Sieve forwarding loop destroys e-mail

[Joseph Brennan]

Matt Garretson <mattg at assembly.state.ny.us> wrote:

> Along similar lines, any well-written Procmail recipe which redirects
> mail typically checks for, or adds, an "X-Loop" header before
> forwarding anything.

Yes, it's an old solution.

The crucial difference is that if one writes a bad procmail recipe,
the message loops round and round until one of the MTAs considers
the hop count exceeded and bounces it to sender, but if one writes
a bad sieve rule, the message _is silently lost_.  That's a much
harsher penalty.

And we stand a chance of here of doing _better_ than procmail.  If
we insert a header roughly like 'X-Sieve-Seen: user hostname' when
we forward, we can look for it in incoming messages and say we
won't forward again.  So the penalty for writing a bad sieve forward
rule would be that it doesn't forward.  That's better than bounce
to sender, and way better than losing the message.


> If the "editheader" Sieve extension gets implemented, then a well-
> written sieve script should be able to do the same type of thing.
> To me this seems a bit more sane than expecting lmtp or sieve to
> accomplish it automatically.

I've been called crazy before!



Joseph Brennan
Lead Email Systems Engineer
Columbia University Information Technology