Re: [abnf-discuss] [art] [Technical Errata Reported] RFC7601 (5435)

[Peter Occil <[e-mail]>]

You have pointed to an unexpected weakness of ABNF.  As a result, it's unfortunate that many ABNF productions in RFCs can't generally be used as is in order to build a parser, since ABNF itself has no strong concept of matching.  

For example, RFC 5234 sec. 3.2 is silent on which order alternative productions are to be matched (although this silence can be useful, I admit, in order for the incremental alternatives feature, sec. 3.3, to work).  The production "received-token" in RFC 5322 is illustrative:

   received-token  =   word / angle-addr / addr-spec / domain

Both "word" and "domain" include the production "atom".  Thus, whether a received-token parser parses an "atom" as a "word" or as a "domain" depends on the context in which the "atom" appears, the kind of parser (greedy or not), or whether earlier alternatives take precedence over latter alternatives or vice versa, none of which is defined in RFC 5234. "Obviously", if the "atom" is part of a domain, then the received-token should be the whole "domain" regardless of the details just mentioned.

[ Additional context: https://www.ietf.org/mail-archive/web/art/current/msg00571.html ]

From: John R. Levine
Sent: Sunday, July 22, 2018 10:33 PM
To: Peter Occil
Cc: Murray Kucherawy; [e-mail]; Alexey Melnikov; [e-mail]; [e-mail]
Subject: RE: [art] [Technical Errata Reported] RFC7601 (5435)

> And also, the erratum I submitted can be correct for handling certain 
> parsers that skip the CFWS in the “pvalue”, set the pointer to after the 
> CFWS (if any), and move on to parsing the next “CFWS propspec” within 
> the (current) “resinfo” production (in other words, the parser does 
> “greedy” matching).

ABNF doesn't have to be LL(k) or LR(k), so greedy parsers won't work, at 
least not without rewriting the rules to factor out ambiguities.

Regards,
John Levine, [e-mail], Primary Perpetrator of "The Internet for Dummies",
Please consider the environment before reading this e-mail. https://jl.ly