help needed debugging a crash

[Dag]

Mandag den 16. april 2012 13:36:17 Jaroslaw Staniek skrev:
> On 16 April 2012 13:10, Dag <danders at get2net.dk> wrote:
> > Mandag den 16. april 2012 12:52:39 Jaroslaw Staniek skrev:
> >> On 16 April 2012 12:42, Dag <danders at get2net.dk> wrote:
> >> > :Hi, I have a crash in plan on exit which I don't know how to squash.
> >> > 
> >> > https://bugs.kde.org/show_bug.cgi?id=298075
> >> > 
> >> > It happends sometimes (today: once in ~30 runs). I hoped it was a
> >> > koffice
> >> > issue because I had not seen it in calligra untill a few days ago.
> >> > 
> >> > The crash is always in a dbus object in qobject.cpp:854 where it tries
> >> > to
> >> > access a declarative data object:
> >> > 
> >> > 853 if (d->declarativeData)
> >> > 854      QAbstractDeclarativeData::destroyed(d->declarativeData, this);
> >> > 
> >> > When running gdb, I can see that d->declarativeData != 0, but the
> >> > destroyed() function pointer == 0, hence crash.
> >> > 
> >> > I cannot find any use of declaratives in dbus, but I might not have
> >> > looked
> >> > in the right place.
> >> > 
> >> > The only theory I have is that there a bug somewhere that overwrites
> >> > d->declarativeData.
> >> > 
> >> > How to catch this?
> >> > Anybody?
> >> 
> >> Could Valgrind help?
> > 
> > I have one crash when running in valgrind but afaics there is nothing
> > obvious. I get some "invalid read size", "Use of uninitialised value" and
> > "Conditional jump or move depends on uninitialised value" but it does not
> > seem to originate in plan specific code.
> 
> Not even in calligra libs' code?
> 
> If an invalid pointer is passed to Qt API and accessed by Qt code
> itself, it would be catched at Qt level indeed.
> So this may happen.
Yes, a couple:
KoApplication constructor and KoDocument::openFile()

I send you the log file if you want to look at it.
-- 
Mvh.
Dag Andersen