Andrew, Ken: I really like to see this happen -- here are some
comments that actually might make things a bit easier to implement, by
making them less ambitious.

I have no scientific evidence, but I'm not a big fan of keywords.
My guts tell me that an extensive keyword system may not be worth it.

I think I like the organization of a reference book best: there's a
hierarchical table of contents at the front, and there's an index in
the back. Translate this to computers, and you have something like
Ken's classification hierarchy, but without symbolic links, plus full
text search on the descriptions. The descriptions of course should be
structured (type, name, version, author, date, url, platform,
see-also, etc.) but additional full-text indexing a paragraph or two
of description makes it much easier to find things using odd search
criteria.

You could consider the hierarchical table of contents a keyword scheme
(and it has some of the same properties -- it needs occasional
extension but that's mostly a manual process) but the difference is
that each module is in exactly one place. That's useful because it
makes it possible to map it to a real file system without using
symbolic links (Ken already explained why that's a bad idea) and it
also makes it possible to get a feel for the size of the collection as
a whole, since every item is counted exactly once.

Some additional ways to access things might make things a lot more
usable:

- most popular (e.g. by recent download count, if available)

- new arrivals

- a personal selection (e.g. tchrist's selection from CPAN)

But just collecting the raw records for all items and making them
searchable is probably the most useful thing to start with. The FAQ
wizard might even be good enough!

--Guido van Rossum (home page: http://www.python.org/~guido/)

_______________
LOCATOR-SIG: Discussions about a Python Locator for resource discovery

send messages to: locator-sig@...
administrivia to: locator-sig-request@...
_______________