeration, too.

> I have to agree with Nicolas that rules about directory ownership should 
> apply consistently, and not be ignored when it's inconvenient.

Well, I prefer semantic correctness of dependencies and home directories.
In my point of view, /usr/share/mc _belongs_ to Midnight Commander, and
hence only mc ought to own that directory. Now, gnumeric does not _require_
mc, it only needs /usr/share/mc, but even if that directory is not present,
RPM installs the files anyway, although they are not used and not needed.
So, there is no strict dependency.

-- 
Btw, why doesn't RPM prune empty directories which don't belong to
any package left in the database after package removal?