Re: [apps-discuss] draft-yevstifeyev-abnf-separated-lists-02

["Thomson, Martin" <[e-mail]>]

+1 to the rest of this.  Other stuff below.

On 2010-12-17 at 04:40:30, Dave CROCKER wrote:
> Note:  We still do not have statements of support for standardizing a
> mechanism,
> in spite of the continued activity on this thread.  So far, this is a
> fun
> exercise, but that's all.

+1.  I'm still open to being convinced though.  The HTTP-specific extensions to ABNF are a bit of an irritant, as are other such specification-specific extensions.  As it stands, the draft is not enough to convince me.

> { There are many different list forms; the one given here builds in
> certain
> assumptions that do not always apply.  /d }
> 
> 
> >    (*LWS element n*m(*LWS separator *LWS element))

In fact, using more prescriptive prose and specific examples is probably better.  Explain in prose, then provide a few actual examples, with examples of the syntax that conforms to these grammars as well.  The style of RFC 5234 is a good guide; I don't like the fact that this draft strays so far from that.

The entire draft might be summarized so much more succinctly and clearly in this fashion:


--8<--
3.11  Separated list: n*m#s rule

A hash following a repetition indicates a separator that appears between each of the multiple instances of the element.  Any variable or specific repetition can be followed by a separator.  The full form is:

  <a>*<b>#<separator>element

Where <separator> is any valid element - including grouped elements - followed by optional whitespace.

A rule of the form:

   4*7#"," foo

is equivalent to:

   foo 3*6("," foo)

Similarly, a rule of the form:

   *#[*WSP ";" *WSP] foo

is equivalent to:

   [foo *([*WSP ";" *WSP] foo)]

4.1 ABNF Definition of ABNF
	
The <repeat> rule is extended from that in [RFC5234] from:

   repeat = 1*DIGIT / (*DIGIT "*" *DIGIT)

to:

   repeat = 1*DIGIT / (*DIGIT "*" *DIGIT) [ "#" element *c-wsp ]

-->8--

{ I missed the optional c-wsp in my previous comments - that's necessary to separate the two consecutive <rulename> constructs in a repetition.  For instance:

   example = 3*10#comma list-item
}

> In reviewing the way Section 3 of RFC 5234 specifies things, I believe
> that the
> specification, here needs to be made without the square brackets and
> without the
> parentheses.  (I think someone earlier commented about the parens.)

That was probably me.  You are correct.  Based on this, the grammar should be as above.

> >    indicating exactly <n> elements separated by <a>.
> >
> >    The ABNF for this construction is:
> >
> >    n*m#(a)(element) = (n_1*m_1(element 1*SP a) f(element))
> 
>     n*m#a element) = (n_1*m_1(element 1*SP a) f(element))

All this n_1, m_1 and f stuff is confusing and unnecessary.  It relies on what amounts to another modification to the grammar.  The definition of <f> still doesn't make sense.  u-(u-1) is 1, always.

> >    1*3#(";")element indicates the list of elements separated by
> 
> Question to the audience:  are the parentheses needed here?  The
> quotation marks
> would seem to provide sufficient lexical uniqueness, but this is the
> sort of
> nuance I usually get wrong.

Parentheses are not required here - if we assume that the separator grammar conforms to <element>.  They would be permitted, but the draft still does not describe the grammar for the separator.

--Martin