Address Copilot review on #136

- _get_coords_for_basis: destroy matInterp/vecScale after use so the
  per-variable eager cache fill and the nuke_coords_and_rebuild refill
  loop don't leak PETSc objects in long adapt runs.
- test_0780_ve_stokes_first_solve_mpi: require mpi(min_size=4). The
  bug only reproduces at the 4x2 partition geometry; min_size=2 would
  let CI pass with a partition that never hit the deadlock.
- Relabel the two bugfix comment blocks from TODO(BUG) to BUGFIX(#130)
  so they don't read as unfinished work.

Underworld development team with AI support from Claude Code

Author: lmoresi

src/underworld3/discretisation/discretisation_mesh.py

@@ -1653,10 +1653,10 @@ def nuke_coords_and_rebuild(
         # Invalidate projected boundary normals (rebuilt lazily on access)
         self._projected_normals = None
 
-        # TODO(BUG): issue #130 — refill the coord cache for every already-
-        # registered variable. Variables created before this rebuild would
-        # otherwise have their cache entry (from __init__) wiped above and
-        # refill lazily from rank-local code paths (rbf_interpolate), which
+        # BUGFIX(#130): refill the coord cache for every already-registered
+        # variable. Variables created before this rebuild would otherwise
+        # have their cache entry (from __init__) wiped above and refill
+        # lazily from rank-local code paths (rbf_interpolate), which
         # deadlocks when the collectives inside _get_coords_for_basis are
         # reached by only a subset of ranks.
         for _var in list(self.vars.values()):
@@ -2733,6 +2733,12 @@ def _get_coords_for_basis(self, degree, continuous):
 
         dmnew.restoreGlobalVec(coordsNewG)
         dmnew.restoreLocalVec(coordsNewL)
+        # Clean up the PETSc interpolation objects built above. Without this
+        # they accumulate until Python GC runs — noticeable in long adapt
+        # loops that re-fill the coord cache per variable.
+        matInterp.destroy()
+        if vecScale is not None:
+            vecScale.destroy()
         dmnew.destroy()
         dmfe.destroy()
 

src/underworld3/discretisation/discretisation_mesh_variables.py

@@ -388,8 +388,8 @@ def __init__(
         self.mesh.vars[self.clean_name] = self
         self._setup_ds()
 
-        # TODO(BUG): issue #130 — pre-populate the mesh's coordinate cache
-        # for this variable's basis. mesh._get_coords_for_basis contains MPI
+        # BUGFIX(#130): pre-populate the mesh's coordinate cache for this
+        # variable's basis. mesh._get_coords_for_basis contains MPI
         # collectives (DMClone, createInterpolation, globalToLocal) that
         # deadlock when triggered lazily from rank-local code paths (e.g.
         # rbf_interpolate inside global_evaluate_nd's per-particle loop):

tests/parallel/test_0780_ve_stokes_first_solve_mpi.py

@@ -27,12 +27,12 @@
 pytestmark = [
     pytest.mark.level_2,
     pytest.mark.tier_a,
-    pytest.mark.mpi(min_size=2),
+    pytest.mark.mpi(min_size=4),
     pytest.mark.timeout(60),
 ]
 
 
-@pytest.mark.mpi(min_size=2)
+@pytest.mark.mpi(min_size=4)
 def test_ve_stokes_first_solve_does_not_deadlock():
     """
     First VE_Stokes.solve() must complete under MPI on a partition-sensitive