Re: Selecting packages from thousands
Ian Jackson <ian@davenant.greenend.org.uk> wrote:
> I submit that the desired user interface is that the user is offered
> (by defautlt) a set of `categories' or `keywords' or whatever,
> something like
>
> Standard internet protocol clients.
> Software development tools.
>
> and gets to choose whether they want them. Each of those categories
> corresponds to a list of packages, and the resulting installation is
> the union of them. Perhaps categories will have different levels of
> support which might correspond to Priorities.
>
> I think we're all pretty much agreed so far.
Probably -- at least, I agree.
> Clearly these categories are not sections, because a package may be in
> none, or more than one, or be installed regardless (eg, a Required
> package).
>
> The next question to ask ourselves is: who decides on these categories
> and who decides on which packages are in them ? Clearly the
> categories need to be decided on centrally, so their descriptions can
> be specified centrally.
Probably.
> I think that if we have a stronger Priority mechanism, which
> distinguishes Optional packages that most people will want from those
> which most people should probably ignore, we can just allow package
> maintainers to decide which categorie(s) their package fits into.
>
> Then the categories are like keywords. So we end up with:
>
> * A file in the distribution alongside the Packages file. We could
> call it Categories. It contains a list of the categories. Each
> category would have one or more levels of support; for each level of
> support it lists the keyword/priority combinations that should be
> included.
>
> * Package installation tools should by default select packages
> according to whether they match the selected level of any category.
> But, the user can go in at a more detailed level and (de)select
> individual packages.
>
> What do people think ?
I'm worried about bottlenecks -- there are a lot of people in Debian
and we can't expect all of them to be able to give this priority. So
we ought to do something to help with differences of opinion, occasional
strong contributions, etc. [In other words, centralized authority
is only good for final approval..]
So, I think it would be good to (early on) define a few basic tools to
work on such files (for example, implementing category intersection,
union and difference). [I expect these tools to be trivial, but I
also expect that some valuable contributions would come from people
who don't know how to write such a tool.]
[But this is more of an implementation issue than a policy issue.]
--
Raul
Reply to: