Loïc Minier a écrit :
Hi,
On Wed, Mar 01, 2006, Filipus Klutiero wrote:
synaptic permits removal for "uncommon" situations, but *should* install
recommendations by default. I still disagree that it's a bug not to do
so though.
You disagree that's it's a bug not to implement the dependencies
described in Policy?
Hehe, I'm not sure whether this is a proper reformulation of what I
stated :)
Anyway, be warned that you'll have a hard time to back a bug on a dpkg
front-end with Policy, which is "APT-agnostic". If a dpkg front-end is
found to violate policy due to its package management, it will probably
end up being a dpkg bug. As I guess you agree that it's not a bug for
dpkg -i to ignore recommendations, I think you should see that synaptic
isn't violating Policy neither.
Back to my point: the simple fact this is not implemented renders this
type of dependency *dangerous* for package maintainers to use. I can't
rely on the behavior or Recommends to be installed by default any more,
hence I'm forced to upgrade deps to Depends, and to remove freedoms to
my users. Thanks.
It's hard to see why a recommendation would be dangerous. As you know,
packages can't rely on recommendations, so if it's dangerous to not have
a recommendation installed, it shouldn't be a recommendation. You are
not forced to change deps and remove freedoms to your users. As a
maintainer you should simply ignore the bugs of package managers, as
well as other package managers issues if you consider them bugs.
I certainly *don't*, we have correct package managers already, aptitude
is an example, and the old and not-so-maitained-anymore dselect is
handling this way better than synaptic. How can you consider
pushing users into using "simpler" or "nicer" software when you are in
fact pushing them into breaking recommends?
"Breaking recommends" is not a good description of what synaptic is
doing with them. Anyway, aptitude looks like a bad example of a package
manager to state as superior to synaptic given #299009. I must admit
that this is where my experience with it ends, so I don't know how
broken or not broken it is beyond that.
Package recommendations have to be the default, and don't have to be
explicitely selected by users.
Again, my only disagreement with this is about "have to" versus
"should". As a wishlist bug this is clearly valid and a high priority
enhancement to do, but then it's a matter of writing it as usual, and as
Michael said. Cheers to the patch writer.
