[cap-talk] search capability

[Sam Mason]

On Mon, Sep 29, 2008 at 12:09:51AM +0100, David-Sarah Hopwood wrote:
> Rob Meijer wrote:
> > 1) A single indexer would require an abundance of authority in order to be
> >    able to create an index of the full graph.
> 
> The user's shell has sufficient authority.

As far as I know, most search systems tend to have a single global
index that's built by a privileged entity (i.e. root in a unix system).
The results of searches on this global index are filtered before
being returned to the user to ensure that they can actually see the
node/document in question.

The most fiddly case to me seems to be the content management system
(CMS) of some large organization.  Here there would be many thousands of
users and if each was forced to maintain their own index the majority of
it would be identical between users.  Storing thousands of copies of an
almost identical multi-gigabyte index seems wasteful.

Turning things around and being able to ask each capability to search
its transitively reachable objects seems possible, but, as Alan said,
seems to make indexing awkward.  I believe Facebook has encountered
and solved these problems already, but unfortunately I can't remember
where I read about it.  In Facebook, users can only see their friends'
content; hence being a friend of somebody could be seen as having a
capability to them.


  Sam