Re: NET WM Spec implementation and changes

["Michael Rogers" <mrogers cs ucl ac uk>]

> Hmm. My impression was that this portion of the ICCCM was
> superceded by the X Session Management Protocol (first used
> by CDE, I guess), which moved the responsiblity for session
> management from the window manager to a (possible seperate)
> session manager. The session manager notifies those clients
> which connect to it when they should save their state, and
> it is then the client's responsibility to save whatever
> information they feel is appropriate. The window manager
> therefore would no longer have a responsibility to record
> any of the layout information (unless it wished to remain
> backwards compatible with the older ICCCM session management.)

This sounds like a sensible approach - the session manager can confidently 
identify which window in the new session corresponds to which window in the 
old session; the window manager would have to rely on things like window class 
and window role being unique and persistent.

However, the session manager (and the clients) may not be aware of things like 
multiple desktops which are relevant when restoring a window's position. While 
there's a hint in version 1.9e of the spec to let clients know which desktop 
they're on, we can't expect every legacy app to be modified to save and 
restore this information as well as its window geometry. So to implement 
multiple desktops correctly the window manager ends up having to save and 
restore window information, match windows from the new session with windows 
from the old session, etc. And once you've put that code in the window manager 
you might as well avoid duplicating it in every client...


Michael