allow creating references for empty archival datasets

virtualizarr/manifests/manifest.py

@@ -141,6 +141,15 @@ def __init__(self, entries: dict) -> None:
         self._offsets = offsets
         self._lengths = lengths
 
+    @classmethod
+    def empty(cls) -> "ChunkManifest":
+        """Create an empty chunk manifest."""
+        return cls.from_arrays(
+            paths=np.array((), dtype=np.dtypes.StringDType),
+            offsets=np.array((), dtype="uint64"),
+            lengths=np.array((), dtype="uint64"),
+        )
+
     @classmethod
     def from_arrays(
         cls,
@@ -257,6 +266,9 @@ def dict(self) -> ChunkDict:  # type: ignore[override]
         Entries whose path is an empty string will be interpreted as missing chunks and omitted from the dictionary.
         """
 
+        if len(self) == 0:
+            return cast(ChunkDict, {})
+
         coord_vectors = np.mgrid[
             tuple(slice(None, length) for length in self.shape_chunk_grid)
         ]

virtualizarr/readers/kerchunk.py

@@ -243,6 +243,9 @@ def variable_from_kerchunk_refs(
     if chunk_dict:
         manifest = ChunkManifest._from_kerchunk_chunk_dict(chunk_dict)
         varr = virtual_array_class(zarray=zarray, chunkmanifest=manifest)
+    elif len(zarray.shape) != 0:
+        manifest = ChunkManifest.empty()
+        varr = virtual_array_class(zarray=zarray, chunkmanifest=manifest)
     else:
         # This means we encountered a scalar variable of dimension 0,
         # very likely that it actually has no numeric value and its only purpose

virtualizarr/tests/test_readers/test_kerchunk.py

@@ -61,3 +61,24 @@ def test_dataset_from_df_refs_with_filters():
     ds = dataset_from_kerchunk_refs(ds_refs)
     da = ds["a"]
     assert da.data.zarray.filters == filters
+
+
+def test_empty_netcdf4():
+    zarray = {
+        "chunks": [50, 100],
+        "compressor": None,
+        "dtype": "<i8",
+        "fill_value": 100,
+        "filters": None,
+        "order": "C",
+        "shape": [100, 200],
+        "zarr_format": 2,
+    }
+    refs = gen_ds_refs(zarray=ujson.dumps(zarray))
+    del refs["refs"]["a/0.0"]
+
+    ds = dataset_from_kerchunk_refs(refs)
+    assert "a" in ds.variables
+    assert isinstance(ds["a"].data, ManifestArray)
+    assert ds["a"].sizes == {"x": 100, "y": 200}
+    assert ds["a"].chunksizes == {"x": 50, "y": 100}

virtualizarr/tests/test_writers/test_kerchunk.py

@@ -39,6 +39,35 @@ def test_accessor_to_kerchunk_dict(self):
         result_ds_refs = ds.virtualize.to_kerchunk(format="dict")
         assert result_ds_refs == expected_ds_refs
 
+    def test_accessor_to_kerchunk_dict_empty(self):
+        manifest = ChunkManifest.empty()
+        arr = ManifestArray(
+            chunkmanifest=manifest,
+            zarray=dict(
+                shape=(2, 3),
+                dtype=np.dtype("<i8"),
+                chunks=(2, 3),
+                compressor=None,
+                filters=None,
+                fill_value=np.nan,
+                order="C",
+            ),
+        )
+        ds = Dataset({"a": (["x", "y"], arr)})
+
+        expected_ds_refs = {
+            "version": 1,
+            "refs": {
+                ".zgroup": '{"zarr_format":2}',
+                ".zattrs": "{}",
+                "a/.zarray": '{"shape":[2,3],"chunks":[2,3],"dtype":"<i8","fill_value":null,"order":"C","compressor":null,"filters":null,"zarr_format":2}',
+                "a/.zattrs": '{"_ARRAY_DIMENSIONS":["x","y"]}',
+            },
+        }
+
+        result_ds_refs = ds.virtualize.to_kerchunk(format="dict")
+        assert result_ds_refs == expected_ds_refs
+
     def test_accessor_to_kerchunk_json(self, tmp_path):
         manifest = ChunkManifest(
             entries={"0.0": dict(path="test.nc", offset=6144, length=48)}