[abnf-discuss] Deviant form of ABNF in RFC 989/1049/1341/1521/2045/2397

[Carsten Bormann <[e-mail]>]

I have a question that predates my active time in the IETF.

(I don’t think this question will have a practical impact on the evolution of ABNF, but there are still a few documents that need to reference RFC 2045 or RFC 2397, and the answer would be good to know when doing that.)

Before ABNF was standardized (starting with RFC 2234, now in RFC 5234), some RFCs were published with a form of BNF that is claimed to be the one from RFC 822 (without the # operator), but has at least one glaring difference: It uses “:=“ instead of “=“ to separate rule name from rule content.
Of course, this makes ABNF tools error out.

In a recent discussion about a document, I wrote:

> RFC 2397 uses a form of BNF that is taken from RFC 822, but with some gratuitous changes.
> This form was already used in RFC 2045 and its predecessors RFC 1521 and RFC 1341.  I do not know the origin of these changes (I find them first in RFC 989 and 1049); they are not mentioned in Section 2 of RFC 2045/1521/1341 — the authors may not have been fully aware that they were changing the notation.  Interestingly, the changes are not used in RFC 2396, which uses the original RFC 822 notation in its recently published RFC 2234 “ABNF” form.  In summary, the [BNF used in] RFC 2397 is “almost” ABNF.

Does anyone remember the history that made this alternative form appear, teeter on for a while, and then pretty much go away?
Was there any other difference apart from “:=“?

Grüße, Carsten