Fix _gvec not synced after solver runs, causing zero checkpoint data

After PETSc solves, solution data was copied only to _lvec (local vector)
but _gvec (global vector) was never updated. Since write_timestep and all
checkpoint methods write from _gvec, every saved field was zeros.

Fix adds _sync_lvec_to_gvec() to _BaseMeshVariable using the correct
subdm scatter pattern, and calls it:
- In all 3 base solver finalizations (SNES_Scalar, SNES_Vector, SNES_Stokes_SaddlePt)
- In all write/checkpoint methods (belt-and-suspenders)

The operation is idempotent so the double coverage is safe.

Underworld development team with AI support from Claude Code

Author: lmoresi

src/underworld3/cython/petsc_generic_snes_solvers.pyx

@@ -1669,9 +1669,11 @@ class SNES_Scalar(SolverBaseClass):
         self.u.vec.array[:] = lvec.array[:]
         self.mesh._stale_lvec = True
 
-        # Invalidate cached data views - PETSc buffer may have changed
-        # Handle both EnhancedMeshVariable (has _base_var) and direct _MeshVariable
+        # Sync _gvec so downstream consumers (write, stats) see the result
         target_var = getattr(self.u, "_base_var", self.u)
+        target_var._sync_lvec_to_gvec()
+
+        # Invalidate cached data views - PETSc buffer may have changed
         if hasattr(target_var, "_canonical_data"):
             target_var._canonical_data = None
 
@@ -2438,9 +2440,11 @@ class SNES_Vector(SolverBaseClass):
         self.u.vec.array[:] = lvec.array[:]
         self.mesh._stale_lvec = True
 
-        # Invalidate cached data views - PETSc buffer may have changed
-        # Handle both EnhancedMeshVariable (has _base_var) and direct _MeshVariable
+        # Sync _gvec so downstream consumers (write, stats) see the result
         target_var = getattr(self.u, "_base_var", self.u)
+        target_var._sync_lvec_to_gvec()
+
+        # Invalidate cached data views - PETSc buffer may have changed
         if hasattr(target_var, "_canonical_data"):
             target_var._canonical_data = None
 
@@ -3842,6 +3846,12 @@ class SNES_Stokes_SaddlePt(SolverBaseClass):
                 var.vec.array[:] = clvec.getSubVector(pressure_is).array[:]
         self.mesh._stale_lvec = True
 
+        # Sync _gvec so downstream consumers (write, stats) see the result
+        for name, var in self.fields.items():
+            target_var = getattr(var, "_base_var", var)
+            target_var._sync_lvec_to_gvec()
+            if hasattr(target_var, "_canonical_data"):
+                target_var._canonical_data = None
 
         self.dm.restoreGlobalVec(clvec)
         self.dm.restoreGlobalVec(gvec)

src/underworld3/discretisation/discretisation_mesh.py

@@ -1851,6 +1851,7 @@ def petsc_save_checkpoint(
         ### Empty meshVars will save just the mesh
         if meshVars != None:
             for var in meshVars:
+                var._sync_lvec_to_gvec()
                 viewer(var._gvec)
 
         viewer.destroy()
@@ -1903,6 +1904,7 @@ def write_checkpoint(
 
         if meshVars is not None:
             for var in meshVars:
+                var._sync_lvec_to_gvec()
                 iset, subdm = self.dm.createSubDM(var.field_id)
                 subdm.setName(var.clean_name)
                 self.dm.globalVectorView(viewer, subdm, var._gvec)
@@ -1912,6 +1914,7 @@ def write_checkpoint(
         if swarmVars is not None:
             for svar in swarmVars:
                 var = svar._meshVar
+                var._sync_lvec_to_gvec()
                 iset, subdm = self.dm.createSubDM(var.field_id)
                 subdm.setName(var.clean_name)
                 self.dm.globalVectorView(viewer, subdm, var._gvec)

src/underworld3/discretisation/discretisation_mesh_variables.py

@@ -977,6 +977,9 @@ def save(
         # Keep vector available for future access
         pass
 
+        # Ensure global vector is up-to-date before writing.
+        self._sync_lvec_to_gvec()
+
         viewer = PETSc.ViewerHDF5().create(filename, "a", comm=PETSc.COMM_WORLD)
         if index:
             raise RuntimeError("Recording `index` not currently supported")
@@ -1078,6 +1081,9 @@ def write(
         dmnew.localToGlobal(lvec, gvec, addv=False)
         gvec.setName("coordinates")
 
+        # Ensure global vector is up-to-date before writing.
+        self._sync_lvec_to_gvec()
+
         viewer = PETSc.ViewerHDF5().create(filename, "w", comm=PETSc.COMM_WORLD)
         viewer(self._gvec)
         viewer(gvec)
@@ -1601,6 +1607,23 @@ def vec(self) -> PETSc.Vec:
 
         return self._lvec
 
+    def _sync_lvec_to_gvec(self):
+        """Ensure the global vector reflects the current local vector.
+
+        After PETSc solves (Stokes, Projection, etc.), data lives only in
+        ``_lvec``.  Any operation that reads ``_gvec`` (writes, norms, etc.)
+        must call this first.  The scatter uses the variable's sub-DM so it
+        works for any field order (P1, P2, DG-0, …).
+
+        The operation is idempotent — calling it when ``_gvec`` is already
+        up-to-date is harmless.
+        """
+        if self._lvec is None or self._gvec is None:
+            return
+        indexset, subdm = self.mesh.dm.createSubDM(self.field_id)
+        subdm.localToGlobal(self._lvec, self._gvec, addv=False)
+        indexset.destroy()
+
     @property
     def old_data(self) -> numpy.ndarray:
         """