categorical index

[Paul Kienzle]

As recently discussed on help-octave, the categorical indexing
in octave-forge is quite popular with some people.  As I said
in response, building the index requires access to octave's
DOCSTRINGS files.

This is not a scalable design.  Octave should start out with
a base categorical index.  octave-forge should add its
functions to the appropriate categories.  Once we have
a cpan-like package repository, each package could add
its own functions to the appropriate categories, creating new
categories as needed.   When a package is removed, its entries
in the index should also be removed.

Given the index, we need a function to display the available
categories, and to display the functions within a category.
This is easy enough to write (I have a prototype already),
but it should be incorporated into the help function.

Annotations e.g., comparing and contrasting the quad
and quadl functions, explaining where you would use
one instead of the other is more ambitious.

The matlab model is to have a special file called contents.m
which is displayed when you ask for help on a particular
toolbox.  However, this does not support third party additions
to a toolbox, such as octave-forge's additions to octave.

Should we assume that a base index covering octave
and octave-forge is good enough, and any other additions
can be handled akin to contents.m?

Or should we come up with a more sophisticated design,
using, e.g., the metakit database:
        http://www.equi4.com/metakit.html
to dynamically generate the index information on the
fly as LOADPATH is modified?

Comments?

Paul Kienzle
address@hidden