Dependency loops considered harmful?

[Michael Schwendt]

On Wed, 3 Sep 2008 13:36:41 -0500 (CDT), Jon Ciesla wrote:

> 
> > On Wed, 3 Sep 2008 12:10:13 -0500 (CDT), Jon Ciesla wrote:
> >
> >> > For example, why games data depend on binaries?
> >>
> >> So that if we try to upgrade the version of the binaries and not the
> >> data,
> >> it will fail.
> >
> > Unconvincing.
> >
> > Why? Because the game program pkg could require a specific version of
> > the data pkg to achieve the same. If game version gets bumped, dep
> > would break, since you would need to update data pkg, too. Game requires
> > compatible data. So far so good.
> >
> > However, if the data pkg requires the game pkg (versioned! or else it
> > would not be strict enough for your needs), this only increases the
> > risk that you need to bump'n'rebuild the data pkg if the game version
> > is increased without needing any changed data. It's a superfluous
> > dependency that only tries to enhance "yum remove" a bit.
> 
> How?  For game data, the convention is to require on name and version, but
> not release.  If new data is required, it will change the version.  If
> not, we only increment the binary rpm's release, so the data rpm matches
> on version and needs no rebuild.

game-0.8  Requires  game-data = 0.8
game-data-0.8  Requires  game    (which version? and why?)

Case 1) 
game-data-0.8  Requires  game = 0.8
What happens if game-1.0 is released with unchanged game-data?
Then you need to update game-data for nothing else than the broken dep.

Case 2)
game-data-0.8  Requires  game
No version. Hence no dep breakage in all cases where you may update game
without updating game-data. Instead, a strict dependency is installed
in pkg game:  game-%{version}  Requires  game-data = %{SOME_version}
Here you can control the game-data version within pkg "game".


Dep 1) is really just for "yum remove".