[conspire] The deal about "Reply-To"

[Rick Moen]

So, Bill Hubbard (who just joined us) innocently raised the much-feared
"Why doesn't this list set Reply-To?" question on a nearby LUG list
where one of the officers has banned all discussion of it, or even of
matters _adjoining_ on that subject, on grounds that the subject leads
to flamewars.

I should hasten to add that it's obvious where he (the officer) is coming
from, and one can only applaud his intention:  Flamewars _are_ notorious
on that subject, and nobody with any common sense wants them.  The part
that seems unreasonable is imposing a blanket (undocumented) ban on _any_
discussion of _any_ merely adjoining topic, including those exchanges
conducted calmly, without repetition or haranguing, and including _new_
observations (i.e., ones seldom or never seen).  That's excessive -- and
a mild affront to those of us who make a _point_ of those latter types
of exchanges specifically to _avert_ dumb, tedious flamewars.
 
But, if those threads aren't allowed in some places run by people who
need to lighten up and not be such control freaks, at least they're
perfectly OK in less heavy-handed places.  Like here.  Which I'll be
doing, below, just to prove the point and explain the issue (since you
asked, Bill H.).

First, I have to apologise for a paragraph that got mangled:  I write
this stuff really quickly, and I did a cut-and-paste screwup:

  [That officer] disapproves of people detailing netiquette matters on
  [the LUG's] mailing list, and (e.g.) on one past occasion referred to
  my saying "Cross-posting is generally a bad idea, and often leads to
  problems" as being a "list cop" and/or as "criticising" a fellow
  poster, the latter of which by public decision of the two immediately
  prior [LUG name] presidents is a misdeed punishable by sanctions
  including summary permanent banning without prior warning from all
  [LUG name] mailing lists.

Er, I meant to say:  1.  The LUG officer disapproves of people
discussing netiquette matters, claiming that's being a "list cop" or
constitutes "criticising" a fellow poster.  The two prior LUG 
presidents declared any such "criticising" a permanent-ban-worthy
offence.  2.  The officer I spoke of (not long ago) classified one of my
posts as "criticising a member" for no better reason than my saying
(carefully being _very_ generic):  "Cross-posting is generally a bad
idea, and often leads to problems."

I pointed out to that guy in follow-up private discussion that I'd done
nothing of the kind, and (several times) asked him to please correct
his (public) claim to that effect.  He never did -- and it's the one
thing I asked of him.  (He _did_ admit that he'd misread my post.)

Anyhow, all of that is part of why I could not answer your question on
the other list:  The politics of that place is such that I get declared
guilty of "flaming" or "cricitising a fellow poster" when all I've done
is pointed out a generic truth or calmly answered someone's question.
Screw that:  We don't have that sort of baggage here, and can do better.


Anyhow:  The IETF's RFC822 specification (the core of the SMTP protocol
definition) defines the Reply-To header as a means for indicating a
different address (different from the From: address) to which any
Reply-to-sender mails from a responding mail client should go.  E.g.,
suppose I know that linuxmafia.com is going to be unreachable because
the machine's about to be rebuilt, and so want any replies to my address
to go to rick at deirdre.net (my alternate address) instead.  I send:

  From: rick at linuxmafia.com
  Reply-To: rick at deirdre.net
  To: a at a.com, b at b.com, c at c.com

If correspondent "a at a.com" does a reply-to-sender, i.e., writes to just
me (as sender) and if his mail client honours my Reply-To header, his
mail goes out like this:

  From: a at a.com
  To: rick at deirdre.net

On the other hand, if he does a reply-to-all (and his mail client
honours my Reply-To), it goes out like this:

  From: a at a.com
  To: rick at deirdre.net, b at b.com, c at c.com

You might be thinking:  What's this about reply-to-sender versus
reply-to-all?  Ah, this is exactly where part of the flamewars
originate:  A lot of non-technical e-mail users don't want to think
about that, and just want magic to ensure that mail goes where they want
it to, and not anywhere else.  But SMTP doesn't have facilities for
that:  They want something that isn't there.

And yes, "Eudora Version 6.1.2.0" (which you use) _does_ do both types
of reply as separate commands.  Check for yourself.


People usually aren't entirely clear on what Reply-To is _supposed_ to
do (described above), but where they _really_ get confused is where you
add mailing list addresses to the picture.  Let's do that (and
understand that rick at linuxmafia.com, a at a.com, b at b.com, and c at c.com are 
all list-members):

  From: rick at linuxmafia.com
  Reply-To: rick at deirdre.net
  To: list at example.com

Subscriber "a at a.com" receives a copy retransmitted to him by the mailing
list software at example.com -- with all the headers still the way
rick at linuxmafia.com wrote them.  Imagine that he does reply-to-sender
(and his mail client honours my Reply-To):

  From: a at a.com
  To: rick at deirdre.net

On the other hand, if he does a reply-to-all (and his mail client
honours my Reply-To), it goes out like this:

