rpatterson.mailsync Integration between mswatch, OfflineIMAP, and Gnus for realtime mail =================== rpatterson.mailsync =================== Integration between mswatch, OfflineIMAP, and Gnus for realtime mail .. contents:: This package provides some scripts that wrap parts of mswatch and OfflineIMAP and integrate with Gnus to provide a local maildir that is synchronized with a remote maildir as changes occur, instead of polling on a regular basis. This provides for near instant delivery of new mail while also reducing resource utilization. Integration with the Emacs mail and newsreader, Gnus, is also provided in such a way that your single threaded Emacs process is blocked much less as changes occur to the maildirs. Requirements ============ * `mswatch <http://mswatch.sourceforge.net>`_ * `OfflineIMAP <http://software.complete.org/software/projects/show/offlineimap>`_ * `Gnus <http://gnus.org>`_ Actually, they're all kinda optional. The OfflineIMAP sync wrapper script can be used with mswatch without Gnus. The Gnus checkers can be used without OfflineIMAP. For that matter, the OfflineIMAP sync wrapper script can be used without mswatch, but why would you? :) Installation ============ All you need is easy_install_:: $ easy_install rpatterson.mailsync The mailsync-gnus.el library will be installed in the site-lisp directory in the egg. To use the library, you'll need to add this path to your emacs load-path. It should be something like the following but be sure to substitute the question marks with the appropriate values for your version of Python, the version of rpatterson.mailsync, and your easy_install site-dirs:: /usr/lib/python?.?/site-packages/rpatterson.mailsync-?-py?.?.egg/site-lisp/ Alternatively, if you have a /usr/local/emacs/site-lisp directory, the mailsync-gnus.el library can be installed into that directory if you install rpatterson.mailsync from a source distribution. You can still use easy_install to get the source distribution:: $ easy_install --editable --build-directory=/usr/local/src rpatterson.mailsync $ cd /usr/local/src/rpatterson.mailsync/ $ python setup.py install Once the library is on Emacs' load-path, to use the mailsync/gnus-check function, you'll need to make sure it's loaded in your Emacs. You can do this by adding the following to your .gnus.el:: (load-library "mailsync-gnus") If you'll be using mailsync_offlineimap to synchronize your mail, start by configuring OfflineIMAP as per the OfflineIMAP documentation. Once OfflineIMAP is verified working, verify mailsync_offlineimap by running it from the console. On any machine where you'll be using the maildir watchers, either with or without mswatch, see the mswatch documentation and verify that watch_maildirs works. Install rpatterson.mailsync wherever you want to use mailsync_gnus_watch. Run it from the console to test it. It should check all groups in Gnus and output a blank line when it's started up and then it should check any changed maildir folder groups in Gnus when changed and then omit the folder name. Use "mailsync_gnus_watch --help" to see what options are available to modifying the watcher behavior. To use mswatch, see the mswatch documentation but base your ~/.mswatchrc on the following and see the "MAILSYNC:" comments for what to change:: # minimum time after first queued mailbox change to synchronization (default: 10s) #base_delay 10 # minimum time between two syncs or failed attempts (default: 60s) #inter_delay 60 # minimum time between two syncs or failed attempts for specific lists #inter_delay 30 important_list #inter_delay 600 high_volume_list another_list # maximum waiting time between failed attempts (default: 600s) #max_delay 600 # program (and arguments) to run to sync the mail stores (required) # sync mswatch-sync # MAILSYNC: use the following to have mswatch use your OfflineIMAP # setup to sync folders on change. sync mailsync_offlineimap # prefix this string to sync's mailboxes; # useful as mbsync channel or OfflineIMAP account. # the first string ("mydomain") is always prefixed # the second string (":") is prefixed only when syncing a particular mailbox mailbox_prefix foo : # a store to watch, call it "local" (required) store local { # program (and args) that will monitor this store for changes (required) # see 'man watch_maildirs' to tell watch_maildirs where to find mail # watch watch_maildirs # MAILSYNC: use the following to have your local Gnus check # folders as they change, otherwise, just use the above. watch mailsync_gnus_watch } # the other store to watch, call it "mydomain" (required) store