fix: scalar variable manifests get shape () not (1,) from kerchunk refs

virtualizarr/manifests/manifest.py

@@ -239,7 +239,7 @@ def __init__(
             path, offset, length = entry.values()
             entry = ChunkEntry.with_validation(path=path, offset=offset, length=length)  # type: ignore[attr-defined]
 
-            split_key = parse_manifest_index(key, separator)
+            split_key = () if shape == () else parse_manifest_index(key, separator)
             paths[split_key] = entry["path"]
             offsets[split_key] = entry["offset"]
             lengths[split_key] = entry["length"]

virtualizarr/parsers/kerchunk/translator.py

@@ -178,7 +178,13 @@ def manifestarray_from_kerchunk_refs(
     chunk_dict, metadata, zattrs = parse_array_refs(arr_refs)
     # we want to remove the _ARRAY_DIMENSIONS from the final variables' .attrs
     if chunk_dict:
-        manifest = manifest_from_kerchunk_chunk_dict(chunk_dict, fs_root=fs_root)
+        chunk_grid_shape = determine_chunk_grid_shape(
+            metadata.shape,
+            metadata.chunks,
+        )
+        manifest = manifest_from_kerchunk_chunk_dict(
+            chunk_dict, fs_root=fs_root, shape=chunk_grid_shape
+        )
         marr = ManifestArray(metadata=metadata, chunkmanifest=manifest)
     elif len(metadata.shape) != 0:
         # empty variables don't have physical chunks, but zarray shows that the variable
@@ -202,6 +208,7 @@ def manifestarray_from_kerchunk_refs(
 def manifest_from_kerchunk_chunk_dict(
     kerchunk_chunk_dict: dict[ChunkKey, str | tuple[str] | tuple[str, int, int]],
     fs_root: str | None = None,
+    shape: tuple[int, ...] | None = None,
 ) -> ChunkManifest:
     """Create a single ChunkManifest from the mapping of keys to chunk information stored inside kerchunk array refs."""
 
@@ -216,7 +223,7 @@ def manifest_from_kerchunk_chunk_dict(
             raise TypeError(f"Unexpected type {type(v)} for chunk value: {v}")
 
         chunk_entries[k] = chunkentry_from_kerchunk(v, fs_root=fs_root)
-    return ChunkManifest(entries=chunk_entries)
+    return ChunkManifest(entries=chunk_entries, shape=shape)
 
 
 def chunkentry_from_kerchunk(

virtualizarr/tests/test_integration.py

@@ -47,7 +47,7 @@ def test_kerchunk_roundtrip_in_memory_no_concat(array_v3_metadata):
     manifest = ChunkManifest(entries=chunks_dict)
     metadata = create_v3_array_metadata(
         shape=(2, 4),
-        chunk_shape=(2, 4),
+        chunk_shape=(2, 2),
         data_type=np.dtype("float32"),
     )
     marr = ManifestArray(

virtualizarr/tests/test_parsers/test_kerchunk.py

@@ -199,6 +199,23 @@ def test_kerchunk_parquet_sparse_array(tmp_path, local_registry):
         assert "1.0" not in manifest
 
 
+def test_scalar_variable_manifest_shape(refs_file_factory, local_registry):
+    """Scalar variables from kerchunk refs should produce a manifest with shape (), not (1,)."""
+    zarray = '{"chunks":[],"compressor":null,"dtype":"<i4","fill_value":null,"filters":null,"order":"C","shape":[],"zarr_format":2}'
+    zattrs = '{"_ARRAY_DIMENSIONS":[]}'
+    refs_file = refs_file_factory(
+        zarray=zarray,
+        zattrs=zattrs,
+        chunks={"a/0": ["/test1.nc", 100, 4]},
+    )
+    parser = KerchunkJSONParser()
+    with open_virtual_dataset(
+        url=refs_file, registry=local_registry, parser=parser
+    ) as vds:
+        assert vds["a"].shape == ()
+        assert vds["a"].data.manifest.shape_chunk_grid == ()
+
+
 def test_handle_relative_paths(refs_file_factory, local_registry):
     # deliberately use relative path here, see https://github.com/zarr-developers/VirtualiZarr/pull/243#issuecomment-2492341326
     refs_file = refs_file_factory(chunks={"a/0.0": ["test1.nc", 6144, 48]})