Implement first-class List type

pandas/__init__.py

@@ -61,6 +61,7 @@
     PeriodDtype,
     IntervalDtype,
     DatetimeTZDtype,
+    ListDtype,
     StringDtype,
     BooleanDtype,
     # missing
@@ -261,6 +262,7 @@
     "Interval",
     "IntervalDtype",
     "IntervalIndex",
+    "ListDtype",
     "MultiIndex",
     "NaT",
     "NamedAgg",

pandas/_libs/lib.pyx

@@ -834,6 +834,9 @@ cpdef ndarray[object] ensure_string_array(
             if isinstance(val, bytes):
                 # GH#49658 discussion of desired behavior here
                 result[i] = val.decode()
+            elif isinstance(val, np.ndarray):
+                # TODO(wayd): is_float_object actually returns true for this...
+                result[i] = str(val.tolist())
             elif not util.is_float_object(val):
                 # f"{val}" is faster than str(val)
                 result[i] = f"{val}"

pandas/_testing/asserters.py

@@ -54,6 +54,7 @@
     TimedeltaArray,
 )
 from pandas.core.arrays.datetimelike import DatetimeLikeArrayMixin
+from pandas.core.arrays.list_ import ListDtype
 from pandas.core.arrays.string_ import StringDtype
 from pandas.core.indexes.api import safe_sort_index
 
@@ -824,6 +825,11 @@ def assert_extension_array_equal(
             [np.isnan(val) for val in right._ndarray[right_na]]  # type: ignore[attr-defined]
         ), "wrong missing value sentinels"
 
+    # TODO: not every array type may be convertible to NumPy; should catch here
+    if isinstance(left.dtype, ListDtype) and isinstance(right.dtype, ListDtype):
+        assert left._pa_array == right._pa_array
+        return
+
     left_valid = left[~left_na].to_numpy(dtype=object)
     right_valid = right[~right_na].to_numpy(dtype=object)
     if check_exact:

pandas/core/api.py

@@ -40,6 +40,7 @@
     UInt32Dtype,
     UInt64Dtype,
 )
+from pandas.core.arrays.list_ import ListDtype
 from pandas.core.arrays.string_ import StringDtype
 from pandas.core.construction import array  # noqa: ICN001
 from pandas.core.flags import Flags
@@ -103,6 +104,7 @@
     "Interval",
     "IntervalDtype",
     "IntervalIndex",
+    "ListDtype",
     "MultiIndex",
     "NaT",
     "NamedAgg",

pandas/core/arrays/arrow/array.py

@@ -428,7 +428,7 @@ def _box_pa_scalar(cls, value, pa_type: pa.DataType | None = None) -> pa.Scalar:
         """
         if isinstance(value, pa.Scalar):
             pa_scalar = value
-        elif isna(value):
+        elif not is_list_like(value) and isna(value):
             pa_scalar = pa.scalar(None, type=pa_type)
         else:
             # Workaround https://github.com/apache/arrow/issues/37291
@@ -1350,7 +1350,16 @@ def take(
                 # TODO(ARROW-9433): Treat negative indices as NULL
                 indices_array = pa.array(indices_array, mask=fill_mask)
                 result = self._pa_array.take(indices_array)
-                if isna(fill_value):
+                if is_list_like(fill_value):
+                    # TODO: this should be hit by ListArray. Ideally we do:
+                    # pc.replace_with_mask(result, fill_mask, pa.scalar(fill_value))
+                    # but pyarrow does not yet implement that for list types
+                    new_values = [
+                        fill_value if should_fill else x.as_py()
+                        for x, should_fill in zip(result, fill_mask)
+                    ]
+                    return type(self)(new_values)
+                elif isna(fill_value):
                     return type(self)(result)
                 # TODO: ArrowNotImplementedError: Function fill_null has no
                 # kernel matching input types (array[string], scalar[string])