My students and I have built a prototype system, Javamlm, supporting user subscriptions on a per-thread basis using standard MUAs. I approached Barry about adding threads to Mailman, and he kindly encouraged me to do so. I'm now trying to figure out the best way to add the functionality and welcome any suggestions. You can get the full description of our system at http://www.mills.edu/ACAD_INFO/MCS/SPERTUS/Papers/lisa2001.pdf. I'll summarize important features here in the hope of getting feedback on the best way to fold them into Mailman.
FUNCTIONALITY
A user of mailing list foo creates a new thread with a base name of bar by sending email to foo-new-bar@host. The system appends an integer to bar to generate a unique thread name, e.g., bar1. All subscribers to mailing list foo receive the first message of any thread. The message headers are: To: foo-bar1@host From: original-sender@wherever If a user chooses reply, the new message goes to the original sender (assuming a decent MUA). It a user chooses reply-all, the new message becomes part of the thread bar1.
When subscribing (or subsequently), each user specifies whether they want to be subscribed to new threads by default. If so, the user gets all messages in a new thread (unless they explicitly unsubscribe). If a user chooses NOT to be subscribed to new threads by default, they only get the initial message unless they explicitly subscribe. Each message contains (un)subscribe information both as a URL and mailto to make the operations as lightweight as possible; e.g.,
To unsubscribe from this conversation, send email to foo-bar1-unsubscribe@javamlm.mills.edu or visit http://javamlm.mills.edu/scripts/override?listname=foo&conversation=14&preference=0. To unsubscribe from foo, send email to foo-unsubscribe@javamlm.mills.edu.
Note that thread membership is determined solely by the address to which a message is sent, not the message's subject, references, or other headers.
IMPLEMENTATION
We implemented threads by including a threadID in each message's metadata, which was stored in a relational database, as was information about each thread, subscription, etc. The following SQL query determines to whom to send a message in thread 37:
SELECT email FROM Subscriber WHERE deleted = FALSE AND ((Subscriber.preference = 1 AND NOT EXISTS (SELECT * FROM Override WHERE Override.subscriber_id = Subscriber.subscriber_id AND Override.thread_id = 37 AND Override.preference = 0)) OR (Subscriber.preference = 0 AND EXISTS (SELECT * FROM Override WHERE Override.subscriber_id = Subscriber.subscriber_id AND Override.thread_id = 37 AND Override.preference = 1)))
The full schema is in the paper. I'm including the query here to give you an idea of the complexity and to show that list membership is calculated on the fly.
The biggest implementation question in my mind is whether I should store subscription information in a relational database (so I can make the above efficient query) or build on top of the exiting pickles.
I welcome any comments and pointers.
Ellen