X configuration paradigm, and a proposal

[Jim Gettys]

On Mon, 2004-11-22 at 16:32 -0600, Ted Kaminski wrote:
> On Thu, 2004-11-11 at 14:23, Jim Gettys wrote:
> > The point is that X is only one component of a fully configurable
> > desktop; it should be a good citizen and do what it is told, rather than
> > the current, obsolete view of X owning everything, and having to be
> > restarted to chance a configuration, and fighting with other
> > applications that might need the same resources.
> 
> Has anyone considered moving configuration out of the X server
> completely? This could be done by having the X server do nothing but
> start listening when it first starts, and wait for a configuration
> client to connect and tell it which devices to start using.

Yes, we are actively considering this.

We are actively prototyping doing this with input device right now,
and there have been serious discussions of moving mode selection out of
the X server, and just telling the X server what the configuration of
the frame buffer is.

The input device part is pretty straightforward, the mode selection
piece is more "challenging".

Help, as always, greatly appreciated.
				- Jim

> 
> Compatibility could be preserved by shipping two clients along with
> Xorg, the autodetect client and the current configuration format client.
> (nicely compartmentalizing that code)
> 
> This solves a number of problems:
>  1. Configration format. Human readable vs Machine readable is up to the
> client.
>  2. Dependancies. The X server would never have to start depending on
> HAL or gconf, for example.
>  3. Hotplug. The design essentially requires that hotplug be supported.
> 
> and creates one problem (that I can foresee)
>  1. Security. The X server is root, and the user is telling it which
> devices to use. Uh oh.
> 
> This idea doesn't directly address the problem of binary driver
> installation complexity, but the suggested solution for that (multiple
> file configuration format?) could certainly be implemented as a third
> configuration client shipped with Xorg (and become the new default).
> 
> The work necessary to achieve this seems to center around two things:
>  1. Hotplug. This already appears to be a relatively short term goal.

Underway: Kristian Hogsberg is leading the charge, and I'm helping out.
More help, as always, desirable.

>  2. Nonroot server. Unless I'm mistaken there's some slow work towards
> this, correct?

The technical issue is mostly that a lot/most of PC hardware is brain
dead, without the right hooks.  X has certainly run on workstation
hardware in the past without root.

>  (3. As long as input stuff is getting shaken up in the process, is
> there any reason we continue to pretend there's only ONE "core" mouse
> and ONE "core" keyboard?)

No, we have to fix this.

Shared projector systems have serious need for multiple simultaneous
streams of input.

I recently had a chance to play with the projector table at MERL, and we
can expect similar such devices as commercial products over the next
couple years; it provides a compelling demonstration of the need.

				- Jim