The problem is not seen in r19532 64bit windows binary.

Hmm, the back trace in the debugger hits the same error as in bug 1759, but I'm also getting an assert crash after I unhook the debugger, Slicer stays open, and I do File -> Exit:
ASSERT: "d->IsEntered == true" in file /projects/birn/nicole/Slicer4/Slicer4/Base/QTGUI/qSlicerAbstractModuleWidget.cxx, line 65

Is the crop volume module missing an enter() call? It has one, but is over riding the superclass, so adding in a call to the supercalss enter method and seeing if that helps.

svn 19547 fixes the assert problem (could be seen if started up slicer, entered CropVolume, then tried to switch to another module). Didn't see the crash on scene close this time, will test more.

Crash also happens if load a volume, go to editor to create a label volume, save scene, close scene.

If I change the scene view node destructor:

• this->Nodes->GetNodes()->RemoveAllItems();
• //this->Nodes->GetNodes()->RemoveAllItems();
• this->Nodes->Clear(1);

it changes the crash to happen if you restore a scene view and then close the scene (crash in referencing nodes of the crosshair node), but saving the mrml file (with scene view) then closing the scene doesn't crash due to volume node image data.

Still trying to figure out what the scene view saved scene thinks it's referenced/ing nodes are pointing to.

this->Nodes->GetNodes()->RemoveAllItems(); seems extremelly dangerous as it removes nodes without informing scene and node observers.

The RemoveAllItems calls triggers Delete on the nodes it contains and the node removed event does trigger a bit of an observer clean up.
The destructor of the volume nodes calls SetAndObserveImageData(NULL) which does remove observations on the image data, but the next problem is that it tries to iterate through it's display nodes and the ids are pointing to invalid nodes at that point.
The Clear call seems to be less dangerous, but I still need to look at the RestoreScene call to fix any dangerous calls there.

The Crop Volume error seems to be fixed as of svn 19588....

svn 19597 reverts some changes to how the scene views are saved/restored and this crash is not longer happening. Yanling, could you verify that it works for you and set this bug status to closed if so?