Re: Any more features?

[raster rasterman com]

On 26 Nov, Owen Taylor scribbled:
->  
->  Paul Warren <pdw@ferret.lmh.ox.ac.uk> writes:
->  
->  > Hi,
->  > 
->  > Whilst I think that the features that have been suggested in the last week
->  > or so are good, I think that we really need to put a "feature freeze" on
->  > our spec, and then get the thing up to a publishable standard.
->  > 
->  > The features that have been suggested are:
->  > 
->  > _NET_PING for determining if an app is still alive when closing it.
->  > _NET_PID to give the WM access to the clients pid.
->  > _NET_INSTANCE_ID to allow association of windows and processes.
->  > 
->  > Do we definitely want to include those 3? Are there any others?
->  
->  I think it is worth pointing out that these protocols/properties
->  are definitely getting close to session management.
->  
->  The SM_CLIENT_ID property allows the unique identification
->  of the session managed client for a certain window, and ICE
->  gives you a ping command.
->  
->  There are some issues with stock X11R6 session management
->  that need to be eventually addressed:
->  
->  - How do you identify the session-management client
->    for a process you fork/exec? (gnome-session uses an
->    artificial client id for the -sm-client-id 
->    command line argument and then assigns a real client
->    ID when the connection is established.)
->  
->  - How do you "snapshot" a particular app instead of the
->    entire session? (E.g., you drag from the titlebar
->    and create a launcher for that app, in its exact
->    state). When you relaunch such an app, how does the
->    window manager associate its geometry with the
->    geometry that it had before?
->  
->  But, my general feeling is that session-management issues
->  should be dealt with separately or left to a later version
->  of the spec. and I'd consider PING/PID/INSTANCE_ID
->  to be session-management issues.
->  
->  The first version should concentrate on the management
->  of _windows_.

the problem is session management doesnt let us map a widnow Id to an
application - not trivially - and it requires all apps to suddenly use
a very complex api and library. the 3 properties above provide a
trivial app->process mapping. they can be grafted on "efter the fact"
trivially with LD_PRELOAD hacks and / or patches to Xlib.

they also provide a way for more than one program to do the ammmping
independantly - with session management, unless the session manager
provides external information, only the SM knows the mappign between
session client ID and client.

so i think these are still very important.

-- 
--------------- Codito, ergo sum - "I code, therefore I am" --------------------
The Rasterman (Carsten Haitzler)    raster@rasterman.com     raster@valinux.com
                                    raster@enlightenment.org raster@linux.com
				    raster@zip.com.au