Re: GNUstep and session management

[Richard Frith-Macdonald]

On 2005-10-07 07:53:06 +0000 Chris Vetter <cbv@gmx.net> wrote:

> On Fri, 07 Oct 2005 06:58:51 +0000, Richard wrote:
> [...]
> > I think we should provide a (documented) mechanism to shut down an 
> > application for GNUstep.  This should probably be some message that the 
> > application will respond to like -powerOff, which will send a local 
> > NSWorkspaceWillPowerOffNotification within the receiving application and 
> > then call [NSApp terminate].
>
> Might be a good idea. However what if you (you in general) explicitly *want*
> your application to *ignore* NSWorkspaceWillPowerOffNotification (for
> whatever reason)?

Well, I was simplifying a bit for clarity ... the whole standard termination 
procedure is actually a bit more complex than that, and the value that the 
delegate returns is used by -terminate to determine whether to actually 
terminate, to abandon termination, or to defer termination.  That is to say, 
the ability to ignore termination is already present in MacOS-X, and we 
should aim for compatible behavior.

> OTOH, why not just have (sort of a forced) observer for
> NSWorkspaceWillPowerOffNotification in NSApplication that just calls
> -terminate: on it's respective delegate? No need to add another
> GNUstep-specific method (-powerOff). Just create the observer in
> NSApplication-setDelegate: and you're good to go.

The only problem with this is, as Roman pointed out, the order of delivery 
of the notification to different observers is undefined.
This means that any built-in observer which calls -terminate could be 
executed before other observers which should be doing some particular 
power-off cleanup before the app terminates.
We need to make sure that all observers receive 
NSWorkspaceWillPowerOffNotification *before* -terminate is called (unless 
one of the observers explicitly calls -terminate itsself ... in which case 
this is presumably what the developer who wrote that particular app wanted).

The only simple way I can see to do that is have a new method which sends 
the notification and then calls -terminate.

I don't see it as a problem that this would be a new GNUstep method ... 
since MacOS-X probably has something similar (used by the 'force quit' panel 
in the finder), but doesn't document it.  Apple can get away with that ... 
as they have their own session manager application which is able to use 
private undocumented APIs.  We need to make our API public, and try to keep 
it really simple so that, if apple make their mechanism public, we chan 
change to be compatible with minimal work.