Use default cast for sliced list arrays if pyarrow >= 4

src/datasets/arrow_dataset.py

@@ -82,7 +82,7 @@
 
 logger = get_logger(__name__)
 
-if int(pa.__version__.split(".")[0]) == 0:
+if int(config.PYARROW_VERSION.split(".")[0]) == 0:
     PYARROW_V0 = True
 else:
     PYARROW_V0 = False
@@ -935,7 +935,7 @@ def cast_(
         schema = pa.schema({col_name: type[col_name].type for col_name in self._data.column_names})
         dataset = self.with_format("arrow")
         dataset = dataset.map(
-            lambda t: cast_with_sliced_list_support(t, schema),
+            lambda t: t.cast(schema) if config.PYARROW_VERSION >= "4" else cast_with_sliced_list_support(t, schema),
             batched=True,
             batch_size=batch_size,
             keep_in_memory=keep_in_memory,
@@ -994,7 +994,7 @@ def cast(
         format = self.format
         dataset = self.with_format("arrow")
         dataset = dataset.map(
-            lambda t: cast_with_sliced_list_support(t, schema),
+            lambda t: t.cast(schema) if config.PYARROW_VERSION >= "4" else cast_with_sliced_list_support(t, schema),
             batched=True,
             batch_size=batch_size,
             keep_in_memory=keep_in_memory,

src/datasets/config.py

@@ -37,6 +37,8 @@
 
 
 # Imports
+PYARROW_VERSION = importlib_metadata.version("pyarrow")
+
 USE_TF = os.environ.get("USE_TF", "AUTO").upper()
 USE_TORCH = os.environ.get("USE_TORCH", "AUTO").upper()
 

src/datasets/utils/file_utils.py

@@ -27,7 +27,6 @@
 
 import numpy as np
 import posixpath
-import pyarrow as pa
 import requests
 from tqdm.auto import tqdm
 
@@ -371,7 +370,7 @@ def cached_path(
 
 def get_datasets_user_agent(user_agent: Optional[Union[str, dict]] = None) -> str:
     ua = "datasets/{}; python/{}".format(__version__, config.PY_VERSION)
-    ua += "; pyarrow/{}".format(pa.__version__)
+    ua += "; pyarrow/{}".format(config.PYARROW_VERSION)
     if config.TORCH_AVAILABLE:
         ua += "; torch/{}".format(config.TORCH_VERSION)
     if config.TF_AVAILABLE:

tests/test_arrow_writer.py

@@ -6,6 +6,7 @@
 import pytest
 from packaging import version
 
+from datasets import config
 from datasets.arrow_writer import ArrowWriter, OptimizedTypedSequence, TypedSequence
 from datasets.features import Array2DExtensionType
 from datasets.keyhash import DuplicatedKeysError, InvalidKeyError
@@ -57,7 +58,7 @@ def test_try_incompatible_extension_type(self):
         self.assertEqual(arr.type, pa.string())
 
     def test_catch_overflow(self):
-        if version.parse(pa.__version__) < version.parse("2.0.0"):
+        if version.parse(config.PYARROW_VERSION) < version.parse("2.0.0"):
             with self.assertRaises(OverflowError):
                 _ = pa.array(TypedSequence([["x" * 1024]] * ((2 << 20) + 1)))  # ListArray with a bit more than 2GB
 

tests/test_table.py

@@ -6,6 +6,7 @@
 import pyarrow as pa
 import pytest
 
+from datasets import config
 from datasets.table import (
     ConcatenationTable,
     InMemoryTable,
@@ -745,7 +746,7 @@ def test_concatenation_table_cast(
             for k, v in zip(in_memory_pa_table.schema.names, in_memory_pa_table.schema.types)
         }
     )
-    if pa.__version__ < "4":
+    if config.PYARROW_VERSION < "4":
         with pytest.raises(pa.ArrowNotImplementedError):
             ConcatenationTable.from_blocks(blocks).cast(schema)
     else: