Xfail NaN <-> NaT tests on non-x86 and warn on cast

pd.Series([np.nan]).astype('datetime64[ns]')[0] = pd.NaT on x86
but 1970-01-01 on arm* because float NaN -> int is undefined:
numpy/numpy#8325
pandas-dev/pandas#17792
pandas-dev/pandas#26964

On s390x it's the maximum _positive_ value (2**63-1 ns = year 2262)

Author: Andreas Tille <[email protected]>, Graham Inggs <[email protected]>, Rebecca N. Palmer <[email protected]>
Bug-Debian: https://bugs.debian.org/877754
Forwarded: no


Gbp-Pq: Name xfail_tests_nonintel_nannat.patch

Author: Debian Science Team

pandas/core/dtypes/cast.py

@@ -1,6 +1,10 @@
 """ routings for casting """
 
 from datetime import datetime, timedelta
+import warnings
+import platform
+import re
+warn_nannat_platform = "Non-x86 system detected, float -> datetime/timedelta may not handle NaNs correctly - https://bugs.debian.org/877754" if not bool(re.match('i.?86|x86',platform.uname()[4])) else False
 
 import numpy as np
 
@@ -891,6 +895,8 @@ def astype_nansafe(arr, dtype, copy: bool = True, skipna: bool = False):
             f"'{dtype.name}[ns]' instead."
         )
         raise ValueError(msg)
+    if warn_nannat_platform and (is_datetime64_dtype(dtype) or is_timedelta64_dtype(dtype)) and np.issubdtype(arr.dtype, np.floating) and not np.isfinite(arr).all():
+        warnings.warn(warn_nannat_platform)
 
     if copy or is_object_dtype(arr) or is_object_dtype(dtype):
         # Explicit copy, or required since NumPy can't view from / to object.
@@ -1263,6 +1269,8 @@ def maybe_cast_to_datetime(value, dtype, errors: str = "raise"):
                     value = iNaT
             else:
                 value = np.array(value, copy=False)
+                if warn_nannat_platform and np.issubdtype(value.dtype, np.floating) and not np.isfinite(value).all():
+                    warnings.warn(warn_nannat_platform)
 
                 # have a scalar array-like (e.g. NaT)
                 if value.ndim == 0:

pandas/tests/dtypes/cast/test_downcast.py

@@ -7,6 +7,9 @@
 
 from pandas import DatetimeIndex, Series, Timestamp
 import pandas._testing as tm
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86',platform.uname()[4]))
 
 
 @pytest.mark.parametrize(
@@ -77,6 +80,7 @@ def test_downcast_conversion_empty(any_real_dtype):
     tm.assert_numpy_array_equal(result, np.array([], dtype=np.int64))
 
 
+@pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
 @pytest.mark.parametrize("klass", [np.datetime64, np.timedelta64])
 def test_datetime_likes_nan(klass):
     dtype = klass.__name__ + "[ns]"

pandas/tests/frame/indexing/test_where.py

@@ -8,6 +8,9 @@
 import pandas as pd
 from pandas import DataFrame, DatetimeIndex, Series, Timestamp, date_range, isna
 import pandas._testing as tm
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 
 class TestDataFrameIndexingWhere:
@@ -340,6 +343,7 @@ def test_where_bug_transposition(self):
         result = a.where(do_not_replace, b)
         tm.assert_frame_equal(result, expected)
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)#not found
     def test_where_datetime(self):
 
         # GH 3311

pandas/tests/frame/test_analytics.py

@@ -23,6 +23,9 @@
 import pandas._testing as tm
 import pandas.core.algorithms as algorithms
 import pandas.core.nanops as nanops
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 
 def assert_stat_op_calc(
@@ -790,6 +793,7 @@ def test_sum_prod_nanops(self, method, unit):
         expected = pd.Series(result, index=["A", "B"])
         tm.assert_series_equal(result, expected)
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
     def test_sum_nanops_timedelta(self):
         # prod isn't defined on timedeltas
         idx = ["a", "b", "c"]

pandas/tests/indexes/datetimes/test_datetime.py

@@ -7,6 +7,9 @@
 import pandas as pd
 from pandas import DataFrame, DatetimeIndex, Index, Timestamp, date_range, offsets
 import pandas._testing as tm
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 randn = np.random.randn
 
@@ -63,6 +66,7 @@ def test_time_overflow_for_32bit_machines(self):
         idx2 = pd.date_range(end="2000", periods=periods, freq="S")
         assert len(idx2) == periods
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
     def test_nat(self):
         assert DatetimeIndex([np.nan])[0] is pd.NaT
 

pandas/tests/reductions/test_reductions.py

@@ -23,6 +23,9 @@
 )
 import pandas._testing as tm
 from pandas.core import nanops
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 
 def get_objs():
@@ -1142,6 +1145,7 @@ def test_mode_mixeddtype(self, dropna, expected1, expected2):
         expected = Series(expected2, dtype=object)
         tm.assert_series_equal(result, expected)
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
     @pytest.mark.parametrize(
         "dropna, expected1, expected2",
         [

pandas/tests/series/test_constructors.py

@@ -28,7 +28,9 @@
 )
 import pandas._testing as tm
 from pandas.core.arrays import IntervalArray, period_array
-
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 class TestSeriesConstructors:
     @pytest.mark.parametrize(
@@ -960,6 +962,7 @@ def test_construction_to_datetimelike_unit(self, arr_dtype, dtype, unit):
 
         tm.assert_series_equal(result, expected)
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
     @pytest.mark.parametrize("arg", ["2013-01-01 00:00:00", pd.NaT, np.nan, None])
     def test_constructor_with_naive_string_and_datetimetz_dtype(self, arg):
         # GH 17415: With naive string
@@ -1272,6 +1275,7 @@ def test_NaT_scalar(self):
         series[2] = val
         assert isna(series[2])
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
     def test_NaT_cast(self):
         # GH10747
         result = Series([np.nan]).astype("M8[ns]")

pandas/tests/test_algos.py

@@ -35,6 +35,9 @@
 import pandas.core.algorithms as algos
 from pandas.core.arrays import DatetimeArray
 import pandas.core.common as com
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 
 class TestFactorize:
@@ -1046,6 +1049,7 @@ def test_dropna(self):
             expected = Series([2, 1, 1], index=[5.0, 10.3, np.nan])
             tm.assert_series_equal(result, expected)
 
+    @pytest.mark.xfail(condition=not is_nannat_working,reason="https://bugs.debian.org/877754",strict=False)
     def test_value_counts_normalized(self):
         # GH12558
         s = Series([1, 2, np.nan, np.nan, np.nan])