Parse 0 fill value as "" for str dtype

Author: moradology

src/zarr/core/array.py

@@ -58,6 +58,7 @@
     _warn_order_kwarg,
     concurrent_map,
     parse_dtype,
+    parse_fill_value,
     parse_order,
     parse_shapelike,
     product,
@@ -3901,6 +3902,7 @@ async def init_array(
 
     from zarr.codecs.sharding import ShardingCodec, ShardingCodecIndexLocation
 
+    fill_value_parsed = parse_fill_value(fill_value, dtype, zarr_format)
     dtype_parsed = parse_dtype(dtype, zarr_format=zarr_format)
     shape_parsed = parse_shapelike(shape)
     chunk_key_encoding_parsed = _parse_chunk_key_encoding(
@@ -3947,7 +3949,7 @@ async def init_array(
             dtype=dtype_parsed,
             chunks=chunk_shape_parsed,
             dimension_separator=chunk_key_encoding_parsed.separator,
-            fill_value=fill_value,
+            fill_value=fill_value_parsed,
             order=order_parsed,
             filters=filters_parsed,
             compressor=compressor_parsed,
@@ -3985,7 +3987,7 @@ async def init_array(
         meta = AsyncArray._create_metadata_v3(
             shape=shape_parsed,
             dtype=dtype_parsed,
-            fill_value=fill_value,
+            fill_value=fill_value_parsed,
             chunk_shape=chunks_out,
             chunk_key_encoding=chunk_key_encoding_parsed,
             codecs=codecs_out,

src/zarr/core/array_spec.py

@@ -102,7 +102,9 @@ def __init__(
     ) -> None:
         shape_parsed = parse_shapelike(shape)
         dtype_parsed = np.dtype(dtype)
-        fill_value_parsed = parse_fill_value(fill_value)
+        # No zarr_format available here...
+        # fill_value_parsed = parse_fill_value(fill_value, dtype, zarr_format=2)
+        fill_value_parsed = fill_value
 
         object.__setattr__(self, "shape", shape_parsed)
         object.__setattr__(self, "dtype", dtype_parsed)

src/zarr/core/common.py

@@ -150,9 +150,12 @@ def parse_shapelike(data: int | Iterable[int]) -> tuple[int, ...]:
     return data_tuple
 
 
-def parse_fill_value(data: Any) -> Any:
+def parse_fill_value(fill_value: Any, dtype: Any, zarr_format: ZarrFormat) -> Any:
+    if zarr_format == 2:
+        if dtype is str or dtype == "str" and fill_value == 0:
+            fill_value = ""
     # todo: real validation
-    return data
+    return fill_value
 
 
 def parse_order(data: Any) -> Literal["C", "F"]:

tests/test_metadata/test_v2.py

@@ -298,3 +298,18 @@ def test_zstd_checksum() -> None:
         arr.metadata.to_buffer_dict(default_buffer_prototype())[".zarray"].to_bytes()
     )
     assert "checksum" not in metadata["compressor"]
+
+
+def test_0_fill_str_type():
+    array = zarr.create_array(
+        store=zarr.storage.MemoryStore(),
+        dtype=str,
+        shape=(5,),
+        chunks=(2,),
+        fill_value=0,
+        zarr_format=2,
+        overwrite=True,
+    )
+
+    # Ensure the array initializes correctly with the fill value
+    np.testing.assert_array_equal(array[:], ["", "", "", "", ""])