xserver: Branch 'master' - 2 commits

[Alan Hourihane]

On Tue, 2006-11-21 at 20:27 +0000, Alan Hourihane wrote:
> On Tue, 2006-11-21 at 20:24 +0000, Alan Hourihane wrote:
> > On Tue, 2006-11-21 at 21:52 +0200, Daniel Stone wrote:
> > > On Tue, Nov 21, 2006 at 07:22:24PM +0000, Alan Hourihane wrote:
> > > > On Sat, 2006-11-18 at 23:01 +0200, Daniel Stone wrote:
> > > > > On Sat, Nov 18, 2006 at 11:57:09AM -0800, Alan Hourihane wrote:
> > > > > >     Shutdown input devices if FatalError occurs during startup.
> > > > > >     Fixes Xdmx problems when the input device has been initialized
> > > > > >     and the keyboard map has been destroyed.
> > > > > 
> > > > > Hi Alan,
> > > > > Could you please elaborate on which problem this actually fixes?
> > > > > (Particularly as Xdmx doesn't even build currently.)  It's quite ugly,
> > > > > and presumably implies that FatalError/AbortDDX isn't doing its job of
> > > > > bringing the server down correctly.
> > > > 
> > > > Any server that initialises the keyboard and does anything funky with it
> > > > (as Xdmx does) via InitAndStartDevices() and then a FatalError occurs
> > > > (say because of a bad fontpath), the keyboard (as well as other input
> > > > devices) don't get cleaned up because they never get called with
> > > > DEVICE_OFF/DEVICE_CLOSE.
> > > > 
> > > > If there's some cleaner way, please go ahead.
> > > 
> > > If Xdmx really needs to have each device's close hook called, then it
> > > should be calling CloseDownDevices in AbortDDX, or OsCleanup should call
> > > CloseDownDevices itself.  You've only guarded against two or three fatal
> > > errors here, and there are many more ...
> > 
> > Exactly. Sounds like OsCleanup() should be the place ?
> 
> Or maybe in AbortServer() after OsCleanup() - what do you think ??

I've done this option. Let me know if you think it should be somewhere
else.

Thanks for reviewing Daniel.

Alan.