New Comps Groups

[Jeremy Katz]

On Mon, 2006-11-27 at 21:52 +0100, Nicolas Mailhot wrote:
> Le lundi 27 novembre 2006 à 15:33 -0500, Brian Pepple a écrit :
> > On Mon, 2006-11-27 at 13:24 -0500, Bill Nottingham wrote:
> > > We already have a 'package search' interface for finding packages - is
> > > listing 100 (or however many) python-* packages better than this? In
> > > what way? Are they not getting pulled in for dependencies when necessary?
> > 
> > I'm in agreement with Bill on this.  Pretty much all the python-*
> > packages should be pulled in as dependencies.  Am I missing something
> > here?
> 
> It's pretty much impossible to autodetect missing comps entries unless
> every package is systematically put in comps. No autochecking means low
> QA.

But the entire point is that everything _SHOULDN'T_ be there.  If so,
then it's no better than a list[1]

> Also if a group is too big it should be broken up in lighter
> finer-grained ones IMHO. Choosing the right group is much less work than
> writing the package description, and often more useful for users.

And when a user now has to go through 100 groups to find the one they
want?

Jeremy

[1] Grouping is somewhat arbitrary by nature and different people will
have different ideas on how packages should be grouped.  By your
argument, I could just as well make the categories letters of the
alphabet and group based on the first letter of the package name. But
that *doesn't* provide anything useful for users.