Well, yeah, but I still think that you can get 'good enough' without too much pain. But thats always what 'some other programmer' thinks when commenting on the task some one else has before them- goes with the territory. Any conceptually non trivial task follows the bell curve law of difficulty. Things always appear easiest before you really start traversing the state space, and after you find the path.

Anyway, search is well studied and its own entire industry. There are probably much better solutions out there even ahead of the proprietary solution line than what we're likely to stumble on with idle chat.

But even given all that, I think you could narrow on author, keyword and time to get down to just a handful of threads in which to actual phrase matching- so much so that you wouldn't have to use any fancy algorithms. Or just ignore phrase matching for now and just match on keyword, time and author and let the search user go in and work through the material.

Also- keep in mind you have a small but dedicated user base. You could provide a really basic search (time and author) and then let users who search start tagging threads and individual messages they pull up. Pull a page from the semantic web, so to speak.

Yeti