[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]
Subject: More on i113
While working on a more detailed proposal for 113, it
appears to me that these tables need a bit more work than I thought. (Again I see these tables as more than just accessory: they
are necessary to nail down corner cases, and are ultimate ref material for
developers.) In addition to items currently in 113, I propose the following
– depending on reactions on the mailing list, I would update 113
appropriately: 1- As mentioned before, for each one of the tables, events that
may occur fall in two categories: (a) those generated by the RM component (e.g. RMD generates and
sends a Fault) and under full control of the RM component, (b) those “received” from outside , e.g. RMS
gets a Fault message. for (a) events, it is OK to use “N/A” for
the non-relevant states (the RM component has control over generating these
events), but we cannot just use “N/A” for (b) events, that the RM
component must be prepared to handle in whatever state it is in, even if such
events occur when they shouldn’t. We need to tell what is the effect of
receiving (b) events in every state (even if most of the times, sate remains
the same). Can’t just brush it off with N/A… 2- There are still several TBD values in these tables –
some of them are in particular related to the case where, say the RMS, gets a
fault like “Seq Closed Fault” or “Seq Terminated Fault”,
while RMS has not even closed or terminated the Seq (mostly, a decision from
RMD). I assume an RMS should update to “closed” when getting a Seq
Closed Fault, even if it has never sent CloseSequence (like it does for
termination). This has to appear in the table. Another case of questionable transition, is the “Elapse
Expires duration” event. Should close IMO instead of terminate, as RMS
may want to be able to query a final Ack. 3- there are events ( lines) in these tables that
actually do not cause any state transition. E.g. in RMS table: “new
message”, “retransmit of unack message” , “SeqAck
(non final)”, “Nack”. But it seems we are interested in
reporting what should the RMS behavior be for these in each current state. I’d
suggest to do this outside these state transition tables, e.g. in another table
where we consider specific events that do not cause any transition, - but need
to tell what should the RMS (RMD) behavior be depending on the state it is in -,
(kind of “decision table”). Jacques |
[Date Prev] | [Thread Prev] | [Thread Next] | [Date Next] -- [Date Index] | [Thread Index] | [List Home]