Your first point is precisely why I think the new BNF is somewhat awkward --
boundary markers are always preceeded by a CRLF as well as being followed by
one. The preceeding CRLF is generally a more important characteristic (in
fact, the trailing CRLF is irrelevant insofar as boundary checking goes
because implementations have to assume that trailing blanks might be
added). The emphasis on the trailing CRLF that's implied by the BNF is
therefore a red herring, and I don't like formal red herrings.
However, there is no way to put the CRLF back into the innermost
portion of the BNF without reinsering the requirement of an initial
CRLF. And I certainly agree that losing the initial CRLF is a good thing
that's worth a little awkwardness.
Ned