inputd vs. hal

[Johannes Berg]

Hi all,

I wanted to do one trivial thing last night: change my display 
brightness using the keyboard and noticed that it didn't work anymore. 
It used to work with gnome-power-manager by hal broadcasting that I 
pressed brightness-up or brightness-down.

After scratching my head for a while I realized that the problem was 
that I had inputd running. For the uninitiated, inputd is a small, 
currently experimental, daemon written by Michael that grabs all 
/dev/input/event* devices exclusively, and for each one creates (using 
the uinput facility) a new virtual one. The point is to allow having 
things coming from two different input devices be joined together into a 
single event. A useful example (and currently the only thing 
implemented) is having CTRL+left-button map to right-button for a mouse 
or more importantly the touchpad on the powerbooks. Also, inputd is 
intended to replace the kernel facility for creating a virtual middle 
and right mouse button on ADB based Apple laptops.

Now, since inputd was running, it had grabbed all /dev/input/event* 
devices so that hal didn't get any events from them any more. Note that 
of course hal was notified about the fact that new devices showed up, 
but none of them were actually bound by any hald-addon-keyboard because 
none of them had a physical device link since they were virtual.

I'm uncertain about how to handle this. It will obviously require a hal 
modification because it is essential that hal pick up events from such 
virtual devices that were added for each physical device. inputd takes 
care to create a device with the same descriptor as the original (same 
bustype etc), it only prefixes the device name with "[inputd] " at the 
moment. However, there doesn't seem to be a clean way of detecting the 
fact that a device is virtual, other than deducing from the fact that it 
has no physical device link yet is USB ;)

Possibly the best solution would be to define a new BUS_VIRTUAL constant 
and force all uinput devices to that from the kernel, and add it to the 
switch statement in hal's probe-input.c (the one that I previously 
cleaned up :) )

Thanks for reading,
johannes