Paraview 5.9.1 ghostCells


we have a custom cxx reader plugin. When updating ParaView from v5.8.1 to v5.9.1, we ran some small tests to ensure that the plugin is still doing what its supposed to do.

I observed that when I load the same data file with v5.8.1 and v5.9.1 in parallel (in this case 2 processes) and compare the data in a SpreadSheetView the solution differs. The difference is that when using v5.9.1 the ghostCells are displayed such that the data occurs twice. Once as regular cell and once marked as vtkGhostType.

Any idea why cells marked as vtkGhostType appear in the SpreadSheetView for v5.9.1?

In the reader plugin, each process reads a subset of the data file and then creates a vtkUnstructuredGrid. At the process boundaries ghostCells are added and marked as DUPLICATE.
The vtkUnstructuredGrid is then added to a vtkMultiBlockDataSet, which is the output of the reader. The number of the vtkUnstructuredGrids created depends on the input data, in the present case there is only one vtkUnstructuredGrid.

Might it be related to changes in the vtkMultiblockDataSet as discussed here: Deprecating MultiBlock datasets


maybe @Yohann_Bearzi can help

This is very likely to be caused by the changes in vtkMultiBlockDataSet. vtkDataTabulator needs to make sure that ghosts are filtered out. I’ll create an issue.

Edit: here’s the issue

@ThomasAC I just tried a 5.8.1. build, and the ghost data appears twice as well in the spread sheet. Am I missing something?

To reproduce:

  • Run ParaView with 2 ranks
  • Create a Wavelet, hit apply
  • Load Threshold, hit apply
  • Load Ghost Cells Generator, hit apply
  • Open a Spread Sheet View
  • Select Cell Data
  • Click on vtkGhostType to sort ghost values in a decreasing order in the Spread Sheet View, you’ll see ghosts with a value equal to 1.

I reproduce what Yohann sees, but using 5.7.0.

We could add an option in the SpreadSheet view to not show ghost cells/points, but it seems the current behavior is the same as in past versions.

I checked with my 5.8.1 build again and indeed the ghost data appears twice, too. Not sure what I did to overlook them before.

An option to hide ghost cell data, also when writing to csv, would be helpful. Written data files could be made independent of the number of processes used then.

Regarding the migration from vtkMultiBlockDataSet to vtkPartitionedDataSet mentioned, should the reader be updated in the near future?

vtkMultiBlockDataSet will still be supported for at least the next ParaView release, but updating to vtkPartitionedDataSetCollection when you can is recommended.