Re: Seeking clarification of ABNF "name formation" reference

[Julian Reschke <[e-mail]>]

Clive D.W. Feather wrote:
> If "value range" means the "%x20-3F" construct, then that has higher
> precedence than comment, I would suggest. More significantly, repetition
> has lower precedence than grouping and optional: this is shown firstly by
> the formal grammar, and secondly because "*(A B)" means zero or more sets
> of A-then-B, while if repetition had higher precedence then it would mean
> zero or more As followed by one B (that is, the same as "((*A) B)").

Maybe we could have the individual items in the precedence list 
reference either subsections in section 3, or rule names in section 4?

>> The reference to "Name formation" has been in this form since RFC 2234.
> 
> I assumed it was <prose-val>.
> 
> However, as Philip points out, it's not defined anywhere in sections 2 or
> 3. Rather, 2.1 implies that the "<rulename>" form can occur on the left of
> the <defined-as> if the author wants. This means that the grammar should
> have the following definitions for rulename:

Maybe we should check whether that is used in practice, and what the 
parser implementations do with it.

>     rulename = ub-rulename / "<" ub-rulename ">"
>     ub-rulename = ALPHA *(ALPHA / DIGIT / "-")
> 
> I've used "ub-rulename" to mean "unbracketed rulename"; feel free to change
> it.
> 
> Is it intentional to allow <rulename> to end in a hyphen?
> 
> As well as needing a textual description of <prose-val>, there are two other
> considerations:
> (1) the comment implies that "<" cannot occur in a <prose-val>, but the
>     formal grammar allows it;

Right.

> (2) with the above change, the grammar is ambiguous, because "<rulename>"
>     can be either a <rulename> or a <prose-val>. This needs disambiguating
>     in some way, either in a comment or by requiring a prose-val to contain
>     a non alphanumeric character (which might be space).

Best regards, Julian