[Bug] Broken VTK output for Compositional Multiphase Flow/Well simulations

[francoishamon]

Describe the bug When trying to use the VTK output for a compositional simulation in the tutorials, I got the following error:

***** LOCATION: /g/g90/hamon1/geosx/GEOSX/src/coreComponents/fileIO/vtk/VTKPolyDataWriterInterface.cpp:442                                     
***** Controlling expression (should be false): true                                                                                           
***** ERROR                                                                                                                                    
***** LOCATION: /g/g90/hamon1/geosx/GEOSX/src/coreComponents/common/DataTypes.hpp:917                                                          
***** Controlling expression (should be false): true                                                                                           
***** Rank 0: TypeID not recognized.                                                                                                           

** StackTrace of 8 frames **                                                                                                                   
Frame 0: geosx::vtk::VTKPolyDataWriterInterface::WriteField(geosx::dataRepository::WrapperBase const&, vtkSmartPointer<geosx::vtk::VTKGEOSXDat\
a>, long, long&) const                                                                                                                         
Frame 1: void geosx::vtk::VTKPolyDataWriterInterface::WriteElementFields<geosx::CellElementSubRegion>(vtkSmartPointer<vtkCellData>, geosx::Ele\
mentRegionBase const&) const                                                                                                                   
Frame 2: geosx::vtk::VTKPolyDataWriterInterface::WriteCellElementRegions(double, geosx::ElementRegionManager const&, geosx::NodeManager const&\
) const                                                                                                                                        
Frame 3: geosx::vtk::VTKPolyDataWriterInterface::Write(double, int, geosx::DomainPartition const&)                                             
Frame 4: geosx::EventBase::Execute(double, double, int, int, double, geosx::dataRepository::Group*)                                            
Frame 5: geosx::EventManager::Run(geosx::dataRepository::Group*)                                                                               
Frame 6: main                                                                                                                                  
Frame 7: __libc_start_main                                                                                                                     
Frame 8:                                                                                                                                       
===== 

This happens when the VTK writer is trying to output an array4d< real64 > such as phaseCompFraction.

If I roll back one commit to go before PR #1106 (which fixed the allocation of the constitutive models), then the VTK writer does not try to output any of the MultiFluid properties and the VTK writer works.

To Reproduce Steps to reproduce the behavior:

1. Modify a compositional integrated test to output with VTK
2. Run the test

Expected behavior It should work, and preferably, output the constitutive properties.

Platform (please complete the following information):

• LLNL/Quartz
• gcc 8.1.0
• GEOSX commit e107210

Additional context

• If I modify the following lambda to not error out when the type is not recognized: https://github.com/GEOSX/GEOSX/blob/e107210dbd9737394988695d2e6af5e8df48c623/src/coreComponents/fileIO/vtk/VTKPolyDataWriterInterface.cpp#L247-L249 it does not fix the problem and I get a crash later.
• If I try to add the type info for array4d in DataTypes.hpp, I get all sorts of compilation errors in DofManager, FieldSpecification, etc where the ApplyArrayTypeLambda2 is used.

@AntoineMazuyer any ideas?

[AntoineMazuyer]

Yes it's a known issue. I'll take care of that

[AntoineMazuyer]

What is the plot level of such an array4d ?

[AntoineMazuyer]

@francoishamon did pr #1131 solved the problem ?

[francoishamon]

@AntoineMazuyer Yes it solved the problem in the sense that VTK is no longer trying to output the array4d that was causing the problem. Do you think there is a way to modify the VTK output so that it outputs this field (phaseCompFraction)? Silo can output this field so it would be nice to have it with VTK as well.

[AntoineMazuyer]

@francoishamon it's an array2d isn't it ?

[AntoineMazuyer]

I can see

./constitutive/fluid/MultiFluidBase.hpp:159: arraySlice2d< real64 > const & phaseCompFraction, when I grep but also ./constitutive/fluid/MultiFluidBase.hpp:277: arrayView4d< real64 const > const & phaseCompFraction() const { return m_phaseCompFraction;

[francoishamon]

The phaseCompFraction is an array4d with the following entries:

• 0: element index
• 1: number of quadrature points (always equal to 1 in the finite-volume schemes)
• 2: number of phases
• 3: number of components

https://github.com/GEOSX/GEOSX/blob/33e1a8dd6525f22ac87a6d60fee9f385cb7d7a15/src/coreComponents/constitutive/fluid/MultiFluidBase.cpp#L116

and the arraySlice2d you saw is used for the point-wise update. For the Silo output, the function that takes care of outputting phaseCompFraction is: https://github.com/GEOSX/GEOSX/blob/33e1a8dd6525f22ac87a6d60fee9f385cb7d7a15/src/coreComponents/fileIO/silo/SiloFile.cpp#L965-L972

[francoishamon]

Solved by PR #1131. Thanks @af1990 !