Re: Getting rid of circular dependencies, stage 5
On Tue, Jul 25, 2006 at 04:39:24AM +0200, David Weinehall <tao@debian.org> wrote:
> On Mon, Jul 24, 2006 at 06:32:54PM -0400, Joey Hess wrote:
> > Steve Greenland wrote:
> > > This really seems like something that while they may, very occasionally,
> > > be required, are mostly unnecessary and often misused.
> >
> > Rather, I'd characterise it as a feature that is necessary for any
> > general-purpose depencency-based system to be complete[1], which is
> > totally safe and does not adversely affect any aspect of the system
> > if some simple rules are followed, and which, if used incorrectly, is
> > still orders of magnitude safer than other dpkg features, such as its
> > support for setuid files, or its support for postinst scripts that run
> > arbitrary code at install time.
>
> Well, if foo depends on foo-data, and foo-data depends on foo, I find
> it really hard to see the point of splitting the two into distinctive
> packages...
I can see two common cases:
- Huge arch independant data. If you put that in the arch dependant
package, that clutters the archive. I did that for xulrunner for
example, with a libxul-common package.
- A single source package providing several binary packages using
the same set of data. apache2 does that with apache2-mpm-* and
apache2-common.
I both cases, the circular dependency would be useful to avoid
installing the common data without the software. Consequently, when
you apt-get remove the software, you don't get an orphan data package.
Mike
Reply to: