Chombo: field viewer ("View in 3D") without Python — pure-Java volume and membrane VTU writers

[jcschaff]

Summary

The browser field viewer could not open any stored Chombo result: like MovingBoundary before #2155, the Chombo VTU grids came only from the Python VTK service, which the data server is never given (vcell.vtk.pythonDir). This PR writes them in Java, following #2155's design, and fixes what stood between a real Chombo run and the viewer.

• Pure-Java writers (VtkService.writeChomboVolumeVtkGridAndIndexData / writeChomboMembraneVtkGridAndIndexData, concrete on the base class, used on the desktop and the server with no Python fallback). VtuWriter gains:
  • volume: polygons (2D); voxels, tetrahedra, then cut cells as VTK_POLYHEDRON with their own faces (3D), in VTK 9.4's layout (version 2.3, face_connectivity/face_offsets/polyhedron_to_faces/polyhedron_offsets). This is what the Python service writes today; it no longer converts cut cells to tetrahedra (Chombo VisMesh is not watertight: cropVoxels clips each voxel independently, leaving EB-surface cracks and exposed interior walls #1895).
  • membrane: the surface points, then VTK_LINEs in 2D or VTK_TRIANGLEs in 3D.
  • ChomboIndexData in cell order, as Python wrote it.
• Hdf5PostProcessor rejects statistic name 'mean' (written by the chombo solver) and aborts all data access for the run #1894 fixed: Hdf5PostProcessor reads Chombo's <var>_mean statistic as the average. Before this, every Chombo result failed to open, fresh runs included.
• Directories: a desktop quick run (LocalVCDataIdentifier) uses its own directory. Server paths are unchanged.
• Field viewer:
  • /grid reports the embedding dimension. A 3D membrane (triangles only) used to come back as dimension: 2, so the viewer drew it top-down in the plane, like a 2D run. It is now drawn as a surface in 3D.
  • /timeseries&snap=nearest snaps probes onto a Chombo membrane variable's lines or triangles.
  • No viewer JS change was needed.

Evidence

Real data. I ran two Chombo runs locally with the vcell-chombo solver bundle (localsolvers/mac64/vcell-chombo), from Solver_Suite_6_2.vcml: "chombo 2D" (8×8 disk) and "chombo 3d" (8×8×8 ball: 56 voxels, 224 cut-cell polyhedra, 888 membrane triangles). Both are committed as a fixture (about 540 KB) in vcell-core/src/test/resources/org/vcell/vis/chombo/. The s2 (membrane) and RanC_nuc (volume) initial conditions encode the voxel each value was computed in, so the tests can check that each value is paired with the right cell. The dev cluster wasn't reachable from here, so no archived /simdata run was used.

Python equivalence (ChomboVtuPythonEquivalenceTest, real pythonVtk via Poetry, VTK 9.4.1), on VisMeshes built by ChomboMeshMapping from both runs. Java and Python produce identical cells, cell types, connectivity and polyhedron faces. The index files match byte for byte, and points agree to Float32 precision (max |dx| ≈ 4.6e-7):

Chombo 105373152 volume:   65 points, 32 cells, 32 indices
Chombo 105373152 membrane: 20 points, 20 cells, 20 indices
Chombo 104116603 volume:   907 points, 280 cells (224 polyhedra), 280 indices
Chombo 104116603 membrane: 446 points, 888 cells, 888 indices

Tests (each fails on the old code)

test	on old code
ChomboVtuWriterTest (2D, 3D): every volume and membrane value pairs with its cell; membrane on the sphere/circle	required System property "vcell.vtk.pythonDir" not defined (and No Statistic with name mean before #1894's fix)
ChomboVtuPythonEquivalenceTest (2D, 3D)	same
ChomboServerVtuPathTest: data server path, pythonDir unset	same
Hdf5PostProcessorTest.testReadChomboStatistics	No Statistic with name mean
FieldViewerServerChomboLocalRunTest (2D, 3D): /info lists _Membrane; /grid is a triangle surface in 3D with physical coordinates; /field pairs one value per triangle (and per volume cell); /timeseries snaps a click onto the membrane	3D: dimension 2; 2D: the membrane click misses (no snap); with the old core, pythonDir not defined
browser test_chombo_membranes.py (new chomboRun2d/chomboRun3d fixtures; Chromium, WebKit, Firefox: 12 passed)	3D membrane renders in 2D mode ("drag to pan")

Locally:

• mvn clean install dependency:copy-dependencies -DskipTests passes.
• vcell-core org.vcell.vis.**, Hdf5PostProcessorTest, CartesianMeshChomboTest and cbit.vcell.solvers.mb.* pass.
• vcell-client org.vcell.client.viz.* passes (all 12 classes).

Nothing serialized over the API changed, so I didn't run validate-openapi-spec.sh.

Not done / gaps

• No desktop end-to-end check through the debug bridge: the dev cluster was unreachable from this machine. The desktop path (LocalDataSetController → FieldViewerServer → browser) is covered by the local-run tests and the browser fixtures.
• Membrane kymographs over the VTU seam are still refused. On a 2D Chombo membrane the viewer offers the curve tool, and the server then rejects the request.
• Not exercised here: multi-level (refined) Chombo meshes. Both fixture runs have one level. chomboToPhysical uses the solver mesh's size, while ChomboMeshMapping uses the finest level's.

🤖 Generated with Claude Code