Distribution semantics: database consistency -- transactions

Gary D. Duzan gary@wheel.tiac.net
Wed, 09 Nov 1994 21:49:26 -0500


In Message <Pine.SUN.3.91.941109092823.8828C-100000@crl3.crl.com> ,
   Mike Prince <mprince@crl.com> wrote:

=>
=>
=>On Wed, 9 Nov 1994, Francois-Rene Rideau wrote:
=>
=>> So we need recovery mechanisms.
=>
=>> In any case, that means our object semantics *must* include failures
=>> as standardly supported states.
=>
=>Devils Advocate: But why?  In the real world when you send something off 
=>you sometimes don't care if it comes back.  Now lets say the mechanism 
=>for automatically managing failures of remote happenings is relatively 
=>"heavy".  Should we give the programmer the option to specify whether that 
=>failure should even be checked for, or insist the potentially uneeded 
=>mechanism is in place.  

   Ever heard of futures? Ellie? It is a language mechanism, but
it seems to me to be the right way of dealing with return values,
normal or abnormal, in a concurrent system.

                                      Gary D. Duzan
                         Humble Practitioner of the Computer Arts