BM> Yes, but what Dan Bernstein is doing is far worse IMHO.

Your humble opinion is wrong. Checking for and rejecting
protocol errors is in general rarely worse than heuristically
trying to convert erroneous data into non-erroneous data,
and isn't in this particular case. "Auto-correcting" heuristics
have the propensity of introducing many more new problems, and
also serve to increase the number of faulty implementations by
removing the negative feedback mechanisms that would otherwise
weed them out.

BM> From RFC1123 (which I am sure he is quite familiar with):
BM> 1.2.2 Robustness Principle
BM> [...]

Gresham's Law trumps Postel's Principle, and the explicit
requirements in the RFCs prohibit an SMTP Relay server from
behaving as you imply it should.

<URL:http://homepages.tesco.net./~J.deBoynePollard/FGA/qmail-myths-dispelled.html#MythAboutBareLFs>

BM> Writing buggy software can to some extent be excused, but writing
BM> and using software that deliberately causes interoperability
BM> problems, like qmail, is just plain stupid.

The cause of the interoperability problems is the clients that don't
speak the protocol correctly, not the server that requires that the
protocol be spoken correctly.