fix another typing thing

dask_deltatable/core.py

@@ -69,17 +69,17 @@ def _read_delta_partition(
     pyarrow_to_pandas["types_mapper"] = _get_type_mapper(
         pyarrow_to_pandas.get("types_mapper")
     )
-    return (
-        pa_ds.dataset(
-            source=filename,
-            schema=schema,
-            filesystem=fs,
-            format="parquet",
-            partitioning="hive",
-        )
-        .to_table(filter=filter_expression, columns=columns)
-        .to_pandas(**pyarrow_to_pandas)
+    pyarrow_to_pandas["ignore_metadata"] = pyarrow_to_pandas.get(
+        "ignore_metadata", False
     )
+    table = pa_ds.dataset(
+        source=filename,
+        schema=schema,
+        filesystem=fs,
+        format="parquet",
+        partitioning="hive",
+    ).to_table(filter=filter_expression, columns=columns)
+    return table.to_pandas(**pyarrow_to_pandas)
 
 
 def _read_from_filesystem(

tests/test_acceptance.py

@@ -47,6 +47,10 @@ def test_reader_all_primitive_types():
     expected_ddf = dd.read_parquet(
         f"{DATA_DIR}/all_primitive_types/expected/latest/table_content/*parquet"
     )
+    # Dask and delta go through different parquet parsers which read the
+    # timestamp differently. This is likely a bug in arrow but the delta result
+    # is "more correct".
+    expected_ddf["timestamp"] = expected_ddf["timestamp"].astype("datetime64[us]")
     assert_eq(actual_ddf, expected_ddf)