15:58 < wjt> it'd be useful if an observer could ask MC what channels it 
             *would* have been asked to observe, had it been running when they 
             were opened
15:59 < wjt> because otherwise when you're developing an observer and you 
             restart it, you need to restart all your channels before it 
             notices them
16:00 < sjoerd> being able to get all of the current channels might be good 
                enough for those use-cases though
16:01 < smcv> yeah, like the DispatchOperationList interface
16:01 < smcv> but MC already knows how to match channels against filters, so...
16:01 < sjoerd> yup
16:01 < smcv> it's inherently racy, of course - the observer working out what 
              it ought to be observing races with the handler ack'ing messages
16:02 < smcv> but it might be useful for observers that crash and auto-restart, 
              at least
16:02 < smcv> (in which case you already crashed, so potetially losing a few 
              messages is to be expected)
16:03 < wjt> hell, we could have an RestartMeIfICrash property