foo.com { # program (and args) that will monitor this store for changes (required). # # Uses ssh private/public keys to login without password prompting. # Uses ssh BatchMode so that 'mswatch' promptly detects disconnects. # Uses 'inputkill' to run 'watch_maildirs' so that 'watch_maildirs' # promptly exits if the connection dies. # watch ssh -o BatchMode=yes foo.com inputkill watch_maildirs # MAILSYNC: use the following to have your remote Gnus check # folders as they change, otherwise, just use the above. watch ssh -o BatchMode=yes foo.com inputkill mailsync_gnus_watch } .. _easy_install: http://peak.telecommunity.com/DevCenter/EasyInstall#installing-easy-install TODO ==== * Use optparse for syncers Add built-in help and clearer option and argument handling for the syncers using optparse. * Use delay in watchers/checkers Currently, the checkers run every time the watcher emits a folder name. Most maildir changes emit the same folder multiple times resulting in the checker running reduntantly. With mswatch, the logic for coalescing events using a delay is in the mswatch process. To address this problem, that logic would have to be extended to the watchers. This is another reason among many for reimplementing mswatch. * Invoke OfflineIMAP in the same process Currently, mailsync invokes OfflineIMAP as a subprocess which is a little wasteful given how heavy launching a python app can be. I briefly looked at the OfflineIMAP code to see if it could be easily done but was disappointed. It would be great if any OfflineIMAP people want to show me how I can do this. * Long running syncers Currently, mswatch invokes the sync process anew every time. Pretty much any sync process establishes one or more network connections. There's no need to do this every time. Better to have a long running sync process that mswatch can feed the folders to sync as they require syncing. Implementing this would require modifying or reimplementing mswatch itself. It would also require support in the sync process. Perhaps OfflineIMAP could be used as a library to accomplish this. * Ignore duplicate maildir notifications As described_ on the mswatch page, every sync is redundantly issued twice. To address this would require modifying or reimplementing mswatch itself. .. _described: http://mswatch.sourceforge.net/#limitations * maildrop and gnus splitting? Maybe have maildrop directly invoke Gnus splitting and feed incomming messages to be split at time of delivery. This would eliminate redundant syncs on the INBOX where gnus would otherwise just immediately move the message into another folder. This approach is, however, probably way to heavy for a maildrop pipeline. An alternative approach would be to use a maildir (~/Incomming) separate from the real INBOX maildir (~/Maildir) where mails would first be delivered. This incomming maildir could be monitored using a checker/watcher whose only real purpose is to get gnus to check the incomming maildir and split messages. The split would then move the message to the destination or fallback to the real INBOX maildir which would in turn trigger any real watchers/checkers used by mswatch. This approach requires no additional mailsync support. It also keeps maildrop job small and light, decoupling the gnus splitting work from delivery. An advantage of either approach is that mail would be split on the server even when the local mswatch wasn't connected. As it is now, mail is only split when mswatch is running or if the Gnus running on the server checks all mail periodically. * Use Gamin Use Gamin to provide compatibility across the file modification services available. Changelog ========= 0.4 - 2009-06-28 ---------------- * Propogate non-zero return codes from subprocesses so that mswatch can know to restart things. * Don't check if the sync fails, allows mswatch to keep trying to sync when offline without constantly making the checkers run * Add another syncer for offlineimap with a local-only gnus checker * Translate Gmail and Gnus IMAP folder separators 0.3 - 2008-05-13 ---------------- * Clarify and correct documentation 0.2 - 2008-05-07 ---------------- * Fix handling of site-lisp for easy_install 0.1 - 2008-05-01 ---------------- * Initial release 0.0 - Backstory --------------- It is amazing that `MUAs do indeed still suck`_. In my case this is exacerbated by a strong preference to use Gnus as my MUA. Thunderbird may be great, but since it lacks good integration with Emacs (and no, emacsclient external editor is not enoug), it really doesn't matter. I began by using nnimap to access