fix(metadata): interpret 0 fill value as "0" for Zarr V2 string arrays

zarrs/tests/data/v2_str0_zarr_python_v2.py

@@ -0,0 +1,42 @@
+#!/usr/bin/env -S uv run
+# /// script
+# requires-python = ">=3.12"
+# dependencies = [
+#     "zarr==2.18.4",
+# ]
+# ///
+
+import zarr
+
+path_out = "tests/data/zarr_python_compat/str_v2_fv_0.zarr"
+array = zarr.open(
+    path_out,
+    dtype=str,
+    shape=(5,),
+    chunks=(2,),
+    compressor=None,
+    mode='w',
+)
+array[0] = "a"
+array[1] = "bb"
+array[2] = ""
+print(array.info)
+print(array[:])
+assert (array[:] == ["a", "bb", "", "0", "0"]).all()
+
+path_out = "tests/data/zarr_python_compat/str_v2_fv_null.zarr"
+array = zarr.open(
+    path_out,
+    dtype=str,
+    shape=(5,),
+    chunks=(2,),
+    fill_value = None,
+    compressor=None,
+    mode='w',
+)
+array[0] = "a"
+array[1] = "bb"
+array[2] = ""
+print(array.info)
+print(array[:])
+assert (array[:] == ["a", "bb", "", "", None]).all() # Yikes

zarrs/tests/data/zarr_python_compat/str_v2_fv_0.zarr/.zarray

@@ -1,19 +1,18 @@
 {
-  "shape": [
-    5
-  ],
-  "chunks": [
-    2
-  ],
-  "fill_value": 0,
-  "order": "C",
-  "filters": [
-    {
-      "id": "vlen-utf8"
-    }
-  ],
-  "dimension_separator": ".",
-  "compressor": null,
-  "zarr_format": 2,
-  "dtype": "|O"
+    "chunks": [
+        2
+    ],
+    "compressor": null,
+    "dtype": "|O",
+    "fill_value": "0",
+    "filters": [
+        {
+            "id": "vlen-utf8"
+        }
+    ],
+    "order": "C",
+    "shape": [
+        5
+    ],
+    "zarr_format": 2
 }

zarrs/tests/data/zarr_python_compat/str_v2_fv_null.zarr/.zarray

@@ -1,19 +1,18 @@
 {
-  "shape": [
-    5
-  ],
-  "chunks": [
-    2
-  ],
-  "fill_value": null,
-  "order": "C",
-  "filters": [
-    {
-      "id": "vlen-utf8"
-    }
-  ],
-  "dimension_separator": ".",
-  "compressor": null,
-  "zarr_format": 2,
-  "dtype": "|O"
+    "chunks": [
+        2
+    ],
+    "compressor": null,
+    "dtype": "|O",
+    "fill_value": null,
+    "filters": [
+        {
+            "id": "vlen-utf8"
+        }
+    ],
+    "order": "C",
+    "shape": [
+        5
+    ],
+    "zarr_format": 2
 }

zarrs/tests/zarr_python_compat.rs

@@ -44,6 +44,7 @@ fn zarr_python_compat_fletcher32_v2() -> Result<(), Box<dyn Error>> {
     Ok(())
 }
 
+// Data generated by v2_str0_zarr_python_v2.py
 #[test]
 fn zarr_python_v2_compat_str_fv_0() -> Result<(), Box<dyn Error>> {
     let store = Arc::new(FilesystemStore::new(
@@ -53,11 +54,12 @@ fn zarr_python_v2_compat_str_fv_0() -> Result<(), Box<dyn Error>> {
     let subset_all = array.subset_all();
     let elements = array.retrieve_array_subset_elements::<String>(&subset_all)?;
 
-    assert_eq!(elements, &["a", "bb", "", "", ""]);
+    assert_eq!(elements, &["a", "bb", "", "0", "0"]);
 
     Ok(())
 }
 
+// Data generated by v2_str0_zarr_python_v2.py
 #[test]
 fn zarr_python_v2_compat_str_fv_null() -> Result<(), Box<dyn Error>> {
     let store = Arc::new(FilesystemStore::new(

zarrs_metadata/CHANGELOG.md

@@ -7,6 +7,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Fixed
+- Interpret a `0` fill value as `"0"` for Zarr V2 string arrays (for `zarr-python` 2.x.x compatibility)
+
 ## [0.3.3] - 2025-02-06
 
 ### Fixed

zarrs_metadata/src/v2_to_v3.rs

@@ -257,9 +257,10 @@
             }
         }
     } else if data_type.name() == "string" {
-        // Add a special case for `zarr-python` string data with a 0 fill value -> empty string
+        // Add a special case for string data with a 0 fill value -> "0"
+        // This was the default in zarr-python 2.x.x.
         if let Some(0) = fill_value.try_as_uint::<u64>() {
-            fill_value = FillValueMetadataV3::String(String::new());
+            fill_value = FillValueMetadataV3::String("0".to_string());
         }
     }