Hello Michael,
it seems as if there is a bug in my contribution (VERY obscure one -
took me several days to figure it out). Basically, when several nodes
are deleted by the user, selectionChanged() signal is emitted after the
first of the nodes is removed. This leads to setting the "selected"
property on the other node that is yet to be deleted. This eventually
calls NodeChanged() of the data manager view which calls invalidate() on
the filter model. Turns out that this call stack, that is filter
invalidation /during/ row deletion, screws up the
QSortFilterProxyModel's internal state and leads to a crash in a very
obscure place within Qt code.
Could you please check this for me? The reproduction steps are really
simple. Fire up an empty workbench instance, show hidden nodes and nodes
with no data, select widget3Plane and widget2Plane under Widgets, and
delete them - this leads to crash in my case.
If it does crash - then I found the following solution to work. To avoid
immediate filter invalidation, I postpone it using Qt::QueuedConnection
slot invocation. So if you replace the code of NodeChanged() in
QmitkDataManagerView
from
m_FilterModel->invalidate();
to
QMetaObject::invokeMethod(m_FilterModel, "invalidate",
Qt::QueuedConnection);
then the crash willl be fixed.
Also during my struggles in searching for the causes of this crash, I
have found that in QmitkDataManagerView::RemoveSelectedNodes,
this->GlobalReinit(false) is called after every node is deleted. I doubt
it makes any sense, so I believe it should be moved out of the loop over
the nodes being deleted, so that GlobalReinit() is called only once all
selected nodes are removed. This should increase the performance.
Rostislav.
------------------------------------------------------------------------------
Slashdot TV.
Video for Nerds. Stuff that matters.
http://tv.slashdot.org/
_______________________________________________
mitk-users mailing list
mitk-users@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/mitk-users