dovecot on my server. I was still a Gnus newbie at that time and I ran into many problems with nnimap. I tried using the Gnus agent to address these but then my head and my Emacs exploded in unison. As many others have before, I moved on to sync the ~/Maildir on my server with a full local copy using OfflineIMAP. Then I setup gnus on my laptop to use nnimap talking to Dovecot running locally on my laptop. This worked fabulously with one exception, I had to choose between how quickly new mail would show up and how much bandwidth OfflineIMAP would use. I'm pretty impatient about my email, so with a sync frequency of every minute, I found that OfflineIMAP could easily saturate what little bandwidth was available at a cafe or somesuch. Lengthening the frequency only improved things slightly since every 5 minutes or what have you, my web browsing would still slow to a crawl. I assume I was also slowing down everyone else, please don't tell them. Still I used this approach for a year or more since I could think of nothing better. Time went on and I used more and more of Gnus' features. I have a beautiful BBDB fancy splitting setup. I use automatic scoring which is just a delight. I agree that Gnus is much too hard to learn and configure, but in all honesty, it could be even harder and I'd still use Gnus. Later, with much more Gnus experience under my belt, I blew away my .emacs and .gnus.el determined to use nnimap and the agent now that I was more proficient. I set it all up and it all seemed to work but as time went on I kept running into more and more problems. Over and over again, the Group buffer would show inaccurate counts of new messages or not show new messages at all. I'd find that articles hadn't been downloaded after all so I wasn't able to read them offline. I came to resemble a Skinarian pigeon, genuflecting before Gnus in the face of chaos with mysterious combinations of "C-u g" and "M-x gnus-agent-regenerate". In addition to those and many other small annoyances, there was still the bandwidth vs check frequency issue, only this time every time the mail would be checked my emacs would block as well. Then I learned about IMAP IDLE and was beside myself with excitement. Then I learned that neither OfflineIMAP nor Gnus nnimap supported it and that Thunderbird did and I found myself seriously considering switching. Reluctance to leave Gnus, however, kept me googling for some time and eventually I found mswatch. Really a trio of three utilities, mswatch uses Linux's file modification notification system (inotify or dnotify) to synchronize maildirs as needed. Initially, I tried mswatch with its intended sync program, mbsync, but mbsync cannot sync nested IMAP folders. The author of mswatch was kind enough to document clearly the interfaces used for the separate processes, so I set about writing a script that wraps OfflineIMAP to provide the mswatch interface. For the first time, I finally had a mail setup that delivered new mail quickly and didn't consume obscene amounts of bandwidth. But wait, I use Gnus splitting with BBDB, so whenever mail arrived in the INBOX on my server, it would be synced to my laptop which Gnus would then split into its destination and then it would be synced back to the server. This little dance doubles the bandwidth consumption. Furthermore, I've always wanted my BBDB splitting to be done on my server for those times when I use Gnus in an SSH session or when I access my mail from another IMAP client. So I also wanted to notify my gnus runnong on my server when when mail arrived. So then I set about writing a checker script that wraps the mswatch's watcher processes and uses "emacsclient --eval" to tell Gnus to check the folder where a change has occurred. On my server, this means the mail will be split before notifying the mswatch process that a change has occurred which means that mswatch will also synchronize the folder the message was moved to. No more round trip. This also has the added benefit of keeping Gnus up to date on all of my mail folders without having to use gnus-demon to periodically check all folders. All in all, *much* faster, very current information, and much less interruptions. There are many things about this solution that are hackish and inefficient so I'd say it's somewhere between %75 and %80 there, as it were. Given, however, that no previous solution has even broken %50 I'm happy as a clam for now. .. _`MUAs do indeed still suck`: http://groups.google.com/group/linux.debian.user/msg/7594165a2b6d1c49 Ross Patterson b51d9c661f066d5875649f8019b2a7007e169030 0.4