[Usability] Re: Usability Digest, Vol 20, Issue 22

[michael chang <thenewme91 gmail com>]

On 12/14/05, usability-request gnome org <usability-request gnome org> wrote:
> Message: 4
> Date: Wed, 14 Dec 2005 11:29:59 -0300
> From: Daniel F Moisset <dmoisset grulic org ar>
> Subject: Re: [Usability] Re: [Desktop_architects] Printing dialog and
>         GNOME
> To: Olav Vitters <olav bkor dhs org>
> Cc: usability gnome org, Andreas Tille <tillea rki de>
> Message-ID: <1134570599 28997 9 camel dwarf moisset>
> Content-Type: text/plain; charset="us-ascii"
>
> On Wed, 2005-12-14 at 01:18 +0100, Olav Vitters wrote:
> > >      You seem to guess the end: If a Nautilus process is killed a new
> > >      instance is started.  Well - this is one of the next boring
> > >      missfeatures:
> > >      If I want to kill a process I really want this.  Don't call me
> > >      stupid.  And I do not want to search for a configuration option
> > >      were I can tell that I *really* want to kill this beast.  So I
> > >      just renamed /usr/bin/nautilus before killing the process and
> > >      I'm waiting for the feature that you people start patching the
> > >      Kernel that /usr/bin/nautilus can not be renamed ... :-(
> >
> > That is just session management. When an important app crashes it will
> > restart it. Remove nautilus from the session and you are done. IIRC
> > gnome-panel, metacity and nautilus are set in the session like that.
>
> But now that I see it I think something can be done to satisfy both
> parts (and without adding a setting!):
>
> The session manager is a parent process of the apps, and so it can know
> when important elements of the desktop (like nautilus, or the panel)
> crash and restart them. That is a Good Thing(tm) IMO. However it is true
> that when those process die because of a SIGTERM we can be quite sure
> that they were stopped by deliberate user interaction, so restarting
> them in that case is deliberately ignoring a user request (or even more:
> actively reverting what the user has done, in front of his face), and
> that is bad usability.
>
> Perhaps gnome-session should be changed to avoid restarting apps killed
> by user request (SIGTERM, and perhaps SIGKILL).

I don't know about whether or not pushing 'Ctrl-Alt-Backspace' at the
X-Server sends a SIGTERM or not, but I know that often when X starts
acting up I've been known to use this key-combo to kill the X-server
and rely on the display manager to restart the X server regardless of
the fact that I intentionally closed it.  It would be the same concept
at a higher level.

There's also the issue of some kernel versions tending to
'spontaneously' kill processes when it runs out of memory - if it
kills the wrong one, the whole chain of processes under it down the
line would have to also be killed with it if the process didn't
restart itself.  [For example, if the window manager is killed, you
can't exactly switch windows or desktops or something to issue a
command to restart it.]  We don't know whether these are SIGTERMs or
SIGKILLs or some other signal.

That said, more intuitive behaviour is usually a better idea than
tradition if it works - so the question is whether or not it works,
and who each solution is useful for.  (I know things like Alt-SysRq-B
('emergency' soft reboot built into the kernel) is a debug-only level
option.)

--
~Mike
 - Just the crazy copy cat.