home *** CD-ROM | disk | FTP | other *** search
- Comments: Gated by NETNEWS@AUVM.AMERICAN.EDU
- Path: sparky!uunet!paladin.american.edu!auvm!INNOSOFT.COM!NED
- Errors-to: epmdf@YMIR.BITNET
- X-Envelope-to: PMDF-L@IRLEARN.BITNET
- X-VMS-To: IN%"DPMSYS%RITVAX.BITNET@ymir.claremont.edu"
- X-VMS-Cc: IPMDF
- MIME-version: 1.0
- Content-type: TEXT/PLAIN; CHARSET=US-ASCII
- Content-transfer-encoding: 7BIT
- Message-ID: <01GOFZ0LZ4J696VN7I@YMIR.CLAREMONT.EDU>
- Date: Sun, 6 Sep 92 10:50:29 GMT
- Sender: PMDF Distribution List <PMDF-L@IRLEARN>
- From: Ned Freed <NED@INNOSOFT.COM>
- Subject: RE: Using DELIVER
- Newsgroups: bit.listserv.pmdf-l
- Lines: 112
-
- > I know this has been covered on this list before, but can
- > someone explain again why the VMS forwarding address cannot
- > be used as the determinant for the execution of DELIVER.
-
- PMDF has to be the entity that determines whether or not integrated DELIVER is
- used. This is fundamental -- integrated DELIVER depends on the knowledge that
- it is processing a PMDF-originated message. If the mail first passed through
- VMS MAIL all the advantages of integrated DELIVER would be totally lost.
-
- This then implies that PMDF would have to react to the forwarding addrtess
- before VMS MAIL gets the message. But only VMS MAIL knows about its forwarding
- addresses. PMDF doesn't know what these addresses are or how to interpret them.
-
- It is simply not practical to try and teach PMDF about VMS MAIL forwarding. VMS
- MAIL forwarding is very complex, very non-intuitive, and any attempt to unravel
- it would result in all sorts of hideous bugs until we got all the semantics
- issues worked out. It would also be a big layering violation -- we have no
- business dealing with data that's private to VMS MAIL. (Note that dealing with
- incoming addresses from the foreign interface is a totally different matter
- that has no real relationship to general forwarding address handling.)
-
- The task of just identifying a PMDF forwarding address that points at DELIVER
- is arguably simpler, but not enough to matter. Don't underestimate the
- complexity here -- we're talking about something that hard, that has numerous
- serious pitfalls, and probably would take many iterations to get right.
-
- Finally, one of the key uses of integrated DELIVER is to conditionally handle
- certain mail and leave the rest available for "normal processing". Part of that
- "normal processing" may involve dynamic use of forwarding addresses. But this
- isn't possible if you've used up your forwarding facilities on firing up
- DELIVER. This is another facet of the layering violation that's implicit in
- this suggestion -- it would put the cart before the horse insofar as most uses
- go.
-
- It is therefore impractical, perilous, and totally inappropriate for PMDF to
- use forwarding address values to determine _anything_.
-
- > I recently encountered the following senario:
-
- > A user runs PMDF with integrated DELIVER on his home
- > machine (TECHNO). On another machine (VAXA) which also runs
- > PMDF with DELIVER disabled he has his forwarding address set
- > to forward mail to his home machine. His login directory on
- > VAXA is his home directory on TECHNO mounted on VAXA via
- > DEC's DFS.
-
- PMDF and DELVIER aside, this is an extremely dangerous setup. DFS interacts
- very poorly with VMS MAIL because of VMS MAIL's use of shared indexed files.
- DFS cannot support this sort of shared access, and VMS MAIL is not prepared to
- deal with the errors that result from collisions across DFS.
-
- Based on very nasty prior experience I would never put this strategy into
- production use. I personally have seen this cute little trick cause huge
- problems, up to and including the silent loss of lots of mail. "Minor" problems
- are usually things like finding your entire cluster wedged tight on Monday
- morning.
-
- Incidentally, using NFS doesn't offer a viable alternative. NFS would get
- rid of most of the wedging problems at the expense of risking total
- corruption of any ISAM database you happen to have (like VMS MAIL folders).
-
- > This all worked fine until I enabled DELIVER on VAXA. PMDF
- > ignored the forwarding to TECHNO and instead processed all
- > mail to his account on VAXA using his mail.delivery file
- > from TECHNO.
-
- What you have here is basically an impossible situation to resolve. You have
- built an environment where users and the software they use has to be able to
- tell what system is in use. But you have provided no reasonable mechanism (like
- different home directories) for them to use to distinguish one environment from
- the other. You therefore force sensitivity to environmental issues into every
- single application that anyone writes or uses on your system.
-
- > I realize there are workarounds such as defining different
- > mail.delivery file names for different nodes, but what are
- > the reasons for allowing the file to override a forwarding
- > address.
-
- The fundamental reason is simple -- DELIVER programming logically occurs before
- delivery to the user agent is done. Forwarding in the user agent (which is
- _very_ different from forwarding at the MTA level) is therefore a step that's
- taken after DELIVER programming and not before.
-
- We thought all this through very carefully when integrated DELIVER was
- originally added to PMDF. Various orderings were considered including the one
- you describe.
-
- Defining different MAIL.DELIVERY files would provide a means for users to
- distinguish between different environments. But users still have to code around
- the differences. We could provide access in the MAIL.DELIVERY file to node
- information to make this sort of thing easier. It would make it possible to
- program things in a node-specific way. But the underlying problem remains the
- same no matter what.
-
- Another theoretical approach would be to completely eliminate the need to
- distinguish between the environments. This would in turn eliminate the need to
- forward mail from one place to the other -- why bother if the environments are
- identical? However, this simply cannot be done using DFS, NFS, or any other
- tool I know of short of actual VMS clustering.
-
- > If there are no technical limitations to letting
- > the ~forwarding address be the deciding switch, can the switch
- > be defined in the PMDF option file.
-
- There are good technical reasons as well as excellent strategic reasons for not
- doing this. This is therefore extremely unlikely to ever get implemented in
- PMDF.
-
- However, in my opinion DELIVER usage of this sort is the very least of your
- problems -- you simply have a big accident waiting to happen here.
-
- Ned
-