Whole ton of work done. Let's see...
Autobus and libautobus now use json as the underlying transport. I modified afntest.autobus3 and afntest.autobus4 to use more threads, interfaces, and requests, and re-tested it with the new Autobus and libautobus. The results were astounding. The old, protobuf-using Autobus had a throughput of between 50 and 100 function calls per second. The new one that uses json as its underlying transport has a throughput of over 600 function calls per second. With protobuf as the transport, I could add ten timers to timerd before there was a delay of about a second between changing the timer's state and when the client-side object listener got notified of the change. With json as the transport, I got up to 50 timers with response still being immediate.
Autobus now supports server-side events, although I don't know if they work yet since I haven't added support for them into libautobus yet (so I have no way of testing them).
Timerd now has a new function on its timer interface, announce(timer_number). This function tells speakd to announce the timer's current time. Timerd also supports the timer attribute announce_on_state_change, which, if set to true for a particular timer, causes announce() to be automatically called by timerd when a client manually changes the timer's state. (This doesn't cause announcement when the timer gets reset to stopped when it hits zero while counting down; another attribute will soon provide support for that.)
Speakd now allows multiple spaces in the text passed to say_text between words; extra space will be ignored. This fixed a bug in timerd's new announce function that was causing an exception in speakd's say_text.