Date: 1 Aug 1996 14:56:17 -0000
Message-ID: <19960801145617.8632.qmail@koobera.math.uic.edu>
From: djb@koobera.math.uic.edu (D. J. Bernstein)
To: drums@cs.utk.edu
Subject: Re: Allowing and disallwoing LWSP between elements in ABNF

Here's the data flow in an RFC 822 parser:

   message
    |
    | look for a blank line
    V
   header (+ body)
    |
    | handle continuation lines
    V
   header field (+ header field + ...)
    |
    | look for colons
    V
   body (+ name)
    |
    | tokenize
    V
   sequence of atoms, specials, qstrings, literals, comments, spaces
    |
    | throw away comments and spaces
    V
   sequence of atoms, specials, qstrings, literals
    |
    | parse
    V
   higher-level data

Some people want to combine the last few steps. I don't understand why.
Instead of saying ``throw away comments and spaces'' once, we'd have to
say it in every single parsing rule. Adding a special notation for that
is begging the question: what's the benefit supposed to be? Why can't we
just keep the steps separate?

> We also need the grammar to clearly state exactly which kinds
> of production syntax permit LWSP between elements

The problem you're addressing comes up only a few times. It's easier to
handle it directly---

   minute = "00" / "01" / ... / "59"

---than to spend several paragraphs explaining some fancy meta-solution.
A sensible implementor is just going to apply atoi() to the atom anyway.

---Dan