From: Austin Ziegler <halostatue@...>
Date: 2004-08-23T02:09:12+09:00
Subject: Re: Package requests for the prelim. Ruby Production Archive

On Sun, 22 Aug 2004 23:00:08 +0900, Dave Thomas <dave@pragprog.com> wrote:
> On Aug 21, 2004, at 22:08, Austin Ziegler wrote:
> > Doesn't ri have a problem if two parts of the class are defined in
> > --ri-system and --ri-site, much less a "local" directory?
> Yup! Currently it assumes that any one class is defined in one place,
> as I hadn't anticipated that
> the packaging folks would install into many different documentation
> directories. When I get this book project finished, I'll be trying to
> work with anyone who's interested to change this behavior to better
> suit their models.

Right. I think that this is something that can be dealt with
regardless of the model by the use of a (for lack of a better term)
RI_PATH similar to MAN_PATH. Then, RPA and RubyGems can add to or
remove from a standard RI_PATH location "at will" (or locations; I
would suggest that the RI_PATH could be stored in the site directory,
the system directory, and optionally the "local" directory).

Right now, without even using one of the main packaging systems, I can do:

  % rdoc --ri-system diff/lcs

If I have previously run rdoc/ri generation for Ruby as a whole to
--ri-site, then when I do "ri Array", I will no longer be able to see
the general ri documentation for Array (because Diff::LCS includes an
extension to Array). (There also appears to be a problem with merging
last time I checked this; I think I sent a patch that should fix that,
though. Merging should be the default) I think that it will also cause
problems if I generate the ri documentation in the "local/home"
directory.

As long as ri knows where to find various YAML files (e.g., RI_PATH),
then whatever solution is used to fix the existing problem will fix
the problem for packagers.

-austin
-- 
Austin Ziegler * halostatue@gmail.com
               * Alternate: austin@halostatue.ca