Jacob Palme <jpalme(_at_)dsv(_dot_)su(_dot_)se> writes:
MIME specifies that the boundary string between body parts must be a
string which does not occur in the body part itself. This goal can be
achieved in two ways:
(a) Try out different boundary string, until a string can
be found which does not occur.
Gnus uses this method, FWIW. It also uses sane boundary strings
(combinations of - and =) that actually look like boundaries to people who
see the raw message, rather than the alphanumeric mess that most clients
use and that confuses the eye when looking at the raw message.
(I've long been of the opinion that MIME wouldn't have encountered as much
resistance as it did if the early implementors had used sane boundary
markers and not reiterated the default headers at the beginning of each
part; in other words, if they'd given some thought to how the message
looked to non-MIME-compliant readers.)
(b) Select a sufficiently long random boundary string,
that the probability that it occurs in the body
part is negligible. This method is not really
permitted according to RFC 2046.
I have an impression that many mail programs use method (b). And that
some use method (b) but do not even select a really random boundary
string.
I have the same impression.
Most versions of Eudora also have an interesting bug wherein when forming
nested multiparts, some percentage of the time the exterior boundary will
be a prefix of the interior boundary. I think they've finally fixed this
in the latest release.
--
Russ Allbery (rra(_at_)stanford(_dot_)edu)
<http://www.eyrie.org/~eagle/>