Re: [websec] [apps-discuss] ABNF references (was RE: AppsDir review of draft-ietf-websec-strict-transport-sec)

[Brian E Carpenter <[e-mail]>]

Roy,

On 2012-05-02 20:32, Roy T. Fielding wrote:
> On May 2, 2012, at 10:48 AM, Murray S. Kucherawy wrote:
>> 2) There's a common axiom that says it's safer to refer to a definition rather than to copy it.
> 
> I think we should recognize that as a false axiom and move on.
> 
> We should refer to orthogonal definitions that are subject to
> independent change control -- e.g., protocol elements that are
> defined in another spec because they change at a different
> rate than the referring spec or are used by multiple specs.
> 
> We should copy a definition by value if the referring spec
> depends on the definition (does not allow the parser to change
> even if some other spec were to define it and later extend it).

But that is contrary to the general principle in the IETF of
using normative references and *not* replicating normative material,
to avoid mistakes.

Both approaches have their advantages and disadvantages, but making
ABNF an exception seems problematic, or at least a decision that can't
be taken at Area level.

> My preference is to not use prose definitions at all -- I used
> them as a crutch when I first started writing IETF specs in 1994,
> and they burned me every time.
> 
> And if we go down the slippery slope, I would love to have a
> formal definition of set reduction, as in
> 
>    ALPHA = ALPHANUM - DIGIT
> 
> since I very commonly need rules that only differ by one or two
> characters being removed from the allowed set.

Would you call that "disinheritance" ?

    Brian