On Mon, Dec 18, 2006 at 07:23:29PM +0100, Axel Thimm wrote:
On Mon, Dec 18, 2006 at 05:47:48PM +0000, David Lutterkort wrote:
> On Sat, 2006-12-16 at 14:09 +0100, Axel Thimm wrote:
> > On Fri, Dec 15, 2006 at 07:43:04PM -0600, Jason L Tibbitts III wrote:
> > > >>>>> "TK" == Toshio Kuratomi
<a.badger(a)gmail.com> writes:
> > >
> > > TK> I think tibbs had the opposite viewpoint but I don't remember
if
> > > TK> we got to a point where he decided it didn't matter or we came
to
> > > TK> an agreement or just let it drop.
> > >
> > > I guess the point is that I can't figure out what additional value it
> > > adds, and in general it's bad to package up something that's
> > > completely needless.
> >
> > egg is a packaging method that is orthogonal to what we use. Leaving
> > the eggs around may get users to start using egg-installation and get
> > files on the system unregistered by rpm.
> >
> > Or not? If the above is correct eggs should even be banned just as
> > other non-native package formats are banned (debs or tarballs for
> > example).
>
> The crucial issue are the dependencies that right now have to stay
> within each packaging format; if rpm's can't contain any egg (or gem or
> whatnot) info, users will end up installing the same package twice, just
> to fulfill dependencies completely within each packaging system.
Don't you have the same issue if you install the egg with -Z? If not,
then the (egg-)package dependencies are obvioulsy spooled somewhere on
disk for easy_install and friends to find.
Looks like all has been considered in advance by the egg folks:
--single-version-externally-managed
This boolean option tells the install command to perform an "old
style" installation, with the addition of an .egg-info directory
so that the installed project will still have its metadata
available and operate normally. If you use this option, you must
also specify the --root or --record options (or both), because
otherwise you will have no way to identify and remove the
installed files.
> It would be much more userfriendly if we laid the groundwork for
other
> packaging systems to depend on rpm-installed bits; that mostly means to
> _allow_ inclusion of non-rpm packaging metadata in rpms.
If you like so, having "egg-provides" is fine, of course. Just like we
have foo.pc, but don't keep the full tarball around.
The equivalent to *.pc files seem to be the .egg-info subdirs. So we
don't need to ship the egg file in addition within the rpm file, but
still feed the egg packaging system with information.
--
Axel.Thimm at
ATrpms.net