[BXXPwg] proposal for 1:N interactions

james woodyatt jhw@wetware.com
Thu, 31 Aug 2000 17:10:51 -0700


At 15:15 -0700 2000.08.31, Marshall Rose wrote:
>  > >if i get back an ERR, i'm thinking that nothing happened whatsoever.
>>
>>  So, what do you get if things are happening and you're getting back
>>  positive stuff, but then something goes wrong before everything is
>>  done happening?
>
>you've already done useful work, so sending an ERR isn't appropriate.

I get it now.  Exceptions are not cause for ERR messages-- only 
errors are cause for ERR messages.  (At least, that's how *I* have to 
think about it to keep myself honest.)

Sending an ERR isn't appropriate because the earlier ANS message 
implies there is no protocol error.  There is symmetry after all, 
since ERR always indicates a protocol error, and exceptions are 
always indicated with variants of RPY and ANS messages, depending on 
whether the interaction is 1:1 or 1:N.

Perhaps it would be helpful to make this point clear in the 
appropriate draft?  It would be a real shame if protocol developers 
missed this point and abused the ERR message by using it to report 
exception conditions rather than protocol errors.


-- 
j h woodyatt <jhw@wetware.com>
http://www.wetware.com/jhw