Re: Case distinctions as theoretical exercise

[Doug Barton <[e-mail]>]

Thanks for the response John, more below.


On 03/13/2016 08:10 AM, John C Klensin wrote:
> Paul, Doug,
>
> I want to respond to two issues/questions only right now and am
> going to do so in two separate messages.  Perhaps someone else
> will get to the other questions before I do -- I think they have
> been answered and explained several times already.   This is
> therefore message 1 of 2.
>
> I don't know why we (that "email community") and you (Paul) are
> having so much trouble communicating, but

Having read a bunch of the DANE archives today it does seem that there 
is a problem in communication, but honestly it doesn't seem all that 
one-sided. :)

> --On Saturday, March 12, 2016 4:09 PM -0500 Paul Wouters
> <[e-mail]> wrote:
>
>> ...
>> However, the email community experts themselves have already
>> stated that finding an email server compliant to case 2 is a
>> theoretical exercise only.
>
> I can't speak for others, but I certainly didn't say that.
>
> The specs (and generally-accepted good practice) recommend using
> prohibition, aliasing (of some flavor), or other mechanisms to
> avoid making subtle distinctions in addresses that are likely to
> be confusing to users and/or to stress expectations about human
> memory and precision beyond reasonable limits.  I would
> certainly recommend that a final delivery system not provide a
> mailbox named [e-mail] to a different recipient than a
> mailbox named [e-mail].  I am agnostic about whether
> they should be the _same_ mailbox or associated with the same
> key: some recipient might plausibly use an obscure form of an
> address as a mail classification tool or as part of a very cheap
> sender-checking model.  Such models, whether done that way or by
> subaddresses or the like, are surprisingly effective against
> casual bulk attacks even if much less so against focused and
> determined ones.

I'm perfectly willing to grant that some systems do this. Paul's hashing 
scheme accounts for this, and CNAMEs can be used to handle any aliases 
that are desired.

Further, I think that many of the critics of the draft are ignoring the 
fact that in the overwhelming number of cases the sending user is going 
to have an example of the receiving user's e-mail address, the most 
likely case being that they are replying to a message they already have 
in their MUA. Thus it's overwhelmingly likely that the sending user will 
already have a representation of the receiving user's local part that 
the receiving user considers to be canonical, and thus is likely to have 
placed an OPENPGPKEY RR for.

> The same thing could be said for other distinctions.  For
> example, [e-mail] is a popular form for a
> local-part and has been used on some systems for years.  IIR,
> there is even a standards-track spec that recommends it.  I
> would certainly not suggest that a system that uses that form
> for local-parts also support separate mailboxes, assigned to
> separate recipients, named like "[e-mail]".
> Whether is it better to not support the latter at all, to make
> it an alias for the [e-mail] form, or to
> algorithmically remove the special characters(s), possibly also
> making [e-mail] point to the same mailbox, is a
> matter of taste: the specs are (deliberately) mute on the
> subject.

Agreed, see above.

[snip]

> Now, having said all of that, I want to note that this has
> little or nothing to do with the protocol requirements of
> draft-ietf-dane-openpgpkey-08.   At least as I read it, that I-D
> simply applies a hash function to the mailbox name as given and
> does not attempt to find "equivalent" or "canonical" names.
> That means that the very worse case is a false negative about
> finding a key.  You want to experiment with that, fine -- I
> won't lose any sleep over it.

Good to hear! :)

> I do believe that, if that is the
> protocol spec, there is a lot of surplus and confusing text in
> the I-D that should just come out.

FWIW, I agree that the draft in its current state needs some tightening, 
as it seems to have some leftover tidbits from previous versions. I also 
think that the approach suggested is wrong, and would like to see a 
parallel approach codified (which I will comment on further in another 
message).

Doug