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)

On riscv64 one test case raises an exception
(though I suspect not the general case since there aren't more).
 
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

@@ -100,6 +100,9 @@
         TimedeltaArray,
     )
 
+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
 _int8_max = np.iinfo(np.int8).max
 _int16_max = np.iinfo(np.int16).max
 _int32_max = np.iinfo(np.int32).max
@@ -1195,6 +1198,8 @@ def astype_nansafe(
             f"'{dtype.name}[ns]' instead."
         )
         raise ValueError(msg)
+    if warn_nannat_platform and (is_datetime64_dtype(dtype) or is_timedelta64_dtype(dtype)) and not (is_datetime64_dtype(arr.dtype) or is_timedelta64_dtype(arr.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.dtype) or is_object_dtype(dtype):
         # Explicit copy, or required since NumPy can't view from / to object.
@@ -1625,6 +1630,8 @@ def maybe_cast_to_datetime(
             dtype = ensure_nanosecond_dtype(dtype)
 
             value = np.array(value, copy=False)
+            if warn_nannat_platform and not (is_datetime64_dtype(value.dtype) or is_timedelta64_dtype(value.dtype)) and np.issubdtype(value.dtype, np.floating) and not np.isfinite(value).all():
+                warnings.warn(warn_nannat_platform)
 
             # we have an array of datetime or timedeltas & nulls
             if value.size or not is_dtype_equal(value.dtype, dtype):

pandas/tests/dtypes/cast/test_downcast.py

@@ -7,6 +7,9 @@
 
 from pandas import Series
 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

@@ -16,6 +16,9 @@
     isna,
 )
 import pandas._testing as tm
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 
 @pytest.fixture(params=["default", "float_string", "mixed_float", "mixed_int"])
@@ -360,6 +363,7 @@ 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_reductions.py

@@ -28,6 +28,9 @@
 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(
@@ -845,6 +848,7 @@ def test_sum_prod_nanops(self, method, unit, numeric_only):
         expected = 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

@@ -14,6 +14,9 @@
     offsets,
 )
 import pandas._testing as tm
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 
 class TestDatetimeIndex:
@@ -55,6 +58,7 @@ def test_time_overflow_for_32bit_machines(self):
         idx2 = 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

@@ -27,6 +27,9 @@
 )
 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():
@@ -1346,6 +1349,7 @@ 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

@@ -47,6 +47,9 @@
 )
 from pandas.core.internals.blocks import NumericBlock
 
+import platform
+import re
+is_nannat_working=bool(re.match('i.?86|x86|s390|ppc',platform.uname()[4]))
 
 class TestSeriesConstructors:
     @pytest.mark.parametrize(
@@ -1057,6 +1060,7 @@ 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", NaT, np.nan, None])
     def test_constructor_with_naive_string_and_datetimetz_dtype(self, arg):
         # GH 17415: With naive string
@@ -1434,6 +1438,7 @@ 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

@@ -45,6 +45,9 @@
 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:
@@ -1265,6 +1268,7 @@ def test_dropna(self):
         expected = Series([3, 2, 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 + [2] * 3 + [np.nan] * 5)

pandas/tests/tools/test_to_datetime.py

@@ -25,6 +25,9 @@
     OutOfBoundsTimedelta,
 )
 import pandas.util._test_decorators as td
+import platform
+import re
+is_platform_x86 = bool(re.match('i.?86|x86',platform.uname()[4]))
 
 from pandas.core.dtypes.common import is_datetime64_ns_dtype
 
@@ -1484,6 +1487,7 @@ def test_to_datetime_errors_ignore_utc_true(self):
         tm.assert_index_equal(result, expected)
 
     # TODO: this is moved from tests.series.test_timeseries, may be redundant
+    @pytest.mark.xfail(not is_platform_x86, strict=False, raises=OutOfBoundsDatetime, reason="fails on riscv64")
     def test_to_datetime_unit(self):
 
         epoch = 1370745748