On Wed, 25 Sep 2002 10:20:56 -0700, Larry Osterman wrote:
> Nit: Should it be "should" or "SHOULD" in "The server should respond
> with a tagged BAD" below?
In general, I've avoided placing requirements on server handling of errors.
In my opinion, a compliant server could treat "FETCH *" of an empty mailbox as
the command to set the hard drive on fire, launch nuclear missiles on
Disneyland, and give the system manager a bad haircut...
The philosophy here is GIGO ("Garbage In, Garbage Out"); you can't predict
subsequent behavior if you violate the standard, so don't violate the
standard.
> Also, the UIDNEXT value? Is this right?
It was for lack of a better choice. The problem with saying "last assigned
UID" is that in a newly-created mailbox, there is no "last assigned UID" and
zero is not a possibility.
> < Mailbox is empty, client believes there are 0 messages in the inbox >
> < Message arrives in mailbox >
> C: 1 UID FETCH * (RFC822.HEADER)
> S: * 1 EXISTS
IMHO, this is a timing race. Either UID 1 doesn't exist at the time that the
command is executed, or it does exist. So, the scenario should be one of:
C: 1 UID FETCH * (RFC822.HEADER)
S: * 1 EXISTS
S: 1 OK done
or
C: 1 UID FETCH * (RFC822.HEADER)
S: * 1 EXISTS
S: * 1 FETCH (UID 1 RFC822.BODY {xx}
<Body of newly received message>)
S: 1 OK done
I do not think that:
S: * 1 FETCH (UID 1 RFC822.BODY NIL)
should ever be returned. This is the worst of both possibilities.
Can you look into what Exchange does? Fortunately, I don't think that it
should be a major problem, since I doubt that clients will ever do this.
Clients are much more likely to do "UID FETCH 1:*". I assume that you'd agree
that
C: 1 UID FETCH 1:* (RFC822.HEADER)
S: * 1 EXISTS
S: * 1 FETCH (UID 1 RFC822.BODY NIL)
S: 1 OK done
is a definite bug!
