Brian, I think your note goes to how important it is to write clearly
and to get a lot of eyes on it before we publish it. Well-written
documents, with or without 2119 key words, and with or without
lower-case look-alikes, can still be clear. Fuzzily written documents
will be fuzzy.
In particular:
they mean? It can be very unclear. If a node receives a message containing
an element covered in the spec by "allowed" instead of "OPTIONAL", is the
receiver supposed to interoperate or to reject the message?
Well, this is where 2119 advises that we *use* the key words when
interoperability is at stake. It's fine to be fuzzy when it doesn't
matter, though even then, I'd argue for more explanation:
Every frobotz MUST contain a valid bleeg. The glorp field in the
frobotz is an unsigned integer that is normally between 0 and 666,
inclusive. Values greater than 666 are allowed, but recipients
using older software might not be able to handle such values.
...
When processing a frobotz that does not meet the requirements in
section 3.1.4, it is permissible to reject the frobotz outright, or to
attempt to process the parts of it that make sense; the choice is
an implementation decision. However, any frobotz that does not
contain a valid bleeg MUST be rejected.
That sort of thing.
Barry
On Mon, Mar 28, 2016 at 3:58 PM, Brian E Carpenter
<brian(_dot_)e(_dot_)carpenter(_at_)gmail(_dot_)com> wrote:
There are times when I think RFC2119 was a really bad idea, despite it having
become probably the most frequently cited RFC (inside and outside the IETF).
It seems to create as much confusion as it avoids.
There are four words whose RFC2119 meaning is different from the dictionary
meaning: should, recommended, may and optional. Having special typography
for them is useful, because it signals the RFC2119 meanings. But if a spec
uses, for example, a mixture of SHOULD and should, who knows what the authors
intended? To that extent, the proposed clarification is helpful.
The other words (must, shall, required, not) mean what they always mean.
The only argument for upper-casing them is aesthetic symmetry. If a spec
uses alternatives like mandatory, necessary or forbidden, they are just as
powerful.
So
these definitions are only meaningful if the words are capitalized
can be applied to should, recommended, may and optional if we want,
but strictly doesn't apply to must, shall, required, not, mandatory,
necessary, forbidden, need, or any other such words.
Where we can get into real trouble is if a spec contains should, recommended,
may and optional *plus* other non-categorical (fuzzy) words like ought,
encourage, suggest, can, might, allowed, permit (and I did not pull those
words out of the air, but out of draft-hansen-nonkeywords-non2119). What do
they mean? It can be very unclear. If a node receives a message containing
an element covered in the spec by "allowed" instead of "OPTIONAL", is the
receiver supposed to interoperate or to reject the message?
If we are issuing guidance, it should probably include a specific warning
to use any such fuzzy words with extreme care.
Brian
On 29/03/2016 03:13, Scott O. Bradner wrote:
one minor tweak
On Mar 28, 2016, at 10:09 AM, Barry Leiba
<barryleiba(_at_)computer(_dot_)org> wrote:
The wishy washy descriptive rather than proscriptive language in the
abstract was because I,
the IESG and the community were not of one mind to say that the use of
such capitalized
terms should be mandatory - quite a few people felt that the english
language was at
least good enough to convey the writer’s intent without having to
aggrandize specific words.
Thus the abstract basically was saying: if you want to use capitalized
words here is a standard
way to say what they mean
Ah. Then perhaps the clarification needs to go a little further and
make this clear:
- We're defining specific terms that specifications can use.
- These terms are always capitalized when these definitions are used.
these definitions are only meaningful if the words are capitalized
- You don't have to use them. If you do, they're capitalized and
their meanings are as specified here.
- There are similar-looking English words that are not capitalized,
and they have their normal English meanings; this document has nothing
to do with them.
...and I'd like to add one more, because so many people think that
text isn't normative unless it has 2119 key words in all caps in it:
- Normative text doesn't require the use of these key words. They're
used for clarity and consistency when you want that, but lots of
normative text doesn't need to use them, and doesn't use them.
Barry