pacakge classification and selection (was Re: New Comps Groups)

[Bill Nottingham]

Nicolas Mailhot (nicolas.mailhot at laposte.net) said: 
> Well, maybe comps was not intended to be a lot of things, but when it
> was decided not to fix rpm groups and use comps as the only
> classification method of yum repositories, comps graduated to another
> level.

Then perhaps it's time to revisit that problem.

(insert dream sequence music here)

Say I'm a generic Fedora user. I want to install a music player. I fire up the
software tool.

I now have the choice of *78* different packages under sound and
video. First, I have to sort through the packages that aren't
music players, by reading to see what they are...

-'A Better CD Encoder'? No.
- 'Audio mixer that uses ncurses'? No. And what's a ncurses?
- 'An AdLib (OPL2) music player'? Maybe that's what I want - I wonder what
  AdLib is? Is that like an iPod?
- 'Rewrite of the xawtv webcam app, which adds imlib2 support'? Why do I care?
- 'The Jack Audio Connection Kit'? Well, I'm sure that means something to
  someone.
  
Now, that I can sort of pick out 10-20 packages that might be want
I want, how can I make a good decision? How do I choose between

'Media player for KDE'
'Xine-based media player'
'Multimedia applications for the K Desktop Environment'
'Music Management Application'
'Free multimedia player'
'The X MultiMedia System, a media player'

Ideally, I'd get information like screenshots, user ratings, reviews,
descriptions of specific features. But that's all the information I have.
So I sit there confused, pick one at random, or install them all. I
then hop on my handy dandy Fedora forum somewhere, and ask why one of them
doesn't work for me, and someone says:

<dude> don't use that, that sucks! why would you pick that!

And I am now sad. And that's *with* a list of culled and vetted packages.

Perhaps I could use the 'search' interface for packages. So I search
for 'music player'.

It includes some of the same results (but not most of them, including
pretty much any that you'd want), and such gems as:

hfsplus-tools - Tools to create/check Apple HFS+ filesystems

Yes, that's it!

(As an aside, I'd say that if you wanted a text-based mixer, searching
for 'text mixer' is an infinitely better way to do it than browsing
a list for aumix. And the search doesn't find it.)

Let's try a different example. I'm a python coder, and I want to find
a library for doing cryptography and hashing - MD5, SSL, etc.

So, I go to the 'Python Development' Group in pirut. I scroll through
the list, looking for something that looks relevant. (See above). This
is even better now, because there are 500 sets of python bindings.

Ideally, we'd have a browsing interface that states things like:

API Categories:

Network connections
Cryptography
Graphics Processing
Applied Math Besides Those Two
Non-Applied Math
Math Is Hard - Shopping Cart Handling

and each of them would have links to various packages, which would include:
- API reference
- API/ABI stability
- standardization (is this core python? Is it in the LSB? Is it a recommended
  API by Fedora)
- reviews, examples, etc.

Again, comps does not help you here.

Now, I could search, again. But "python crypto" yields nothing.

So, in short, I think the comps push, while fitting in our current
infrastructure, isn't really solving the *right* problem. We need a
better interface... something like Amazon or freshmeat or <insert
handwaving>.

And, we need better search. Right now, it looks like it's doing direct
globs without wildcards - i.e., if you enter 'foo bar' on a line, it
needs to match *exactly* 'foo bar' in either the summary or description.
Ideally, it should be more like google - split the query into words,
find what matches those words, sort by number of words matched and
proximity. Simple google-like matching (+,-, quoting) would make it
simpler, and fit the methods people are used to.

Bill