Re: Status of circulars dependencies in unstable
On 04/09/2011 14:44, Eugene V. Lyubimkin wrote:
> While I also would want Debian to eventually get rid of circular
> dependencies, I am not sure about (the value of) the benefits.
> For example, even by default dependency resolvers have to consider
> Depends-Recommends loops. And if the user has selected to install Suggests
> by default (yeah, yeah, mostly theoretic-only but still possible), there
> will be a lot of Depends-Suggests or Recommends-Suggests loops.
Depends-Recommends and Depends-Suggests loops (and even more
Recommends-Suggests loops) are totally different than Depends-Depends loop.
If A depends on B, configuration of A will be done after B, unless
there is a Depends-loop between A and B and that the resolver choose to
break the loop here. Note that the loop can involve other packages. In
the past, this has lead to subtle bugs hard to workaround.
Postinst scripts would be more robust if we do not have Depends-loops.
In this case, we would have "if A depends on B, configuration of A will
always be done after B". It is easier to handle.
No such thinks exist for Depends-Recommends loops or other weaker
loops: if A recommends B, it can never expect B to be configurated
> Also these benefits will be quite delayed, even without counting
> third-party repositories dependency resolvers will have to wait at least
> 2 stable releases until they could drop the relevant parts of the code.
> Therefore I think _for this moment_ mandating in the policy will be too
Waiting more times will not decrease the number of releases to wait
before dropping the relevant parts of the code. So, in my point of view,
the sooner, the better.