Add support for arrow large_string in cudf

python/cudf/cudf/core/column/column.py

@@ -1920,6 +1920,17 @@ def as_column(
         return col
 
     elif isinstance(arbitrary, (pa.Array, pa.ChunkedArray)):
+        if pa.types.is_large_string(arbitrary.type):
+            # Pandas-2.2+: Pandas defaults to `large_string` type
+            # instead of `string` without data-introspection.
+            # Temporary workaround until cudf has native
+            # support for `LARGE_STRING` i.e., 64 bit offsets
+            return as_column(
+                pd.Series(arbitrary),
+                nan_as_null=nan_as_null,
+                dtype=dtype,
+                length=length,
+            )
         if pa.types.is_float16(arbitrary.type):
             raise NotImplementedError(
                 "Type casting from `float16` to `float32` is not "

python/cudf/cudf/tests/test_series.py

@@ -2700,3 +2700,11 @@ def test_series_dtype_astypes(data):
     result = cudf.Series(data, dtype="float64")
     expected = cudf.Series([1.0, 2.0, 3.0])
     assert_eq(result, expected)
+
+
+def test_series_from_large_string():
+    pa_large_string_array = pa.array(["a", "b", "c"]).cast(pa.large_string())
+    got = cudf.Series(pa_large_string_array)
+    expected = pd.Series(pa_large_string_array)
+
+    assert_eq(expected, got)

python/cudf/cudf/utils/dtypes.py

@@ -213,6 +213,8 @@ def cudf_dtype_from_pa_type(typ):
         return cudf.core.dtypes.StructDtype.from_arrow(typ)
     elif pa.types.is_decimal(typ):
         return cudf.core.dtypes.Decimal128Dtype.from_arrow(typ)
+    elif pa.types.is_large_string(typ):
+        return cudf.dtype("str")
     else:
         return cudf.api.types.pandas_dtype(typ.to_pandas_dtype())