Vates Simple Interface and virtual desktops on RHEL 6 can crash MantidPlot

[Dennis Mikkelson]

The following steps consistently crash MantidPlot on RHEL 6 and Scientific Linux.

1. Load a TOPAZ file, for example TOPAZ_6251_event.nxs

2. ConvertToMD

3. Bring up the Vates Simple Interface (VSI) for the MD workspace

4. Choose the scatter plot icon

5. Switch to a different virtual desktop

6. Switch back to the virtual desktop with MantidPlot and the VSI

7. At this point the VSI window is still present, but the scatter plot is empty.

8. Choose the scatter plot icon again, and MantidPlot will seg fault.

The parameters used when loading and converting to MD are recorded in the script below:

Load(Filename='TOPAZ_6251_event.nxs',OutputWorkspace='TOPAZ_6251_event')

ConvertToMD(InputWorkspace='TOPAZ_6251_event',OutputWorkspace='TOPAZ_6251_MD',QDimensions='Q3D', dEAnalysisMode='Elastic',LorentzCorrection='1',MinValues='25,-25,-25',MaxValues='25,25,25', SplitInto='2,2,2',SplitThreshold='50',MaxRecursionDepth='13',MinRecursionDepth='7')

[Russell Taylor]

I can reproduce this and running in gdb shows the crash to come because the pqPipelineSource pointer passed into (and then used by) pqObjectBuilder::createFilter at line 90 of SplatterPlotView.cpp is NULL.

So some checks for non-nullness of the pointer and either regeneration of the pipeline or a clean take-down of the VSI would seem to be needed. This also applies to other views (e.g. ThreeSliceView.cpp line 167).

[Michael Reuter]

I FINALLY found the way to keep RHEL from causing the window closing code from executing when switching virtual desktops. BTW, Ubuntu does not suffer from this problem. As the fix is different from what is being requested here, I'm closing this ticket in favor of #6355.

[Russell Taylor]

Yes, it's OK now. Arguably the resolution should be something different, but there are two many hoops to jump through to get to that point so I won't bother trying.

[Stuart Campbell]

This ticket has been transferred to github issue ​7199