[Top] [All Lists]

Re: MIME boundary question

1995-02-15 09:35:52
< From: John Gardiner Myers <jgm+(_at_)CMU(_dot_)EDU>
< hansen(_at_)pegasus(_dot_)att(_dot_)com writes:
<< No more or less than we expect people to write correct parsers for SMTP.
<< So the answer is "Yes".
< SMTP doesn't require large or arbitrary amounts of lookahead in order to
< parse correctly.

Personally, I'd be happy with limiting the amount of lookahead to at most 1K
(or whatever amount we decide on). It makes it easy to parse and it doesn't
require large amounts of lookahead. This seems to be the best compromise
between all the choices.

This allows all boundary strings to be treated as unique, no matter if they
are nested within other multipart sections or not, while also handling
messages which have passed through gateways which add or subtract trailing
blanks from lines.

< As I've mentioned before, allowing boundaries to be substrings of
< boundaries of enclosing multiparts breaks a fair amount of deployed code
< that was correct per 1341.  The 1521 multipart grammar has a number of
< problems, which I've pointed out previously.

When 1521 came out, there were a variety of changes that many deployed
programs had to update to; this one's no different.

If we go back to the 1341 standard of non-unique boundaries, those programs
that have already moved forward to the 1521 rules (or some interpretation of
those rules) will be broken again.

By all means, let's fix any problems that exist with the 1521 text and
grammar, but let's not go backward to a poorer standard while doing so.

                                        Tony Hansen
                                att!pegasus!hansen, attmail!tony

<Prev in Thread] Current Thread [Next in Thread>