Wietse: > Postfix writes the Received: header at message arrival time. The > decision to add a Received: state field must be made before the > "250 OK" in response to end-of-data, after the message is frozen (*). > Once Postfix sends "250 OK" in response to end-of-data, there is no way > that a Received: header can be replaced without violating the RFC > requirement that mail not be lost due to frivolous causes.
Murray S. Kucherawy: > A quarantine instruction from milter is returned by the EOM callback > before the DATA 250 is sent (has to, or SMFIS_REJECT couldn't be > implemented, for example). I think that's the one case that seemed > most obvious to me, and it also seems to fall within these design > parameters. Given that "quarantine" is one of the pre-defined states, that seems an obvious use case. > I don't think you have to worry about the other states the draft > talks about, as they're mostly to do with things that are implemented > outside of an MTA (MLM hold-for-moderation, for example). If this Received: state is recorded by the MLM after delivery from MTA to MLM, then indeed I don't care, since this happens outside of the MTA. But that is not what the example suggests: it has all the fingerprints of Sendmail (software / configuration) versions 8.6.11. > But design issues aside, would you consider implementing it at > some point? Indications of interest like that would be useful > input to the IETF. I suppose it is doable, but what is the killer application that would encourage me to do this? Wietse