Backport PR pandas-dev#42394: REGR: unpickling in 1.3.0 DataFrame cre…

…ated in 1.2.x pandas-dev#42345

doc/source/whatsnew/v1.3.1.rst

@@ -15,6 +15,7 @@ including other versions of pandas.
 Fixed regressions
 ~~~~~~~~~~~~~~~~~
 - Pandas could not be built on PyPy (:issue:`42355`)
+- :class:`DataFrame` constructed with with an older version of pandas could not be unpickled (:issue:`42345`)
 -
 
 .. ---------------------------------------------------------------------------

pandas/_libs/internals.pyx

@@ -569,7 +569,12 @@ cdef class BlockManager:
         public bint _known_consolidated, _is_consolidated
         public ndarray _blknos, _blklocs
 
-    def __cinit__(self, blocks, axes, verify_integrity=True):
+    def __cinit__(self, blocks=None, axes=None, verify_integrity=True):
+        # None as defaults for unpickling GH#42345
+        if blocks is None:
+            # This adds 1-2 microseconds to DataFrame(np.array([]))
+            return
+
         if isinstance(blocks, list):
             # Backward compat for e.g. pyarrow
             blocks = tuple(blocks)

pandas/tests/io/test_pickle.py

@@ -163,9 +163,9 @@ def compare_index_period(result, expected, typ, version):
     tm.assert_index_equal(result.shift(2), expected.shift(2))
 
 
-files = glob.glob(
-    os.path.join(os.path.dirname(__file__), "data", "legacy_pickle", "*", "*.pickle")
-)
+here = os.path.dirname(__file__)
+legacy_dirname = os.path.join(here, "data", "legacy_pickle")
+files = glob.glob(os.path.join(legacy_dirname, "*", "*.pickle"))
 
 
 @pytest.fixture(params=files)
@@ -635,3 +635,13 @@ def test_pickle_big_dataframe_compression(protocol, compression):
         partial(pd.read_pickle, compression=compression),
     )
     tm.assert_frame_equal(df, result)
+
+
+def test_pickle_frame_v124_unpickle_130():
+    # GH#42345 DataFrame created in 1.2.x, unpickle in 1.3.x
+    path = os.path.join(legacy_dirname, "1.2.4", "empty_frame_v1_2_4-GH#42345.pkl")
+    with open(path, "rb") as fd:
+        df = pickle.load(fd)
+
+    expected = pd.DataFrame()
+    tm.assert_frame_equal(df, expected)