chore: keep BatchedCodecPipeline as default

Restore the default codec pipeline to BatchedCodecPipeline.
SyncCodecPipeline remains available and is tested, but is opt-in via
the codec_pipeline.path config setting.

Tests that exercise SyncCodecPipeline-specific behavior (byte-range
writes for partial shard updates) now skip when a different pipeline
is active.

Also drop a few stale # type: ignore comments in sharding.py that
mypy now flags as unused.

Author: d-v-b

src/zarr/codecs/sharding.py

@@ -611,7 +611,7 @@ def _encode_partial_sync(
             total_data_size = n_chunks * chunk_byte_length
             total_shard_size = total_data_size + shard_index_size
 
-            existing = byte_setter.get_sync(prototype=shard_spec.prototype)  # type: ignore[attr-defined]
+            existing = byte_setter.get_sync(prototype=shard_spec.prototype)
             if existing is not None and len(existing) == total_shard_size:
                 key = byte_setter.path if hasattr(byte_setter, "path") else str(byte_setter)
                 shard_reader = self._shard_reader_from_bytes_sync(existing, chunks_per_shard)
@@ -662,7 +662,7 @@ def _byte_offset(coords: tuple[int, ...]) -> int:
                 morton_order_iter(chunks_per_shard)
             )
         else:
-            existing_bytes = byte_setter.get_sync(prototype=shard_spec.prototype)  # type: ignore[attr-defined]
+            existing_bytes = byte_setter.get_sync(prototype=shard_spec.prototype)
             if existing_bytes is not None:
                 shard_reader_fb = self._shard_reader_from_bytes_sync(
                     existing_bytes, chunks_per_shard
@@ -706,9 +706,9 @@ def _byte_offset(coords: tuple[int, ...]) -> int:
             buffer_prototype=default_buffer_prototype(),
         )
         if blob is None:
-            byte_setter.delete_sync()  # type: ignore[attr-defined]
+            byte_setter.delete_sync()
         else:
-            byte_setter.set_sync(blob)  # type: ignore[attr-defined]
+            byte_setter.set_sync(blob)
 
     def _encode_shard_dict_sync(
         self,

src/zarr/core/config.py

@@ -104,7 +104,7 @@ def enable_gpu(self) -> ConfigSet:
             "threading": {"max_workers": None},
             "json_indent": 2,
             "codec_pipeline": {
-                "path": "zarr.core.codec_pipeline.SyncCodecPipeline",
+                "path": "zarr.core.codec_pipeline.BatchedCodecPipeline",
                 "batch_size": 1,
             },
             "codecs": {

tests/test_config.py

@@ -61,7 +61,7 @@ def test_config_defaults_set() -> None:
                 "threading": {"max_workers": None},
                 "json_indent": 2,
                 "codec_pipeline": {
-                    "path": "zarr.core.codec_pipeline.SyncCodecPipeline",
+                    "path": "zarr.core.codec_pipeline.BatchedCodecPipeline",
                     "batch_size": 1,
                 },
                 "codecs": {

tests/test_sync_pipeline.py

@@ -520,7 +520,11 @@ def test_partial_shard_write_roundtrip_correctness() -> None:
 
 
 def test_partial_shard_write_uses_set_range() -> None:
-    """Partial shard writes with fixed-size codecs should use set_range_sync."""
+    """Partial shard writes with fixed-size codecs should use set_range_sync.
+
+    Only the SyncCodecPipeline uses byte-range writes for partial shard
+    updates; skipped under other pipelines.
+    """
     from unittest.mock import patch
 
     store = zarr.storage.MemoryStore()
@@ -533,6 +537,9 @@ def test_partial_shard_write_uses_set_range() -> None:
         compressors=None,
         fill_value=0.0,
     )
+    if not isinstance(arr._async_array.codec_pipeline, SyncCodecPipeline):
+        pytest.skip("byte-range write optimization is specific to SyncCodecPipeline")
+
     # Initial full write to create the shard blob
     arr[:] = np.arange(100, dtype="float64")
 
@@ -552,7 +559,12 @@ def test_partial_shard_write_uses_set_range() -> None:
 
 
 def test_partial_shard_write_falls_back_for_compressed() -> None:
-    """Partial shard writes with compressed inner codecs should NOT use set_range."""
+    """Partial shard writes with compressed inner codecs should NOT use set_range.
+
+    Only meaningful under SyncCodecPipeline (which can use byte-range writes
+    for fixed-size inner codecs). Other pipelines never use set_range_sync,
+    so the assertion is trivially true and the test is uninformative.
+    """
     from unittest.mock import patch
 
     store = zarr.storage.MemoryStore()
@@ -565,6 +577,8 @@ def test_partial_shard_write_falls_back_for_compressed() -> None:
         compressors=GzipCodec(),
         fill_value=0.0,
     )
+    if not isinstance(arr._async_array.codec_pipeline, SyncCodecPipeline):
+        pytest.skip("byte-range write optimization is specific to SyncCodecPipeline")
     arr[:] = np.arange(100, dtype="float64")
 
     with patch.object(type(store), "set_range_sync", wraps=store.set_range_sync) as mock_set_range: