kiowidgets_kdirmodeltest fail

[Albert Astals Cid]

El dimecres, 10 d’octubre de 2018, a les 23:09:49 CEST, David Faure va escriure:
> On mercredi 10 octobre 2018 22:11:12 CEST Albert Astals Cid wrote:
> > El dimecres, 10 d’octubre de 2018, a les 16:33:43 CEST, Jaime va escriure:
> > > Hello,
> > > 
> > >   About
> > > 
> > > https://build.kde.org/job/Frameworks/job/kio/job/kf5-qt5%20SUSEQt5.10/429/
> > > testReport/junit/(root)/TestSuite/kiowidgets_kdirmodeltest/> 
> > >   I'm only able to reproduce the crash 1/7 times I run ctest -j3, never if
> > > 
> > > I run the test alone or run make test.
> > 
> > FWIW I can reproduce the ASAN error quite easily with
> > 
> > while [ true ]; do ASAN_OPTIONS=detect_leaks=0 ./bin/kdirmodeltest
> > testOverwriteFileWithDir; if [ $? -ne 0 ]; then break; fi; done
> > 
> > it doesn't go more than two or three loops.
> > 
> > >   I've tried applying what worked for me in the 4th revision of
> > > 
> > > https://phabricator.kde.org/D10742 to kdirmodel.cpp around line 600, but
> > > without success.
> > > 
> > >   Any idea how to fix it is welcomed.
> > 
> > From what i see, the problem is that you're storing QModelIndexes and using
> > them later when the model has changed.
> > 
> > This is not correct, you can only use a QModelIndex when you just got it or
> > when it's passed it to you, if you need to store it you need to use
> > QPersistentModelIndex.
> > 
> > I don't understand much the test, but i think the attached patch is a good
> > fix, i.e. use the model indexes as soon as they happen and not store them.
> > Also makes the test pass "forever" here.
> 
> Looks good to me.

Pushed.

Cheers,
  Albert