--On Thursday, 12 April, 2007 11:20 -0400 Tony Hansen
<tony(_at_)att(_dot_)com> wrote:
...
This thread is supposed to be about the wording of 1yz's
description. It has wandered way off course back to Issue 17
(different reply codes in multiline responses).
...
Issue 17, the possibility of different reply codes being used
in multiline responses and the question of which one should be
believed, is an example of such an ambiguity. Some people have
taken the sentence beginning "in many cases..." to be
normative and others not. To me, the wording of "in many
cases" is equivalent to a MAY, which is also equivalent to a
MAY NOT. Given that there *are* implementations that have
chosen both interpretations, and we can't remove multiline
responses, I'm now convinced that we're going to need to take
the conservative approach.
I think one thing that is pushing me toward a hard line on this
is that I can take "in many cases..." to mean, not MAY (or MAY
NOT), but that there are some cases in which that would be an
unsafe strategy. In other words, it is (possibly or probably)
wise for a client to not do this unless it somehow knows,
presumably out of band, whether the strategy is safe, even if it
would be ok in many (or most) cases.
This is why it is important to remember that the text in
question is original 821 material, not something that was
introduced by DRUMS into 2821. RFCs were written very
differently in 1981-82 than they were even a decade later, much
less nearly two decades later. In that period, the community
made the shift between a design specification for comment,
implementation, and more comment to standards with conformance
criteria. We had little systematic history of talking regularly
about conformance and mandatory features prior to RFC 1122 and
1123.
And it just makes little sense to try to split hairs over the
exact conformance meaning or interpretation of a comment in an
1982 document that says "in many cases". Had that text been
written in 2001, the situation would be different (and, in all
likelihood, that language would have been changed to a 2119-like
statement during Last Call if not earlier).
best,
john