* [sup-talk] Reply calculation
@ 2009-07-25 18:10 Edward Z. Yang
2009-07-25 18:24 ` Edward Z. Yang
0 siblings, 1 reply; 8+ messages in thread
From: Edward Z. Yang @ 2009-07-25 18:10 UTC (permalink / raw)
I think sup's "To" address calculation algorithm is wrong, and
messes up in some important edge-cases. The two edge
cases I have in mind are:
* When a mailing list sends an unsubscribe/subscribe notification
* When someone posts to a mailing list with an explicit Reply-to
header.
In both of these cases, the mail will commonly have an explicit
Reply-to header. However, Sup will by default disregard this
header in favor for List-id, which is *wrong*.
You can guess, this has bitten me several times.
I think the correct behavior is to only use List-id when Reply-to
is not explicitly set, since List-id is "magical" whereas Reply-to
is commonly explicit.
Cheers,
Edward
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-25 18:10 [sup-talk] Reply calculation Edward Z. Yang
@ 2009-07-25 18:24 ` Edward Z. Yang
2009-07-25 19:03 ` Edward Z. Yang
0 siblings, 1 reply; 8+ messages in thread
From: Edward Z. Yang @ 2009-07-25 18:24 UTC (permalink / raw)
Here is a patch that changes the behavior:
From 58a8d325286703f9057dd5d07094d0c6a061377c Mon Sep 17 00:00:00 2001
From: Edward Z. Yang <ezyang at mit.edu>
Date: Sat, 25 Jul 2009 14:17:08 -0400
Subject: [PATCH] Use Reply-to if it is explicitly set.
Previously, Sup would ignore an explicit Reply-to
in favor of List-id. This is the wrong behavior for
the following cases:
* List subscribe and unsubscribe messages will commonly
have List-id set, but set Reply-to for their own
nefarious purposes (the "you should be able to reply
to this email and have it work")
* People will sometimes send mail to a list with an
explicit Reply-to header to get responses offlist.
This brings Sup's behavior more inline with traditional
mail clients. Since most lists already set Reply-to
(and most mail clients don't respect List-id), there will
probably be no breakage.
Signed-off-by: Edward Z. Yang <ezyang at mit.edu>
---
lib/sup/modes/reply-mode.rb | 8 +++-----
1 files changed, 3 insertions(+), 5 deletions(-)
diff --git a/lib/sup/modes/reply-mode.rb b/lib/sup/modes/reply-mode.rb
index c79c5db..ca423c1 100644
--- a/lib/sup/modes/reply-mode.rb
+++ b/lib/sup/modes/reply-mode.rb
@@ -81,10 +81,8 @@ EOS
AccountManager.default_account
end
- ## now, determine to: and cc: addressess. we ignore reply-to for list
- ## messages because it's typically set to the list address, which we
- ## explicitly treat with reply type :list
- to = @m.is_list_message? ? @m.from : (@m.replyto || @m.from)
+ ## if someone overrode reply-to, it was for a good reason
+ to = (@m.replyto || @m.from)
## next, cc:
cc = (@m.to + @m.cc - [from, to]).uniq
@@ -115,7 +113,7 @@ EOS
} unless not_me_ccs.empty?
@headers[:list] = {
- "To" => [@m.list_address.full_address],
+ "To" => [@m.replyto.full_address || @m.list_address.full_address],
} if @m.is_list_message?
refs = gen_references
--
1.6.3.3
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-25 18:24 ` Edward Z. Yang
@ 2009-07-25 19:03 ` Edward Z. Yang
2009-07-25 19:15 ` Edward Z. Yang
0 siblings, 1 reply; 8+ messages in thread
From: Edward Z. Yang @ 2009-07-25 19:03 UTC (permalink / raw)
Excerpts from Edward Z. Yang's message of Sat Jul 25 14:24:15 -0400 2009:
> Here is a patch that changes the behavior:
Actually, the patch is not quite correct. Let me think about this
some more.
Cheers,
Edward
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-25 19:03 ` Edward Z. Yang
@ 2009-07-25 19:15 ` Edward Z. Yang
2009-07-27 16:49 ` William Morgan
0 siblings, 1 reply; 8+ messages in thread
From: Edward Z. Yang @ 2009-07-25 19:15 UTC (permalink / raw)
After thinking about this some, I think that the only way
to reasonably handle all corner cases is to explicitly ask
the user for a choice in corner cases. Having the default
subtly change based on some non-obvious attributes of an
email is a great way to break muscle memory.
Myself, I have :user in my reply-to.rb hook for now.
