* Elevate "vacation" to a real release program -- check man pages,
install in /usr/ucb, etc. May need some work -- it's still a bit
* Throw in "smart UUCP" mailer that knows about neighbors that can
handle more than one recipient at once. This is easy and should
be very useful for certain sites. Configuration files should
default to "dumb" however.
* Clean up configuration on releases -- send out fewer Berkeley and
more basic configs (e.g., BLI).
* Verify that new name server doesn't confuse sendmail.
* Only add Message-Id: if message is locally generated (???).
* Improve documentation describing how to create a config file --
probably means putting more configs on distrib.
* Listen in UNIX domain (as well as INET)?
* A way to mark addresses in the queue as you process them as "done" --
probably by appending lines to the cf* file. The point here is to
avoid duplicate messages to large lists if a crash occurs during
processing. This is hard.
* Break up sendmail into several smaller programs:
queueup -- insert message into mail queue; call runqueue when
done -- called by user, smtpserver, cleanqueue (to return
runqueue -- process queue entries -- take optional job name list
to run only those jobs -- called by queueup and by cron
printqueue -- print mail queue.
cleanqueue -- return expired messages; basic consistency
smtpserver -- process smtp connections, pass to queueup -- this
can probably become a simple service a la /etc/inetd.
addrtest -- essentially "-bt" mode.
verifyaddr -- verify an address list -- "-bv" mode -- needs a
"noisy" and a "silent" mode (exit status only). This
should be the same code as smtpserver and queueup use
[This may want "shallow" and "deep" modes, where
deep mode tries to connect to the remote site if
on SMTP and issue a "VRFY" command.]
newaliases -- build alias file -- should not be part of sendmail
proper. (See editalias below -- that would probably
preempt -- although a text form will be critical for
It may be better to make runqueue a daemon on 4.2 and have queueup
poke it (using a datagram). This guarantees single threading (is this
an advantage? -- long jobs can block the queue) and minimizes
gratuituous forks/execs. However, I feel strongly that portability
is important -- the fork/exec method should still be included.
editalias -- front end to modify alias database, including
"ownership" (so that people can modify lists that they
own, but not other lists) -- see Grapevine papers --
rmqueue -- remove a job from the queue. Usual ownership rules
* Change config file to be more understandable -- yacc grammar probably.
Perhaps defaults should be compiled in whenever possible.
* Only use rewriting rules to go to and from canonical format. Other
resolution should be defined by the name server (See MMDF).
* Perhaps there should be a "compiled format" for the configuration
file, something between a yacc-style config and the freeze file; very
quick to read w/ low overhead. [Suggest a struct with "char *" being
index into a buffer area; these can be mapped internal <==> external
[] Currently sendmail will rewrite "To: miriam mckusick" to
"To: miriam, mckusick" unless some "special" characters
are used (e.g., ()<>,; -- characters specific to RFC822).
This should be moved to the user agent (Mail, rmail, MH,
etc.). Big job because all user agents must be modified.
[] Would be nice if UUCP shit-format was handled by UUCP
itself. This has serious performance implications unless
actually built into UUCP.
* Non-sendmail related changes:
[] UUCP routing code (part of the name server really) -- can
the work that Peter Honeyman or others have done be useful?
[A significant point is making this work given incomplete
information -- yum yum. Some database work in this area.]
[] Reply code in ucbMail has little relation to current mail
software. "Smarts" should be stripped out.
projects / unix-history / blob