In relation to the recent rant about the new rpmdrake...

* One more disappearing regreted feature:

In previous releases (7.1 I think) there was a primitive rpmdrake tool,
that allowed an interesting feature, though it was not intended by the
developper, and disappeared in the 8.x.

Imagine you were looking for, say, an xfig replacement. You would search
for xfig. In 7.1 the search engine highlighted xfig *in its category*.
Then, you just had to look at xfig's neighbours in the category to find
similar programs.

In 8.x the same search returns only programs that contains "xfig" in a
separate list. (This list is good because you can select several
packages at once.) But you lost the ability to browse the category.

Still, the categories exist (they existed in RedHat 4 and probably even
before), but you can't take advantage of them in the 8.x rpmdrake.


* When this becomes a suggestion of improvement:

To continue with the xfig example, I know that you can look for "vector
drawing" or the like, but sometimes you know a program that is similar
to the one you're looking for, but you don't know what keywords to use.
(Or the keyword is too vague, like "network" of "font" so you have tons
of irrelevant results.)

The improvement could take the form of a button (or contextual menu)
that says "jump to the category where this package belongs".

So, you search "xfig". The xfig package is selected. You click "jump to
category" and voil�, you have sodipodi. (There may be some more relevant
examples).


* And in your wildest dreams:

The essence of this idea is to take advantage of the neighbourhood
relationship between packages, instead of merely a keyword-based
matching.

This is already what you do in a file manager when you use
sort-by-name/date/size.

The idea is to add to the "sort" criteria more similarity criteria:
function (vector drawing), environment (KDE/Gnome/...), dependencies
(qt,xml,...) to jump and/or filter.

In a dream you could imagine an interface like in http://kartoo.com/

There are researches about "self-organized maps" (SOM) where some may
take inspiration.


* Conclusion

When the number of packages in a distribution becomes very large, what
you actually need is a package search engine.

And a good search engine tries hard to take all available information to
focus on relevant answers and provide easy ways to navigate and refine
the query. That's what Google does.

-- 
St�phane Gourichon - Labo. d'Informatique de Paris 6 - AnimatLab
http://animatlab.lip6.fr/ - philo du dimanche http://amphi-gouri.org/


Reply via email to