Cheers,
Edward
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-25 19:15 ` Edward Z. Yang
@ 2009-07-27 16:49 ` William Morgan
2009-07-27 17:09 ` Edward Z. Yang
0 siblings, 1 reply; 8+ messages in thread
From: William Morgan @ 2009-07-27 16:49 UTC (permalink / raw)
Reformatted excerpts from Edward Z. Yang's message of 2009-07-25:
> After thinking about this some, I think that the only way to
> reasonably handle all corner cases is to explicitly ask the user for a
> choice in corner cases.
What was the breakage when favoring reply-to over list-id? I was buying
your arguments...
Is it possible to identify these corner cases? Is it always when there's
a reply-to and a list-id both set?
--
William <wmorgan-sup at masanjin.net>
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-27 16:49 ` William Morgan
@ 2009-07-27 17:09 ` Edward Z. Yang
2009-07-28 15:48 ` William Morgan
0 siblings, 1 reply; 8+ messages in thread
From: Edward Z. Yang @ 2009-07-27 17:09 UTC (permalink / raw)
Excerpts from William Morgan's message of Mon Jul 27 12:49:11 -0400 2009:
> What was the breakage when favoring reply-to over list-id? I was buying
> your arguments...
First and foremost, s/list-id/list-post/ in my previous posts; I was quoting
the wrong field and double checked message.rb just now.
I received mail from a mailing list as a call for applications that should
not be sent back to the list. The mail was constructed with headers like:
From: correct@example.com
To: correct at example.com
Reply-To: correct at example.com
List-Post: mailto:incorrect at example.com
Sup detects List-Post, categorizes the message as a list message, and then
sets the default reply mode as list, which results in List-Post being used
as the to address.
> Is it possible to identify these corner cases? Is it always when there's
> a reply-to and a list-id both set?
Unfortunately, mailing list administrating is notoriously broken. I'm not
sure at all what the right solution is. Take for example this other case:
From: person@example.com
To: list at example.com
Reply-To: person at example.com
List-Post: mailto:list at example.com
Reply-To, in this case, was set by the mailing list server. This makes
having Reply-To be the end-all be-all kind of spurious. If we try
to make Sup do the Right Thing (TM), you probably want to send mail to
the list as a whole, which means you do want to either "reply all" semantics
or "list post" magic. ("reply all" semantics are what you see in traditional
mail clients when you hit the "reply all" button.)
However, consider the next case:
From: persona@example.com
To: list at example.com
Cc: me at example.com
Which is when someone else hit "Reply all" and you got CC'ed. This means
that the mail never passed through the mailing list agent, the List-Post/Reply-To
headers never got set, and the only way to tell that you should reply to the
whole list is to explicitly ask for "Reply all" semantics (Sup defaults to
"Reply" semantics, which is damn confusing if you're not paying attention).
The core problem is that subtle changes in state should not require the user
to do things differently; it breaks muscle memory and makes mistakes easy.
We could try to make it so that Sup requires /no/ user intervention, but
this is seems to be an AI-hard problem. Then, the logical other direction,
is to make the interface as consistent as possible.
Cheers,
Edward
PS. Djb wrote an article along these lines:
http://cr.yp.to/proto/replyto.html
I've skimmed it but I'm kind of skeptical about client support and not sure
if it actually gives useful recommendations.
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-27 17:09 ` Edward Z. Yang
@ 2009-07-28 15:48 ` William Morgan
2009-07-28 18:34 ` Edward Z. Yang
0 siblings, 1 reply; 8+ messages in thread
From: William Morgan @ 2009-07-28 15:48 UTC (permalink / raw)
Reformatted excerpts from Edward Z. Yang's message of 2009-07-27:
> From: correct at example.com
> To: correct at example.com
> Reply-To: correct at example.com
> List-Post: mailto:incorrect at example.com
>
> Sup detects List-Post, categorizes the message as a list message, and then
> sets the default reply mode as list, which results in List-Post being used
> as the to address.
So in this case, following reply-to is correct.
> Unfortunately, mailing list administrating is notoriously broken. I'm not
> sure at all what the right solution is. Take for example this other case:
>
> From: person at example.com
> To: list at example.com
> Reply-To: person at example.com
> List-Post: mailto:list at example.com
>
> Reply-To, in this case, was set by the mailing list server.
In this case, I'd argue that this means the list administrator wants the
default reply behavior to be to the individual and not the list. So I'd
again prefer Sup default to the reply-to address rather than the list
address. (With the caveat that this is overrideable by hooks on a
per-list basis, so if it's a matter of an incompetent list
administrator, or simply disagreeing with them, one can override this
behavior for this list.)
