Nigel Metheringham edited this page Nov 25, 2012 · 1 revision
Clone this wiki locally

Exim 5 Discussion

Please add proposed features for Exim 5 to this page. If the description (or discussion) is substantial then make the item a subpage - ie Exim5/MyFeature and link to it from here. Links to mailing list discussion are also appropriate.

Major Architectural Changes

  • Storage revamp - use a storage API allowing different storage systems to be attached
    • SQL Email Storage and expanded SQL support
  • Configuration revamp - this could cover all sorts of things...
  • Simplification of the string expansion language
  • Pluggable lookup API
  • Pluggable content scanner API
    • RuthIvimey asks: forgive me if I've missed an existing ability, but I'd like to be able to do ACL-time content scanning using several different programs on the same message.
  • Pluggable router API
  • Pluggable transport API
  • Reduce the number of different concepts
    • Eliminate the system filter
    • Have local_scan modules explicitly called from ACL
    • Make routers more like ACLs in that conditions are checked in the order they appear and can appear multiple times
  • Improve header handling
    • In what way? To fix what problems?
    • MIME-encoded headers are confusing
  • The aim for Exim #5 ought to be to modularise significantly, so that several different designs are possible within the same overall architecture. That's already the way things have gone with spam filtering schemes.
  • RichardClayton is surprised that no-one has mentioned queues yet. Quite clearly a key design aim for #5 should be to revisit the assumption that the majority of email can be quickly shuffled off to somewhere else. There is now a page about this here.
  • Other major subsystems should become modules so that multiple parallel implementations can replace the existing design. Maildrop style is one (where the code for several schemes is already present), what else? The big plus is that Exim is mature enough to see what the grand overall architecture should be : building from scratch something with a selectable modules style of design can lead to significant inefficiency as one goes through "we'll give them a choice here" interfaces to no practical purpose.
  • The code for callouts is entirely separate from the smtp transport. This causes a number of problems, including the lack of availability of AUTH and LMTP for callouts. Bugzilla #321 and #345 mention this. A better design would be to redesign the transport so that it can function either as a message deliverer or as a callout mechanism.
  • Exim's handling of resent-headers should be revisited. I think there are now more guidelines about as to how they can be grouped and how multiple sets of resent-headers should be handled. fanf suggests sections 6-8 of http://www-uxsup.csx.cam.ac.uk/~fanf2/hermes/doc/qsmtp/draft-fanf-smtp-rcpthdr.html
  • with an python API, exim could be more powerfull and flexible, possibly we can use somethink like this http://freshmeat.net/projects/exim-python/

Other Aspects

  • Configuration file improvements
  • The ability to pipe mail into some process (exists today), and use the output as the new mail (does not exist yet). The current way of having to call /usr/sbin/sendmail again to feed a modified mail back into the mail system after, say, getting it through spam- or virusscanning is a kludge, at best.
    • RuthIvimey: seconded... and to be able to do so from an (appropriate) ACL as well as a router?
  • Possible use of "autoconf"
  • Make it possible to link Eximon with something other than the Athena widgets -- or indeed, rewrite the whole thing.
  • In the same vein, add a GUI configuration tool that can create at least modestly complex configurations. Perhaps integrated into the reworked eximon? I always liked the interface of "mailtraq", although that was a while ago |:)|
  • Philip Hazel's old ''future ideas'' list
  • Ideas for string expansion changes
  • XML file lookup, which would allow to process and fetch data from a XML file. I don't see how XML hierarchy would make benefit for us, but surely we can search and extract data from tags of a given type.