Simon Josefsson wrote:
Keith Moore <moore(_at_)cs(_dot_)utk(_dot_)edu> writes:
If nobody can find client behaviors that MFT/MRT can achieve, that NR
together with (a) cannot, especially wrt news (MFT/MRT is used in news
too, IIRC), I'd like to see NR move forward.
there are client behaviors that MFT/MRT can produce that NR cannot
produce. in particular, NR has no way to specify situations where
"reply to all" is not sent to a superset of the recipients who would be
sent if "reply to author" were used instead.
as far as I can tell, this is a plus for NR.
Yes. I should have added a "useful" before "client behaviors". Since
it is possible to generate some things with MFT/MRT that NR can't do,
it isn't obvious that all of those MFT/MRT things are useless,
although it appears so. That need to be verified through more review.
I did a case analysis of NR vs. MFT. The cases looked like this:
class field reply to author reply to all superset
F1 From n n y
F2 From n y y
F3 From y n n
F4 From y y y
T1 To n n y
T2 To n y y
T3 To y n n
T4 To y y y
C1 Cc n n y
C2 Cc n y y
C3 Cc y n n
C4 Cc y y y
R2 - n y y
R3 - y n n
R4 - y y y
each class except the R classes is a class of recipients. The R classes
are not recipients of the original message, but are intended recipients
of replies. the superset column is whether the reply to all recipients
are a superset of reply to author recipients.
I believe this is is a complete set of cases.
say you want to send a message to zero or more of each class of
recipients. here's how you'd do it:
NR:
From: F1, F2, F3 (1), F4
Reply-To: (2) F4, T4, C4, R2(1), R3 (1), R4
To: T2, T3 (1), T4
To-NoReply: T1
Cc: F2, C2, C3 (1), C4
Cc-NoReply: C1
MFT:
From: F1, F2, F3, F4
To: T1, T2, T3, T4
Cc: C1, C2, C3, C4
Mail-Followup-To: F2, F4, T2, T4, C2, C4, R2, R4
Mail-Reply-To: F3, F4, T3, T4, C3, C4, R3, R4
Notes:
1. classes that can't be dealt with properly using NR are F3, T3, C3,
R2, and R3. The *3 classes can't be dealt with because NR fields only
specify replyability in general, they don't specify replyability
differently for reply to all and reply to author. In the NR example
above, each of the *3 classes is treated as if it were *4. The R2 class
cannot be dealt with because Reply-To (which is the only way to include
recipients on replies who were not recipients of the original message)
doesn't distinguish between reply to author and reply to all.
2. if all classes in the reply-to field are empty, specify this as
Reply-To: nobody:;
MFT seems more straightforward when viewed from the point-of-view
of this case analysis, but this is partially because the cases were
organized in such a way as to cover the union of what MFT and NR can do.
I see two questions that determine the relative merit of NR vs MFT:
1. how useful are cases F3, T3, C3, R2, and R3?
IMHO, R2 might be useful. The others seem more likely to cause
confusion than to provide behavior that an author finds useful.
2. how much potential is there for intended "NoReply" recipients to fail
to recieve mail because "sendmail -t" doesn't recognize the new fields?
Keep in mind that this is only a problem when the new fields are used.
Presumably existing mail robots that generate messages and send them to
"sendmail -t" won't automagically start generating NoReply fields.
But for user agents that allow users to specify arbitrary header fields
and pass them off to "sendmail -t", this is a problem.
Also, I don't see that there's already a significant deployment of MFT,
so I don't think that's much of a factor.
Keith