TYP: to_arrays, BUG: from_records empty dtypes

doc/source/whatsnew/v1.3.0.rst

@@ -401,7 +401,7 @@ Conversion
 ^^^^^^^^^^
 - Bug in :meth:`Series.to_dict` with ``orient='records'`` now returns python native types (:issue:`25969`)
 - Bug in :meth:`Series.view` and :meth:`Index.view` when converting between datetime-like (``datetime64[ns]``, ``datetime64[ns, tz]``, ``timedelta64``, ``period``) dtypes (:issue:`39788`)
--
+- Bug in creating a :class:`DataFrame` from an empty ``np.recarray`` not retaining the original dtypes (:issue:`40121`)
 -
 
 Strings

pandas/core/frame.py

@@ -609,6 +609,8 @@ def __init__(
                 if is_dataclass(data[0]):
                     data = dataclasses_to_dicts(data)
                 if treat_as_nested(data):
+                    if columns is not None:
+                        columns = ensure_index(columns)
                     arrays, columns, index = nested_data_to_arrays(
                         data, columns, index, dtype
                     )

pandas/core/internals/construction.py

@@ -586,7 +586,9 @@ def dataclasses_to_dicts(data):
 # Conversion of Inputs to Arrays
 
 
-def to_arrays(data, columns: Optional[Index], dtype: Optional[DtypeObj] = None):
+def to_arrays(
+    data, columns: Optional[Index], dtype: Optional[DtypeObj] = None
+) -> Tuple[List[ArrayLike], Index]:
     """
     Return list of arrays, columns.
     """
@@ -607,8 +609,10 @@ def to_arrays(data, columns: Optional[Index], dtype: Optional[DtypeObj] = None):
         if isinstance(data, np.ndarray):
             columns = data.dtype.names
             if columns is not None:
-                return [[]] * len(columns), columns
-        return [], []  # columns if columns is not None else []
+                # i.e. numpy structured array
+                arrays = [data[name] for name in columns]
+                return arrays, ensure_index(columns)
+        return [], ensure_index([])
 
     elif isinstance(data[0], Categorical):
         if columns is None:

pandas/tests/frame/constructors/test_from_records.py

@@ -11,6 +11,7 @@
     CategoricalIndex,
     DataFrame,
     Index,
+    Int64Index,
     Interval,
     RangeIndex,
     Series,
@@ -437,11 +438,11 @@ def test_from_records_empty(self):
     def test_from_records_empty_with_nonempty_fields_gh3682(self):
         a = np.array([(1, 2)], dtype=[("id", np.int64), ("value", np.int64)])
         df = DataFrame.from_records(a, index="id")
-        tm.assert_index_equal(df.index, Index([1], name="id"))
-        assert df.index.name == "id"
-        tm.assert_index_equal(df.columns, Index(["value"]))
-
-        b = np.array([], dtype=[("id", np.int64), ("value", np.int64)])
-        df = DataFrame.from_records(b, index="id")
-        tm.assert_index_equal(df.index, Index([], name="id"))
-        assert df.index.name == "id"
+
+        ex_index = Int64Index([1], name="id")
+        expected = DataFrame({"value": [2]}, index=ex_index, columns=["value"])
+        tm.assert_frame_equal(df, expected)
+
+        b = a[:0]
+        df2 = DataFrame.from_records(b, index="id")
+        tm.assert_frame_equal(df2, df.iloc[:0])

pandas/tests/frame/test_constructors.py

@@ -1169,7 +1169,8 @@ def test_constructor_unequal_length_nested_list_column(self):
         # GH 32173
         arrays = [list("abcd"), list("cde")]
 
-        msg = "Length of columns passed for MultiIndex columns is different"
+        # exception raised inside MultiIndex constructor
+        msg = "all arrays must be same length"
         with pytest.raises(ValueError, match=msg):
             DataFrame([[1, 2, 3, 4], [4, 5, 6, 7]], columns=arrays)