From: <Michael_Harer@motorcity2.lotus.com>
To: drums@cs.utk.edu
Message-Id: <8525636C:0079F431.00@motorcity2.lotus.com>
Date: Fri, 19 Jul 1996 18:42:54 -0400
Subject: handling sendmail nonconformance
Mime-Version: 1.0
Content-Type: text/plain


  I recently came across an existing V8 sendmail practice that caught me by
  surprise and which I consider a good issue to discuss/clarify in a
  subsequent Internet-Draft of the revised SMTP document in hopes of
  improving interoperability.  Here's the IETF email trail on the subject:

  >>>>  Today I isolated an SMTP interoperability problem which was induced
  by
  >>>>   the fact that  V8.6.12 sendmail terminates it's SMTP commands with
  >>>>  linefeed instead of <CR><LF>.  Perhaps this is configurable via
  >>>>   sendmail.cf but doesn't the standard clearly state that SMTP
  commands are
  >>>>   to be terminated by <CR><LF> (mine does)?
  >>>>   I'm hoping that someone will answer the following questions so
  that I can
  >>>>   choose the best strategy to resolve this delema.
  >>>>   1)  Are there many versions of sendmail (or other MTA
  implementations)
  >>>>   that terminate 821 command lines with just a linefeed char?
  >>>>   2) Is this a known sendmail glitch (bug) that happened fairly
  recently
  >>>>   that is fixed ??  OR  Should MTA implementations be modified to
  accept it
  >>>>   since there are many versions of sendmail out there doing this
  already?

  >>> I say NO.  Patching conforming implementations to adopt to broken
  ones
  >>> propogates garbage.  If MTAs refuse to communicate with hosts running
  >>> software that sends LF instead of CRLF, those hosts will get fixed.
  >> Easy to say, but if you're selling SMTP mail system(s) and your
  customers
  >> call you complaining that they can't get mail to john doe, or when
  they're
  >>evaluating the product and can't get mail to their most important
  customer
  >> you're hard pressed to point fingers. Especially when the fix is
  relatively
  >> simple (if you encounter a LF character before you hit a CR when
  reading
  >> the HELO command, then set your incoming line-terminator character(s)
  to
  >> a LF instead of CRLF).
  >> If you correct it coming in, then you're not propagating anything,
  just
  >> following that oldest of 'net rules; be conservative in what you put
  out
  >> and liberal in what you accept...
  >I agree with all of this, but please keep in mind that there is a big
  >difference between an implementation deciding to accept this junk and
  modifying
  >the standards to require implementations to accept it.
  >LF-terminated lines in SMTP are a fact of life every server implementor
  runs
  >into sooner later. Depending on context they will either get the
  offending
  >client fixed or else modify their server to accept this incompliant
  form. Or
  >provide a switch to let the customer choose whether or not strict
  compliance is
  >enforced in this area. (This last is what our PMDF products do.)
  >However, a note about the prevalance of this practice on the Internet,
  >how nonstandard it is, and in spite of this the reality of having to
  deal
  >with it, in the SMTP specification might be a good idea.

  So you can see the differing opinions on the subject.  I chose to add
  support for SMTP commands that end with just a linefeed (in addition of
  course to the standard representation) because of a current customer
  interoperability problem but also because the IETF list feedback
  indicates the practice is prevalant.  I suspect that we would eventually
  be inundated with customer interoperability problems if I don't support
  this sendmail behavior.

   So as one of the responses to my IETF list posting suggested (about
  doc'ing this) and which I strongly agree,  I would like to have this
  topic discussed/clarified in a revised SMTP doc if it isn't a;ready.
  Please let me know  how I should proceed to get this included (i.e. is
  this info enough? or should I provide more specifics, proposed
  refinements...etc). thanks

  please advise,
  Mike