Date: Mon, 8 Jan 1996 08:32:19 -0800 (PST)
From: Mark Crispin <MRC@CAC.Washington.EDU>
Sender: Mark Crispin <mrc@Tomobiki-Cho.CAC.Washington.EDU>
Subject: Re: "be liberal in what you accept" considered harmful
To: John C Klensin <klensin@MAIL1.RESTON.MCI.NET>
Cc: RANDY@MPA15AB.MV.UNISYS.COM, drums@cs.utk.edu
In-Reply-To: <2.2.16.19960108142827.41ffbca8@mail1.reston.mci.net>
Message-Id: <MailManager.821118739.27925.mrc@Tomobiki-Cho.CAC.Washington.EDU>
Mime-Version: 1.0
Content-Type: TEXT/PLAIN; CHARSET=US-ASCII

On Mon, 08 Jan 1996 09:28:27 -0500, John C Klensin wrote:
> I see a transition problem getting back to where we should have been, but
> basically agree with John -- the history of senders claiming that receivers
> are broken because they aren't liberal enough to accept the garbage the
> senders deal out has really been a big contributor the the mess we have
> today.

Two anecdotes about this:

For many years in the 1980s, I had people from the Unix world contending that
my SMTP server on the DEC-20 was "broken" because I did not treat bare LF as
newline.  Eventually found myself obliged to put in a diagnostic that detected
bare LFs and send back an appropriate snarl message to postmaster at the
offending site.  Said offenders were almost always using sendmail, and either
denied that they sent bare LFs instead of CRLF, or claimed that bare LF is the
"Unix standard"...  Thus began a prejudice against sendmail which I have to
this day.

The first episode of this sort of thing happening in IMAP has just occurred.
Old versions of the UW IMAP server used strtok() to parse the criteria in a
SEARCH command.  It turns out that a commercial client depended upon this
behavior and generated SEARCH commands with a trailing space, which in turn
got rejected by just about any other IMAP server (including newer versions of
the UW server).  I got a request from the vendor to list all the "incompatible
changes" to the UW server such as "no longer permit trailing spaces at the end
of SEARCH criteria"...as if I kept ever kept track of protocol violations that
old versions of my server may have passed without checking...

I find myself being a definite fan of "be conservative in what you send", but
I am not so sure I believe in "be liberal in what you accept".  I'm starting
to think that the only way out of this morass is to start requiring that
servers abide strictly by the protocol in what they accept.