Input device design (3)

[Christoph Hellwig]

On Thu, Sep 01, 2005 at 09:37:09AM -0400, Joe Krahn wrote:
> I think we can agree not to send events through an extra level of IPC, 
> except in the case of a remote or virtual devices.
> 
> I like the idea of an I/O library. Consider a library of I/O handlers 
> that packages events using an X-centric view, but still rather generic. 
> An X Input loadable module is then built by combining the code for X 
> server integration with the I/O code from a static library. This 
> minimizes the I/O libraries knowledge of X. The same code can then be 
> used shared or static by non-X programs.
> 
> Does that sound like a good plan?

At least for Linux that's total overkill.  Linux has a proper input
layer, and you can access all input devices through /dev/input/event<N>
or a wide variaty of emulations for old device nodes (/dev/psaux,
/dev/js<N>, /dev/ts<N>), and thanks to uevent you can even feed it
input events from userspace.  No need to reinvent the wheel here, just
add a new input driver for linux that can use these and translate them
to normal and XInput extension X input sources.  Other OSes still keep
their old input drivers or if they have a similar input layer implement
a driver like that.  If there's enough common code though OSes refactor
that into a common library in the X server later.