Find the split-checkpoint payload wherever the writer put it (#640 regressed by #756)

src/underworld3/discretisation/discretisation_mesh_variables.py

@@ -1229,11 +1229,20 @@ def _petsc_payload_name(data_file, data_name):
         if uw.mpi.rank == 0:
             try:
                 with h5py.File(data_file, "r") as h5f:
-                    topologies = h5f.get("topologies")
-                    for topology in (topologies or {}):
-                        dms = topologies[topology].get("dms")
-                        if dms is not None and data_name in dms:
-                            found = data_name
+                    # The payload sat at the file root before #756 and under
+                    # /restart/petsc after it. Look in both: this is only an
+                    # existence check, and a file written by either version
+                    # has to stay readable. Checking one place is what
+                    # regressed #640 -- the writer still emits the per-field
+                    # section and its vector, just one group deeper.
+                    for group in ("topologies", "restart/petsc/topologies"):
+                        topologies = h5f.get(group)
+                        for topology in (topologies or {}):
+                            dms = topologies[topology].get("dms")
+                            if dms is not None and data_name in dms:
+                                found = data_name
+                                break
+                        if found is not None:
                             break
             except (OSError, KeyError):
                 found = None
@@ -1253,10 +1262,20 @@ def _guard_coincident_dofs(self, data_file, data_name, is_v1_1,
         if self._lvec is None:
             self._set_vec(available=True)
         live_rows = self._gvec.getSize() // self.num_components
-        payload = None if is_v1_1 else self._petsc_payload_name(
-            data_file, data_name)
-
-        if payload is not None and saved_rows == live_rows:
+        # Look for the section in either format. `_petsc_payload_name` is a
+        # structural lookup (topologies/*/dms/<name>) and does not care which
+        # wrapper the file uses, and `read_timestep` calls
+        # `_read_native_payload` on a v1.1 file straight after this guard --
+        # so treating v1.1 as payload-free refused the very files the native
+        # path was about to read correctly (#640 regressed by #756).
+        payload = self._petsc_payload_name(data_file, data_name)
+
+        # v1.1 shares one compact geometry across the fields (#756), so there
+        # is no per-field saved cloud to count and `saved_rows` is 0 above by
+        # construction; the row comparison only means something for a legacy
+        # file. The payload carries the section, which is what resolves the
+        # cut, so in v1.1 its presence is the whole test.
+        if payload is not None and (is_v1_1 or saved_rows == live_rows):
             return  # `_read_native_payload` will take it
 
         raise RuntimeError(

tests/test_0864_split_checkpoint_roundtrip.py

@@ -149,3 +149,45 @@ def test_unsplit_mesh_still_uses_the_coordinate_remap(tmp_path):
     coords = np.asarray(loaded.coords_nd)
     assert np.allclose(loaded.data[:, 0],
                        100.0 * coords[:, 0] + 10.0 * coords[:, 1], atol=1e-8)
+
+
+@pytest.mark.level_1
+@pytest.mark.tier_b
+def test_the_payload_is_found_wherever_the_writer_put_it(tmp_path):
+    """The existence check must not be tied to one file layout.
+
+    #756 moved the reload payload from the file root into /restart/petsc.
+    The writer still emitted the per-field section and its vector, one
+    group deeper, but `_petsc_payload_name` looked only at the root, so it
+    reported "no PETSc-native payload" for a file written WITH
+    petsc_reload=True and the #640 guard refused the good path as well as
+    the bad. The round-trip test above catches the symptom; this pins the
+    cause, so a future relocation fails here with the reason rather than
+    as a confusing refusal.
+    """
+    import h5py
+
+    mesh = _split_mesh()
+    var = _stamped_variable(mesh)
+    _write(tmp_path, mesh, var, petsc_reload=True)
+    written = tmp_path / f"rt.mesh.{var.clean_name}.00000.h5"
+
+    # The section and its global vector are both on disk, somewhere.
+    with h5py.File(written, "r") as h5f:
+        groups = []
+        h5f.visititems(
+            lambda name, obj: groups.append(name)
+            if isinstance(obj, h5py.Group) else None)
+    assert any(g.endswith(f"dms/{var.clean_name}/section") for g in groups), groups
+    assert any(g.endswith(f"dms/{var.clean_name}/vecs/{var.clean_name}")
+               for g in groups), groups
+
+    # ... and the guard's existence check finds it there. The method lives on
+    # _BaseMeshVariable; EnhancedMeshVariable is a proxy and does not inherit
+    # from it, so reach for the defining class rather than type(var).
+    from underworld3.discretisation.discretisation_mesh_variables import (
+        _BaseMeshVariable,
+    )
+
+    assert _BaseMeshVariable._petsc_payload_name(
+        str(written), var.clean_name) == var.clean_name