Well, finally, the majordomo BOF summary. John has added a lot of
corrections and quite a bit of information which I just shamelessly
folded in because it was looking too weird to credit him where necessary.
------
A Majordomo Birds-of-a-Feather (BOF) session was held at the USENIX
Large Installation Systems Administration Conference (LISA VIII) on
September 21, 1994. There was good attendance and it was nice to meet
a lot of the people who are on this list (although I was becoming
rather sleep deprived by that point on Wednesday). The BOF had a tight
hour to fit into, as people were scheduled before and after in the same
room.
The BOF attempted to cover some of the concerns the development and
user community had about its future. At last year's LISA the plan was
to produce majordomo 2.0, a radical revision which would take the list
management software from its initial intent (to manage Brent Chapman's
17 SAGE mailing lists) to something which would support a more general
need. The development at that point had changed from Brent to John
Rouillard as lunatic in charge.
The two major revisions points of 2.0 were to restructure the list
config, which was presently resulting in a proliferation of little
files, to a single command script for each list. Secondly, the command
parser needed to be rewritten. Before 2.0 could really get underway,
the old 1.62 version began to mutate. Currently we are at version 1.92,
and John's time is more limited so it will be a while before 1.93 comes
out. The goals behind 2.0 are still valid, they have just been waylaid
by bugs and mutations.
In the realm of progress towards 2.0, Alan Stebbens has rewritten the
command parser which is one chunk of the job. Some work is needed to
fold it in, but it is a good step forward. One note on majordomo
development: if you have an idea, you will want to join majordomo-workers
(at greatcircle.com) and discuss it with folks there. That way the
incompatible strains of majordomos may be kept to a minimum.
One of the concerns as majordomo becomes more popular and subject to
childish games is how to prevent people from creating mail loops by
subscribing a redirector to the list. Alan suggested that incorporating
something like mailagent to do message ID checks would be a good place
to start. Mailagent will probably not become viable until 2.0.
One area that needs work is the resend program. Among the things that
need to be added are:
recording of statistics such as mail message size,
author, date sent, time spent in processing
recording of message ID's and bouncing messages containing
the message ID in a header like field to the
list owner.
If anybody would like to tackle this for 1.93 release, John would
appreciate it.
Also in 1.93 will be a complete rewrite of access list code. The new
permissions stuff will be a table of patterns (regexp and glob style)
versus actions:
/.*cs.umb.edu$/ allow # action allowed
*mit.edu deny # action denied
*.bc.edu password=fooble # action allowed
# if password is fooble
* file=list # action allowed if name
# found in file list
* file=list-owners # ditto except file
# is list-owners
* deny # if it wasn't found deny it
This will affect all options that currently have private options, as
well as advertise, noadvertise, subscribe_policy/unsubscribe_policy.
Performance is a big concern. [An interesting side note here is that
Brent mentioned that because greatcircle.com has been a 3/60 perhaps
a slow machine is a nice flame killer. Since the messages spread
more slowly, there isn't that instant gratification feedback loop that
causes people to go ballistic quite so quickly.] If you are interested
in these issues, I'd also urge you to listen in on majordomo-workers.
Also in 1.93 the fileystem layout will change. There will be standard
Unix like layout structure (bin, lib, etc....) for the elements under
the majordomo root. In addition the layour for mailing list
directories may change. There will be one subdirectory for each type
of file, mailing lists, per list config files, info files etc. This
makes it much easier to create a precompiled global config file to
improve performance, since it looks like reading in hundreds of config
files is excessively expensive.
What was mentioned in the meeting is folks are aware there are problems
and code modifications need to be made.
There is a need for better documentation and an installation time
permission checker (for all the files, both POSIX and non).
John's little wrapper checker will be included in 1.93.
Other miscellaneous notes:
A big win would be to get a taintperl version of the wrapper script.
At the time of the original coding Perl was broken so the first attempt
was aborted. Taintperl works now of course, so it should probably be
done.
There are plans to incorporate a real RFC822 header parser, thus handling
a number of bugs with comment parsing etc.
A consideration was put forth to consider the use of syslog to do
the logging.
There will be no more Error Code X messages, sysexits.h will be used
which will give real messages. Sendmail uses sysexits.h to determine
what happened to the child process.
Brent is looking at changing the copyright status of majordomo. The
copyleft as well as the Berkeley Copyright are being looked at.
The hostile address message was never meant to catch x.400 addresses...
freudian programming if you ask me. :-) The hostile address checks will
be come smarter thanks to some experimentation by Anne Bennett.
-----
And, that's all folks! Thanks to John for providing some needed
editing and correction.
jennifer
--
Jennifer Joy sys/net admin Motorola/RISC HW Austin,TX
jjoy@risc.sps.mot.com 512.891.8561 fax:512.891.3190 pgr:933-7333 #898561
Follow-Ups:
|
|
