Message-Id: <v03005505ad6966a844c5@resnick1.isdn.uiuc.edu>
In-Reply-To: <19960307233301.19938.qmail@koobera.math.uic.edu>
Mime-Version: 1.0
Content-Type: text/plain; charset="us-ascii"
Date: Sun, 10 Mar 1996 23:30:19 -0600
To: djb@koobera.math.uic.edu (D. J. Bernstein)
From: Pete Resnick <presnick@qualcomm.com>
Subject: Re: Message format document outline
Cc: drums@cs.utk.edu

On 3/7/96 at 5:33 PM, D. J. Bernstein wrote:

>In fact, I propose that this be an explicit requirement for DRUMS.
>822bis should be free to resolve ambiguities, and to say that lots of
>valid 822 messages must not or should not be generated, but _every_
>unambiguously valid 822 message must have the same meaning in 822bis.
>
>The reason, again, is that mail messages stick around forever.
>
>Reactions?

At least in part this is not reasonable. Remember, creating a standard in
the IETF does not mean "come up with some standard that may or may not be
current practice but is the 'right thing to do'". An IETF standard is
documentation of current interoperable practices such that new participants
can interoperate with current implementations. Given that, I believe it
might be explicitly required by our charter that valid 822 messages which
do not conform to current interoperable practice *must* be made illegal in
822bis, or at the very least illegal to send. Now, what we say about
interpretation is an intersting matter, but at the very least we need to
warn people about where some mailers generate illegal (822bis) stuff and
either recommend some behavior, or say that it is implementation dependent.

>> According to the rule as you have it written, we would be
>> *required* to interpret these as continuations,
>
>Yes, just as you're required to do so under 822.
>
>Apparently you feel okay violating the 822 rule. So why would you have
>a problem violating the same rule in 822bis?

Current circumstances appear to indicate, as Chris pointed out, that blank
continuations are more likely the seperater between headers and body than
they are truly blank continuations. If that's true, then the best thing to
do for those "messages which stick around forever" is to interpret them as
seperaters rather than continuations. However:

On 3/7/96 at 8:59 PM, D. J. Bernstein wrote:

>I've seen the following phenomenon from more than one MUA:
>
>   To: one@some.domain,
>
>     two@another.domain (An exceedingly long comment here, so long in fact
>that it goes past the edge)

If this is the case, then we've got an empirical question to answer. Are
there more messages generated by mailers which exhibit the above, or more
that use CRLF-SPACE-CRLF for header/body seperaters? My experience has been
the latter, but the fact of the matter should certainly dictate the
recommendation we make.

>You propose that an 822bis parser be allowed to incorrectly parse these
>valid 822 messages, so that it is free to accommodate some non-822
>messages produced by low-quality gateways. Did I get that right?

Not quite. If the greater percentage of mail out there which have blank
continuations were intended to be valid 822 messages, then this would (as
you imply) be bad. But if the vast majority are in fact produced by bad
gateways which were not intended to be valid 822 messages, then it *should*
parse them the opposite of how 822 says they should be parsed. Are we at
least in agreement on the principle?

Does anyone have a take on the empirical data?

pr

--
Pete Resnick <mailto:presnick@qualcomm.com>
QUALCOMM Incorporated
Home: (217)337-1905 / Fax: (217)337-1980