[Uml-devel] Spurious segfault on click on note or enum widgets in class diagram

[Achim Spangler]

Hi Daniel,

Am Mittwoch, 17. Mai 2006 23:28 schrieb Daniel Calviño Sánchez:
> Hi Achim,
> thanks for the patch. However, the real problem isn't in those
> methods. Let me explain.
>
> Those methods were wrong, but because I've missed the precondition in
> the documentation of the methods. I mean, the code is right, but the
> comments should explain that they can't be called with an empty list.
> If the precondition is violated, things can (and will) go awry. So
> it's responsability for the caller methods to ensure that the
> precondition isn't violated.
>
> So, why are they called with an empty list? As far as I can tell, it's
> a synchronization problem. Under certain circumstances, the method
> which handles the mouse move event is called before the method which
> handles the mouse press event has finished. So m_leftButtonDown was
> set to true, but selectSingle wasn't executed yet, so no widget was
> selected. When the selected widgets list is updated, no widget is
> selected so it goes empty, and then the getSmallest/BiggestX/Y methods
> crash.
>
> In my humble opinion, this should be controlled in ToolbarState. For
> example, queueing incoming events if there's one being processed, and
> process the queued ones one at a time once the process of the event
> being handled finishes. This way, no simultaneous handling of
> different events can occur.
Yes - this all sounds reasonable. I described therefore my patch as a simple 
way to cure the effects and not the source of the problem.

It simply fixes the central place that was indicated by the KDE crash report, 
where I found this NULL dereference.
>
> By the way Achim, would you mind sending me the XMI file which causes
> the crash? I've already one file which I can use to test this problem,
> but I'd prefer a wider test base. If you don't want to do it, no
> problem at all, it's only a suggestion :)
I think, that this XMI wouldn't help you, as this problem occured here with 
different relatively small projects. And really not each selection of the 
same widget causes this segfault.

I have the impression that also the order of previous actions (which tool was 
selected before,...).
Please try on your own with a class diagram, where you place among some class 
widgets also some packages, NOTES and ENUM-WIDGETS.
We had here some segfaults with NOTES and ENUM selection.

ONE SUGGESTION:
Could you integrate in those functions some monitoring, so that any occurance 
of that problem leads to a NOT_FOCUS_STEALING_INFORMATION and maybe a 
explicit debug-information file, that is created in the same directory as the 
current XMI file?
I would be willing to test umbrello with this debug setting, and report any 
clues we could gather (any systematic..).

Bye,
Achim
>
> 2006/5/17, Achim Spangler <A.Spangler at osb-ag.de>:
> > Hi again,
> > sorry - copy&paste lead to a wrong fix for the last function in this
> > group.
> >
> > Attached diff fixes this.
> >
> > Bye,
> > Achim
> >
> > Am Mittwoch, 17. Mai 2006 18:10 schrieb Achim Spangler:
> > > Hi,
> > > the current version of umbrello from REPO suffers sometimes a segfault,
> > > when simple items like a note or enum widget is clicked.
> > >
> > > The relevant part of the stacktrace is as follows:
> > > [KCrash handler]
> > > #6  0x08232efe in UMLWidgetController::getSmallestX ()
> > > #7  0x0823319d in UMLWidgetController::updateSelectionBounds ()
> > > #8  0x082333c0 in UMLWidgetController::mouseMoveEvent ()
> > > #9  0x081b6dd3 in ToolBarState::mouseMove ()
> > > #10 0x081b7a56 in ToolBarStateArrow::mouseMove ()
> > > #11 0xb7243dc7 in QScrollView::viewportMouseMoveEvent ()
> > >
> > >
> > > The following functions were a little unsafe:
> > > int UMLWidgetController::getSmallestX(const UMLWidgetList &widgetList)
> > > { int UMLWidgetController::getSmallestY(const UMLWidgetList
> > > &widgetList) { int UMLWidgetController::getBiggestX(const UMLWidgetList
> > > &widgetList) { int UMLWidgetController::getBiggestY(const UMLWidgetList
> > > &widgetList) {
> > >
> > > The problem was:
> > > When the widgetList is empty, so that the function it.toFirst() returns
> > > NULL, those functions dereferenced a NULL pointer without any check.
> > >
> > > I fixed this in the attached patch.
> > > I don't know, whether this is the only reason for the segfault. But
> > > this was in my opinion in any case a riscy code.
> > >
> > > Bye,
> > > Achim
> >
> > --
> > Achim Spangler
> > OSB AG Ingenieur- und IT-Dienstleistungen
> > Klenzestraße 38
> > 80469 München
> > Fon:   +49 (0) 89/23 88 57-49
> > Fax:   +49 (0) 89/23 88 57-40
> > Mobil: +49 (0)179/29 22 888
> > Email: a.spangler at osb-ag.de
> > WWW:  http://www.osb-ag.de/
>
> -------------------------------------------------------
> Using Tomcat but need to do more? Need to support web services, security?
> Get stuff done quickly with pre-integrated technology to make your job
> easier Download IBM WebSphere Application Server v.1.0.1 based on Apache
> Geronimo http://sel.as-us.falkag.net/sel?cmd=lnk&kid0709&bid&3057&dat1642
> _______________________________________________
> Uml-devel mailing list
> umbrello-devel at kde.org
> https://mail.kde.org/mailman/listinfo/umbrello-devel