HAL doesn't work with grabbed or uinput created devices

[Richard Hughes]

On Tue, 2007-01-16 at 19:27 +0100, Michael Hanselmann wrote:
> On Tue, Jan 16, 2007 at 06:17:52PM +0000, Richard Hughes wrote:
> > > Input device -> Kernel (input.c) -> Kernel (evdev.c) ->
> > >     inputd (userland) -> Kernel (uinput.c) -> Kernel (evdev.c) ->
> > >         User application, X, HAL, whatever
> 
> > Crazy stuff - think about the cost of going kernel to user to kernel to
> > user for each keypress. What about:
> 
> Actually, I was thinking about such things in the beginning, too.
> Latency isn't a problem. Context switches might be, but in practice,
> aren't. It's a compromise. You only type like 10 characters per second,
> which is about nothing to a computer.

Have you ever programmed on embedded ARM where context switches are very
expensive? I think your statements above are dangerous assumptions to
make.

> > input device -> kernel -> X -> application ?
> 
> Won't work for people not using X. gpm works with inputd, but wouldn't
> with that way.

How does inputd get it's keyboard input without being run in X?

Richard.