Mesh construction crashes on empty ranks: navigation kd-tree build passes a 1-D empty array (StructuredQuadBox(elementRes=(2,2)) at np4)

[lmoresi]

Found during the #330 batch (2026-07-23): fixing the Swarm.save race unmasked this in tests/test_0003_save_load.py at np4 — the file previously hung before reaching it.

Any mesh construction where a rank ends up with zero local cells crashes:

mpirun -np 4 python -c "import underworld3 as uw; uw.meshing.StructuredQuadBox(elementRes=(2,2))"
# rank with no cells:
#   ValueError: Buffer has wrong number of dimensions (expected 2, got 1)
#   at src/underworld3/ckdtree.pyx:99, via _build_kd_tree_index
#   (discretisation_mesh.py:5154); other ranks then die on MPI_ERR_BUFFER
#   from the resulting collective divergence.

Two layers to it:

1. _build_kd_tree_index builds numpy.array(control_points_list) / numpy.array(centroids_list) — empty lists give 1-D (0,) arrays, which the KDTree memoryview (double[:,::1]) rejects.
2. Shaping them (0, dim) is not enough on its own: KDTree.__cinit__ takes &points[0][0] (ckdtree.pyx:104), which is invalid for zero rows — the constructor needs an explicit empty-index path (related to the fix(swarm): Track-0 stale-cache and migration-semantics fixes (BF-02/05/06/07/08 + #286) #313 empty-rank KDTree query guards, which protected callers but not construction).

Verified identical on unmodified development (54a4fc58). Reproduces with any partition that starves a rank of cells — tiny meshes are just the easy trigger; extreme partition imbalance on real meshes would hit the same wall.

Also worth noting: tests/test_0003_save_load.py runs cleanly under MPI up to this point once #330 is fixed; this is the remaining blocker to making that file part of a parallel gate.

Underworld development team with AI support from Claude Code

[lmoresi]

Fixed by PR #406, merged to development on 2026-07-24 after a public adversarial review round. (Manual close — PRs target development.)

Underworld development team with AI support from Claude Code