> However, consider the next case:
>
> From: persona at example.com
> To: list at example.com
> Cc: me at example.com
>
> Which is when someone else hit "Reply all" and you got CC'ed. This means
> that the mail never passed through the mailing list agent, the
> List-Post/Reply-To
> headers never got set, and the only way to tell that you should reply to the
> whole list is to explicitly ask for "Reply all" semantics (Sup defaults to
> "Reply" semantics, which is damn confusing if you're not paying attention).
There's not much to be done in this case, EXCEPT that if you receive
more than one copy of the message, you should keep the list header
around. Then the only time you're in a funny situation is when you've
received the first but not the other.
This is presumably why Mutt had you register your mailing list addresses
explicitly, which I always found a little irritating.
(Or to have Sup keep around email addresses known to belong to lists,
and match those in the To: field against that, which seems significantly
more complicated.)
> The core problem is that subtle changes in state should not require the user
> to do things differently; it breaks muscle memory and makes mistakes easy.
I agree with this in principle, but I see addressing a message as a
fundamental part of composing it. You can remove the notion of a smart
default reply-to address from your Sup, if you like, by using the
reply-to hook.
And as for the default, I think I'm of the opinion that setting the
default reply address so as to obey the reply-to is correcter than
anything else (including whatever Sup does currently).
--
William <wmorgan-sup at masanjin.net>
^ permalink raw reply [flat|nested] 8+ messages in thread
* [sup-talk] Reply calculation
2009-07-28 15:48 ` William Morgan
@ 2009-07-28 18:34 ` Edward Z. Yang
0 siblings, 0 replies; 8+ messages in thread
From: Edward Z. Yang @ 2009-07-28 18:34 UTC (permalink / raw)
Excerpts from William Morgan's message of Tue Jul 28 11:48:48 -0400 2009:
> > Unfortunately, mailing list administrating is notoriously broken. I'm not
> > sure at all what the right solution is. Take for example this other case:
> >
> > From: person at example.com
> > To: list at example.com
> > Reply-To: person at example.com
> > List-Post: mailto:list at example.com
> >
> > Reply-To, in this case, was set by the mailing list server.
>
> In this case, I'd argue that this means the list administrator wants the
> default reply behavior to be to the individual and not the list. So I'd
> again prefer Sup default to the reply-to address rather than the list
> address. (With the caveat that this is overrideable by hooks on a
> per-list basis, so if it's a matter of an incompetent list
> administrator, or simply disagreeing with them, one can override this
> behavior for this list.)
Nods.
> > However, consider the next case:
> >
> > From: persona at example.com
> > To: list at example.com
> > Cc: me at example.com
> >
> > Which is when someone else hit "Reply all" and you got CC'ed. This means
> > that the mail never passed through the mailing list agent, the
> > List-Post/Reply-To
> > headers never got set, and the only way to tell that you should reply to the
> > whole list is to explicitly ask for "Reply all" semantics (Sup defaults to
> > "Reply" semantics, which is damn confusing if you're not paying attention).
>
> There's not much to be done in this case, EXCEPT that if you receive
> more than one copy of the message, you should keep the list header
> around. Then the only time you're in a funny situation is when you've
> received the first but not the other.
This situation is not that uncommon; many mailing list software will notice
that someone is on a CC list and will purposely omit them.
> I agree with this in principle, but I see addressing a message as a
> fundamental part of composing it. You can remove the notion of a smart
> default reply-to address from your Sup, if you like, by using the
> reply-to hook.
This is true. However, when you compose a non-reply message, Sup prompts
you for To, Cc and Subject, because they are such fundamental components
of all messages you would write (well, maybe not so much Cc) and the
user always has to make a judgment call there.
> And as for the default, I think I'm of the opinion that setting the
> default reply address so as to obey the reply-to is correcter than
> anything else (including whatever Sup does currently).
Fair enough. You may get some complaints because this /will/ break muscle
memory, and still presents the possibility for messages to have different
behavior in subtle ways, as detailed in the Cc versus mailing list example
from above.
Cheers,
Edward
^ permalink raw reply [flat|nested] 8+ messages in thread
end of thread, other threads:[~2009-07-28 18:34 UTC | newest]
Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2009-07-25 18:10 [sup-talk] Reply calculation Edward Z. Yang
2009-07-25 18:24 ` Edward Z. Yang
2009-07-25 19:03 ` Edward Z. Yang
2009-07-25 19:15 ` Edward Z. Yang
2009-07-27 16:49 ` William Morgan
2009-07-27 17:09 ` Edward Z. Yang
2009-07-28 15:48 ` William Morgan
2009-07-28 18:34 ` Edward Z. Yang
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox