Re: QSettings deleted by singleton-cleanup mechanism

[Daniel J Sebald]

On 06/17/2015 05:05 PM, Mike Miller wrote:
> On Wed, Jun 17, 2015 at 16:26:06 -0500, Daniel J Sebald wrote:
> > On 06/17/2015 03:56 PM, John W. Eaton wrote:
> > > On 06/17/2015 03:43 PM, Daniel J Sebald wrote:
> > > > Currently the GUI's QSettings are instantiated and deleted by the Octave
> > > > worker thread using the singleton-cleanup mechanism. Can that be
> > > > changed? Was there a reason to have the worker, which is in another
> > > > thread, do this?
> > >
> > >
> > > I haven't looked, but was this changed when we tried to make it possible
> > > to open Qt plot widgets when Octave is started with --no-gui (the
> > > command line interface but linked with the Qt libraries)?
> >
> >
> > Oh, it probably was.  I think there are other ways that could be done, or it
> > might not even be necessary to have a QSettings in that case if nothing is
> > being used by it.  I'll investigate.
>
> I looked into it a little bit. I think it was done that way far before
> that more recent change.
>
> Both the "welcome wizard" and the Qt plot backend need access to the
> settings, so the settings definitely do need to be around and
> accessible when the main window is not up yet (or ever).

OK.  It turns out there isn't much change needed there.  This is all to 
do with the cleanup stage.  I believe I see how this singleton-cleanup 
came to be used.  There was likely segmentation faults when exiting the 
GUI that proved difficult to solve, so eventually this 
"very-last-action" approach to cleaning up the GUI memory space was done 
and it seemed to work for whatever reason.

However, after trying to make the settings save and resource_manager 
cleanup via qApp aboutToQuit() signal, I eventually tracked down the 
original source of problems, even though it was pretty hard to do this 
in such a transient action as exit.  Basically, the terminal was not 
being deleted by the related dock widget.  There was a similar bug in 
another object.  Any dynamically-created widget has to be deleted when 
its parent is deleted.  If not, the widget hangs around in the 
event-processing loop utilizing a parent and resources that no longer 
exist once everything else is destroyed.

A changeset is attached here:

https://savannah.gnu.org/bugs/index.php?45366

Look for the words '(bug fix)' in the changeset comments.

Dan