Re: [abnf-discuss] ABNF colloquialism for end-of-line

[Carsten Bormann <[e-mail]>]

On Nov 16, 2017, at 23:31, Dave Crocker <[e-mail]> wrote:
> 
> Carsten,
> 
> On 11/15/2017 6:35 PM, Carsten Bormann wrote:
>> Hi Dave,
>> On Nov 15, 2017, at 23:37, Dave Crocker <[e-mail]> wrote:
>>> 
>>> Given that the thread in CBOR says 'matching rules', I'm guessing that the goal here is to describe freeform data coming from the net.  Hence, requiring a simple, canonicalized data form is not appropriate.  (This is an essential point; if it's not correct, then what follows won't be either.)
>> The thread title unfortunately is misleading.
>> The ABNF is not for on-the wire packets, but for defining the syntax of the CDDL language (which then defines the syntax of the on-the-wire data items).
>> So this ABNF is about files on computers, which probably run a form of Linux/Unix or Windows (and very likely not pre-2001 classic MacOS).  So
> 
> I take your point, but suspect there is still an issue.  At the least, being clear /and explicit/ about this in the specification document(s) will be helpful.

Right.


> The issue I suspect is the intended portability of the file.  If the file is intended to be portable, then it, too, needs to be in a canonical form.  It's a type of 'over the wire' even though it isn't part of a wire protocol.

Well, software development has focused on Unix line ends, tolerating DOS line ends in some spaces, for a while.
That seems to work for so many languages, we can just emulate that.

Let’s take a page from RFC 7950 (YANG 1.1):

   line-break          = CRLF / LF

That is essentially the same I was proposing, but the explicit name “line-break” is probably better than NL by some.

> This, then, would require separate translation from native, local representation to the canonical form.  But that's a pretty simple definition effort.

Again, I think that most source control systems and programmers’ editors know how to do that.

>>    EOL = [CR] LF
>> is probably the right way to describe line ends for these files.
> 
> Possibly, unless folk really want
> 
>   EOL = *CR LF

We don’t want to tolerate more than one CR here; these would be isolated CRs in todays line end worlds.

> So while there is an historical basis for saying EOL, I'd think that in this context, it would sufficient and simpler just to have:
> 
>  WS = SP / CR / LF

That generally works, except in certain strings, where it is good to identify actual line ends.

> (why not also include TAB?)

I can’t find the string TAB in RFC 20; I think you probably mean HT.
HT is evil(**) and, approximately since the time 300 bit/s LA36 terminals(***) went out of use, should never be used(*).
Easy fix as applied here: Simply don’t allow HT in specification source files.

Grüße, Carsten

(*) Outside certain very sheltered environments such as Linux Kernel development.
(**) RFC 7386/7396 should be proof enough here.
(***) https://archive.org/details/handbookofintera00duan p. 82; additionally insert fond memories here…