[rfc-i] Re: Normative ABNF [was Re: Re: Normative information in RFC imagery]

[Eric Rescorla <[e-mail]>]

On Sat, May 17, 2025 at 3:00 PM Brian E Carpenter <
[e-mail]> wrote:

> On 18-May-25 09:19, Carsten Bormann wrote:
> > On 17. May 2025, at 22:36, Brian E Carpenter <
> [e-mail]> wrote:
> >>
> >> However, RFC 9651 explains
> >> exactly why it uses all that English:
> >>
> >> " Appendix C. ABNF
> >>
> >> This section uses the Augmented Backus-Naur Form (ABNF) notation
> [RFC5234] to illustrate the expected syntax of Structured Fields. However,
> it cannot be used to validate their syntax because it does not capture all
> requirements.”
> >
> > I’m not sure that the text there accurately reflects what the authors
> wanted to say.
> > I’m not aware of any ABNF anywhere that “captures all requirements”, so
> this is a pretty weird argument.
> > (I already mentioned that “validating” the syntax is only a part of what
> needs to be done here, so maybe they meant that — but that doesn’t make the
> ABNF any less valuable.)
>
> I agree with you. It's been pretty clear since BNF was first used
> (strictly speaking, it was just BF for Algol 58, but that was well before
> my time) that some terminations in BNF *require* explanation in human
> language. That doesn't make the formalism useless.
>

Yes. I'm honestly kind of puzzled by this whole discussion.

Formal languages like (A)BNF, ASN.1, TLS PDUs, etc. specify the semantic
productions that are compatible with a given protocol (language), but they
don't
provide the meaning of those productions; that must be supplied in some
other
fashion. In protocols that's typically in text, whereas in a compiler it's
the code
associated with the production.

It's also the case that some productions which would be permitted by the
grammar are actually impermissible because it's too difficult to specify
them
in the grammar, so the grammar describes a superset of the language.
That's not ideal, but also is not a disaster.

-Ekr