On 07/02/2017 15:45, Barry Leiba wrote:
Hi, Stewart.
The above text raises an interesting problem. If the update system works
then
the text should read [RFC2119]. If the update system does not work than
the text needs to be [RFC2119],[RFCxxxx] as shown, but we also need to move
to a system where we always list the update set at the time of publication,
It took me a while to get what you mean here, but I think I have it:
you're saying that because this document (RFC xxxx) UPDATES RFC 2119,
*any* document that cites RFC 2119 and is published after this one
must automatically be taken to refer to this one and must follow the
terms of this one. So why have the double citation?
You're right, as far as that goes, but if it's important to know for
sure whether a document is following the terms of RFC xxxx or not (and
see below about that), then I think it's important to be explicit
about it. It's a question of clarity: the reader of another document
should not have to wonder, should not have to check publication dates,
should not be uncertain. Having the citation there explicitly is good
for clarity.
In fact, we *do* often (though not always) cite update documents when
we're explicitly talking about a feature that was updated. I think we
do it when calling the reader's attention to the update is
particularly important.
I am afraid I disagree. Either the update system works or it does not.
If you think
people do not follow through the updates, then we need to always include
update
list, or we need a system that always gives them the update when they
request the
base document. So for example someone asking for RFC2119 would find this
automatically but distinctly appended to it.
Ironically in this case the result is no text change to the document
making the
reference since a MUST is a MUST.
Now, as to the other point:
I am also somewhat curious about the practical implication of
misinterpreting
MUST as must, and must as MUST.
For example: if A receives a foo packet, it MUST/must reply with a bar
packet
The interoperability considerations are identical, with but with the
advantage
that MUST draws the eye of the reader to the point and reducing the chance
that they will miss it. Similarly with the other keywords. So is there
really
a problem to be fixed here?
This did come up in earlier discussion about "MUST". Sure, it's
possible that there really isn't a reason to define "MUST" in BCP 14
because the English meaning is clear enough. But "SHOULD" and "MAY"
do need these definitions: the English meanings don't accurately
convey the BCP 14 meaning. And even with "MUST", the BCP 14 meaning
explicitly says that it's a protocol requirement that affects
interoperability or security, and we do seem to think that making that
distinction is important.
In any case, the current BCP 14 defines "MUST", and it's specifically
not a goal of this document to change that older decision.
No, the point is that either type of must is exactly the same
instruction - "do it or it's broken".
Similarly when should is given it is advisory, and there is no forcing
how seriously the
individual takes that advice.
So if someone mistakes a MUST for a must, I don't see how the on the
wire behaviour
can possibly change. Same for the other key words.
- Stewart
Barry