Re: [port-srv-reg] we need to make progress

[Joe Touch <[e-mail]>]

On 9/6/2010 11:00 PM, Stuart Cheshire wrote:
>> On 6 Sep 2010, at 16:02, Joe Touch wrote:
>>
>> ABNF may be hard for you to understand
...
> The issue I'm trying to raise is not the particular error I may (or
> may not) have found, but the deeper concern that I'm not confident that
> it does not have more errors in addition to that. The issue is that
> unless a reasonable proportion of the IETF community can look at the
> ABNF definition and state with confidence that it is obviously right,
> then we may publish something that's not right.

I believe we need to publish something, esp. since some programmers will 
want to use what we publish - or a derivation thereof - as code to 
validate the strings.

> Processing a computer language generally consists of tokenizing
> followed by parsing. Tokenizing consists of turning text into logical
> units; parsing consists of turning those tokens into the language's
> structure. ABNF is a good way of expressing parsing rules; it is often
> not a good way of expressing tokenizing rules.

The ABNF we are using includes regular expressions, which is the common 
way to express tokens.

> The rules for stating
> what's a valid service name is a tokenizing problem, not a parsing
> problem. The awkward and convoluted ABNF in the document reflects this
> poor fit.

It describes attempts to over-optimize the description to make it 
compact - which happened back in mid-Dec, and you were cc'd in that 
discussion.

I checked that thread, and we had been developing the ABNF up to a 
point, then Olafur suggested we paste in the copy from the "clarify" 
doc, which introduced the error (indicated below).

       SRVNAME = (ALPHA / *([HYPHEN] ALNUM)) /
                        ^
                 (1*DIGIT ((HYPHEN ALNUM) / ALPHA) *([HYPHEN] ALNUM))
       ALNUM   = ALPHA / DIGIT   ; A-Z, a-z, 0-9
       HYPHEN  = %x2d            ; "-"
       ALPHA   = <See [RFC5234]>
       DIGIT   = <See [RFC5234]>

 From our discussions back in Dec, and correcting to make this easier to 
read (From RFC5234), the intended ABNF would have been:

       SRVNAME =  ALPHA *([HYPHEN] ALNUM)
       SRVNAME =/ 1*DIGIT ((HYPHEN ALNUM) / ALPHA) *([HYPHEN] ALNUM)
       ALNUM   = ALPHA / DIGIT   ; A-Z, a-z, 0-9
       HYPHEN  = %x2d            ; "-"
       ALPHA   = <See [RFC5234]>
       DIGIT   = <See [RFC5234]>

This missed the multiple hyphen case, which can be corrected easily:

       SRVNAME =  ALPHA *(*HYPHEN ALNUM)
       SRVNAME =/ 1*DIGIT ((HYPHEN ALNUM) / ALPHA) *(*HYPHEN ALNUM)

Joe