  From: a at a.com
  To: rick at deirdre.net, list at example.com

(Some people, again not being very clear on what's going on, will tend
to bemoan user rick getting "duplicate" copies of a at a.com's mail, one
sent directly, and the other sent via the mailing list software.  But
they are assuming that mailing list addresses are somehow treated in
SMTP as special, and that something is supposed to magically determine
that the direct copy is surplus.  They aren't, and they don't.[1])


For whatever reason -- usually, a desire to not have to think about
multiple reply modes -- many people lobby to have mailing list admins
force their mailing lists to "munge" (force) the Reply-To header as part
of the mailing list mechanism.  Let's run through those two example
mails again, except with example.com's listadmin "munging" Reply-To:

Original mail is still:

  From: rick at linuxmafia.com
  Reply-To: rick at deirdre.net
  To: list at example.com

Except now, the mailing list software does _not_ retransmit the message
to subscriber a at a.com unchanged:  It -=discards=- rick at linuxmafia.com's 
Reply-To header (the one intended to ensure that correspondence reaches
an in-service mailbox rather than one out of service for a machine
rebuild).  In its place, it puts a "munged" replacement.  a at a.com gets:

  From: rick at linuxmafia.com
  Reply-To: list at example.com
  To: list at example.com

Imagine that he does reply-to-sender (and his mail client honours 
Reply-To):

  From: a at a.com
  To: list at example.com

Oops!  This was supposed to be a _private_ communication, and now
b at b.com, c at c.com, all the other subscribers, and everyone later
researching the Web archives using Google are getting to see it.  Let's
hope that a at a.com didn't blurt out something likely to get him divorced,
or fired, or punched in the nose.

On the other hand, if he does a reply-to-all (and his mail client
honours Reply-To), it goes out like this:

  From: a at a.com
  To: list at example.com

As you can see, all that's really been accomplished is to
sabotage the originally intended function of Reply-To and
overriding/discarding that header if present, make it almost impossible
to do reply-to-sender at all (as many have found to their horror after
sending extremely private mail to the public's attention), and make
reply-to-all be channeled artificially to see just the list address as a
correspondent.


Please note that I've done basically zero _advocacy_ (arguing,
haranging) in the foregoing:  I've just described in detail how things
work, given the assumption that mail clients respect the Reply-To
header.  Regardless of what one _wants_ to have happen, the above
(unless I've made a major error, which I certainly won't mind people
pointing out) is exactly what happens, and everyone can agree to that.

(If I wanted to be extremely thorough, I could describe a now-expired
_draft_ RFC standard by Daniel J. Bernstein, not broadly accepted but
followed by many including me, to disambiguate the situation using an
additional header called Mail-Followup-To.  But we don't want to go
there.)

The above dispassionate rundown is certainly not a flamewar, but it _is_
tedious and long.  People get tired of seeing it being explained to
newcomers over and over.  Thus, a smart LUG would want to FAQ the
subject (covered in that fashion), not ban it.  

Above and beyond the dispassionate bit, of course lotsa people have
opinions about what should or should not be done with the header on or
with mailing lists.  The fact that this is so, and people's arguments
and counter-arguments, would be dispassionately FAQ by a smart LUG
wanting to be able to refer future inquirers to something and tell them
"Please don't address this subject on our mailing list unless you have
something _new_ to say."


Another root cause of the flamewars is a war of two essays that has been
going on for a number of years:  Chip Rosenthal wrote "Reply-To Munging
Considered Harmful", and then Simon Hill wrote "Reply-To Munging
Considered Useful" as an attempted rebuttal.

Rosenthal's essay was technically accurate but annoyingly and needlessly
polemical -- basically, flamebait.  Hill's rejoinder was, at best,
misleading and commits a number of clear logic errors.  Both essays
increased the noise of online debates greatly, as proponents cited one
or the other (or, oddly, both) as somehow definitive -- usually without
quite understanding them.

For ready reference, you can find both essays, plus my comments, plus
former SVLUG president Marc Merlin's comments, linked from the Linux
Users of Victoria FAQ:  http://www.luv.asn.au/faq/#replytomunge
(I help maintain that FAQ, and supplied a new URL for Hill's essay via
the Internet Archive repository, even though, like Merlin, I think
Hill's analysis is deeply flawed and very misguided.)

Anyhow, that wasn't so painful, was it?  Competent adults _can_ discuss
these things on mailing lists; if someone flies off the handle about it,
well, that's what killfiles are for.  ;->


[1] People who _insist_ that the one and only "Reply" command they've
ever bothered to notice on their mail clients _must_ send any reply to a
list post back to the mailing list can achieve that using macros on
their mail clients (or using procmail, if available).  The recipes for
doing so could be collected and FAQed, too.  And that leaves the
so-called "duplicate" mails:  The mailing list post of mine to which the
officer objected (merely) pointed out how to do that using a per-user
subscription setting in Mailman 2.1 and above -- which, by the way,
defaults to stripping the second copy:  an exception to my general
observation that nothing exists to magically "determine that the second
copy is surplus".  The officer was unhappy that I informed the mailing
list of that, because of it "leading to flamewars".  I can't see the
causal connection, personally, and think calm analysis makes flaming
_less_ likely.  But you decide.