So, it looks like one big difference is that the hid-appleir driver didn't exist in the 3.8 kernel, and presumably LIRC was programming the chip directly before.
So I guess either (a) the driver is broken, or (b) LIRC is still trying to program the chip directly, and manages to confuse hid-appleir.
I guess I'll see if I can disable the driver and see what happens.
So, it looks like one big difference is that the hid-appleir driver didn't exist in the 3.8 kernel, and presumably LIRC was programming the chip directly before.
So I guess either (a) the driver is broken, or (b) LIRC is still trying to program the chip directly, and manages to confuse hid-appleir.
I guess I'll see if I can disable the driver and see what happens.