Re: [higgins-dev] IdAS: view of objects in intermediate states

["Tom Doman" <TDoman@xxxxxxxxxx>]

Jim, Duane, and I discussed this some today and were considering the Basic* implementations that we currently have.  I was thinking that if we can get those implementations to assist in the tracking of changes that, in the JNDI CP, I would simply normalize reads into Basic* instances and, while in memory, those would then reflect any modifications that occurred for subsequent reads.  The CP would simply keep track of the modifications through some sort of notification or callback process and apply those when requested to do so.  Then it wouldn't have to do any fancy dancing to determine what was originally read and what was now modified.  Of course, this approach would blow any lazy evaluation that we might want to do but I see a more dynamic approach as being complicated and brittle.

Tom

>>> "Jim Sermersheim" <jimse@xxxxxxxxxx> 4/26/2007 2:07 PM >>>
We need to document what happens when a consumer has an instance of an object (like an IDigitalSubject), performs some updates on that, and then performs read operations on it prior to calling IDigitalSubject.applyUpdates().
 
I believe that the consumer will expect the read operations to reflect recent changes.  The other choice is to respond to read operations with a view representing the not-yet-applied changes.  I think this will be unexpected on the part of consumers.
 
In addition, we need to document the scenario where a consumer has two instances of the same subject (call them A and B), updates one (A), and reads the other (B).  After thinking about this, I believe that we should document the semantics of this as being undefined.  That is, the CP can choose whether to present the updated or original view when B is read.
 
Jim