Bart Schaefer wrote:
The fundamental problem is NOT the behavior of reply-to-all. The
*really* fundamental problem is that reply-to-all is the wrong
function to be using in the first place. The only reason that
reply-to-all gets used, aside from the Berkeley Mail historical
baggage of the meanings of `r' and `R' as commands, is because the
simple reply function is unable to produce the correct result.
Huh?
My MUA has two commands, "Reply to author" and "Reply to all".
The former sends to (Reply-To -or- From) and the latter sends to
((Reply-To -or- From) -and- To -and- CC).
I think this is the universally implemented status quo. I gather
that Bart thinks this is wrong behavior, but this is that this is
what -every- MUA I've ever seen does.
And I really wish, given the vast installed base, the standard
spelled this out.
} I'm talking about convenient _default_ recipient lists. The user is
} of course free to send mail wherever he wants.
I'm talking about convenient default recipient lists, too. The basic
reply command ought to be where one gets those convenient defaults.
Reply-to-all ought to be the way the user sends mail wherever he wants
(e.g., find all the possible addresses for me, and let me edit some of
them out).
Uh, I think now you're talking about a disjoint set of commands:
"Reply to convenient default recipients" and "Reply to all listed
addresses". I gather that you think the former list of addresses
would be determined by some as-yet-unstandardized-or-otherwise-
agreed-upon mechanism, and the latter would be something like
(From + Reply-To + To + CC.)
I think that's a monumentally bad idea.
Aside from not being even close to the behavior I'd want to see in an
MUA I was using (thanks, but reply-to-author and reply-to-recipients is
exactly what I like) this would result in a confusing change to every
existing MUA: there would still be two reply commands, but they'd be
totally different ones than before; or worse, there'd be four distinct
reply commands. Ugh!
For this to work as described, two things are required: First, the
author has to change From and leave out the Reply-To when he means
"reply to address-list X if you want to just reply to me" and he must
put in Reply-To exactly when he means "reply to address-list Y if you
want to followup." The fields look like this:
Replies to address-list X to reply to me (case 1):
From: X
Sender: <real-author>
To: <destinations>
Replies to address-list Y to follow up (case 2):
From: <real-author>
Reply-To: Y
To: <destinations>
In that second case, this message has destroyed my ability to reply to
just the author -- because every MUA in existence today has a command
that replies to (Reply-To or From), and a command that replies to
((Reply-To -or- From) -and- To -and- CC), but no command that even gives
me the *opportunity* to reply to From (if Reply-To is present.): that
address would never appear in the defaultly-offered set of headers: one
would have to put it there with cut and paste.
Perhaps this is what the author of that hypothetical message intended.
But if so, the author is a doofus.
The second thing required is that user agents interpret Reply-To in
such a way that *neither* `reply' nor `reply-using-destination-fields'
takes any addresses from the To field in the event that Reply-To is
present.
That is already the status quo. Right? (For all four of the possible
reply commands that have been mentioned, or at least, the two (possibly
three) of them that have ever been implemented in the real world.)
Someone, I guess Keith, wrote:
I realize that many UAs don't handle "reply-to-all" this way,
in particular a lot of UAs use Reply-to only as a replacement for
From. But the DRUMS document already clarifies that reply-to is the
author's preference as to where replies should be sent.
I haven't read the referred-to document, but for the sake of argument,
I'll take your word for it that it says that a MUA's "reply to all
recipients" command should reply to
A: (Reply-To -or- (From -and- To -and- CC))
rather than the de-facto standard of
B: ((Reply-to -or- From) -and- To -and- CC).
But if it says that, then its authors are operating in a delusional
fantasy world, because that's *not* going to happen.
My reading of RFC 822 (and, I gather, Bart's reading?) is that it does
not favor one over the other, as it talks only about the "reply to
author" command, and not the "reply to all recipients" command, which it
relegates to the status of "extension."
Even if 822 *did* say that B was preferred to A, well, that's just too
darned bad, because A got implemented and B did not, and if the standard
is to mean anything, it should be made to say A. Unimplemented
standards are useful as objets d'art, but little else.
Any attempt to address either "duplicate suppression" or "default
addresses for wide and narrow replies" (which are two totally different
problems that people have been (badly) attempting to address using the
Reply-To header) has to take this status quo into account if it hopes to
get anywhere at all.
Me, I'd much rather live with duplicates than have mailing lists fuck up
my ability to reply to the author of a message without my "reply to
author" command suddenly (and surprisingly) behaving like my "reply to
all" command. I just don't see the duplicates problem as being much of
a problem, nor do I see much of a need for a third reply-to-some-
default-set-of-recipients command. But that's just me; if you want to
standardize and implement them, more power to you -- just don't break or
invalidate the installed base along the way!
(I'm sure that there's a special circle in hell for mailing list
managers who set the Reply-To field to point at the list.)
--
Jamie Zawinski jwz(_at_)netscape(_dot_)com
http://www.netscape.com/people/jwz/
What the world needs now is killfiles that actually kill. -Craig Dickson