REF: share recarray constructor code (#40129)

Author: jbrockmendel

pandas/core/frame.py

@@ -178,10 +178,10 @@
     arrays_to_mgr,
     dataclasses_to_dicts,
     dict_to_mgr,
-    masked_rec_array_to_mgr,
     mgr_to_mgr,
     ndarray_to_mgr,
     nested_data_to_arrays,
+    rec_array_to_mgr,
     reorder_arrays,
     to_arrays,
     treat_as_nested,
@@ -580,7 +580,7 @@ def __init__(
 
             # masked recarray
             if isinstance(data, mrecords.MaskedRecords):
-                mgr = masked_rec_array_to_mgr(data, index, columns, dtype, copy)
+                mgr = rec_array_to_mgr(data, index, columns, dtype, copy)
 
             # a masked array
             else:
@@ -590,11 +590,7 @@ def __init__(
         elif isinstance(data, (np.ndarray, Series, Index)):
             if data.dtype.names:
                 # i.e. numpy structured array
-                data_columns = list(data.dtype.names)
-                data = {k: data[k] for k in data_columns}
-                if columns is None:
-                    columns = data_columns
-                mgr = dict_to_mgr(data, index, columns, dtype=dtype)
+                mgr = rec_array_to_mgr(data, index, columns, dtype, copy)
             elif getattr(data, "name", None) is not None:
                 # i.e. Series/Index with non-None name
                 mgr = dict_to_mgr({data.name: data}, index, columns, dtype=dtype)

pandas/core/internals/construction.py

@@ -117,8 +117,12 @@ def arrays_to_mgr(
     return create_block_manager_from_arrays(arrays, arr_names, axes)
 
 
-def masked_rec_array_to_mgr(
-    data: MaskedRecords, index, columns, dtype: Optional[DtypeObj], copy: bool
+def rec_array_to_mgr(
+    data: Union[MaskedRecords, np.recarray, np.ndarray],
+    index,
+    columns,
+    dtype: Optional[DtypeObj],
+    copy: bool,
 ):
     """
     Extract from a masked rec array and create the manager.
@@ -136,16 +140,10 @@ def masked_rec_array_to_mgr(
     arrays, arr_columns = to_arrays(fdata, columns)
 
     # fill if needed
-    new_arrays = []
-    for col in arr_columns:
-        arr = data[col]
-        fv = arr.fill_value
-
-        mask = ma.getmaskarray(arr)
-        if mask.any():
-            arr, fv = maybe_upcast(arr, fill_value=fv, copy=True)
-            arr[mask] = fv
-        new_arrays.append(arr)
+    if isinstance(data, np.ma.MaskedArray):
+        new_arrays = fill_masked_arrays(data, arr_columns)
+    else:
+        new_arrays = arrays
 
     # create the manager
     arrays, arr_columns = reorder_arrays(new_arrays, arr_columns, columns)
@@ -159,6 +157,24 @@ def masked_rec_array_to_mgr(
     return mgr
 
 
+def fill_masked_arrays(data: MaskedRecords, arr_columns: Index) -> List[np.ndarray]:
+    """
+    Convert numpy MaskedRecords to ensure mask is softened.
+    """
+    new_arrays = []
+
+    for col in arr_columns:
+        arr = data[col]
+        fv = arr.fill_value
+
+        mask = ma.getmaskarray(arr)
+        if mask.any():
+            arr, fv = maybe_upcast(arr, fill_value=fv, copy=True)
+            arr[mask] = fv
+        new_arrays.append(arr)
+    return new_arrays
+
+
 def mgr_to_mgr(mgr, typ: str):
     """
     Convert to specific type of Manager. Does not copy if the type is already

pandas/tests/frame/test_constructors.py

@@ -279,6 +279,7 @@ def test_constructor_rec(self, float_frame):
         tm.assert_index_equal(df2.columns, Index(rec.dtype.names))
         tm.assert_index_equal(df2.index, index)
 
+        # case with columns != the ones we would infer from the data
         rng = np.arange(len(rec))[::-1]
         df3 = DataFrame(rec, index=rng, columns=["C", "B"])
         expected = DataFrame(rec, index=rng).reindex(columns=["C", "B"])