
      $Id: TODO,v 1.2.4.41 2000/03/16 17:12:18 ajp Exp $

There may still be a few race conditions present in this code.  Also,
the discovery of duplicate objects and discovery of addresses are not
completely reliable since they use udp.  There are some spots that
need more error checking, and perhaps certain operations should be
retried up until a timeout.

It also may not be a bad idea for the udp daemon to keep a cache of
the objects its tcp daemon has that it knows about.  This cache will
have a timeout associated with it, since objects can very likey
dissappear and then reappear on another node in the cluster.

This package should probably include a make test target in the
Makefile and a C-program linked with the library to test the
operations after it has started both daemons on the localhost.

I plan to store the objects in hash tables (one per type of object/
synchronization mechanism) and allocate them dynamically.  Doing this
may require reworking the naming scheme the client library uses to
make this more conducive to a hash table.  This will allow more
objects to exist on the server and will expedite searching.   Also,
we'll probably deprecate the "get load" request.

The documentation is still very sparse.  I plan to remedy this soon
(hopefully by the first stable, non-development release).

A manual (online or otherwise) would be nice, and would help those
who are just starting out using pvmsync to get their programs off the
ground.


