fix!: use pandas.NaT for missing values in dbdate and dbtime dtypes (…

…#67)

* fix!: use `pandas.NaT` for missing values in dbdate and dbtime dtypes

This makes them consistent with other date/time dtypes, as well as internally
consistent with the advertised `dtype.na_value`.

* adjust pandas version support for median

BREAKING-CHANGE: dbdate and dbtime dtypes return NaT instead of None for missing values

Release-As: 0.4.0

db_dtypes/__init__.py

@@ -145,8 +145,8 @@ def _datetime(
             raise TypeError("Invalid value type", scalar)
 
     def _box_func(self, x):
-        if pandas.isnull(x):
-            return None
+        if pandas.isna(x):
+            return pandas.NaT
 
         try:
             return x.astype("<M8[us]").astype(datetime.datetime).time()
@@ -250,8 +250,8 @@ def _datetime(
             raise TypeError("Invalid value type", scalar)
 
     def _box_func(self, x):
-        if pandas.isnull(x):
-            return None
+        if pandas.isna(x):
+            return pandas.NaT
         try:
             return x.astype("<M8[us]").astype(datetime.datetime).date()
         except AttributeError:

db_dtypes/core.py

@@ -16,7 +16,6 @@
 
 import numpy
 import pandas
-from pandas import NaT
 import pandas.api.extensions
 from pandas.api.types import is_dtype_equal, is_list_like, pandas_dtype
 
@@ -27,8 +26,8 @@
 
 
 class BaseDatetimeDtype(pandas.api.extensions.ExtensionDtype):
-    na_value = NaT
-    kind = "o"
+    na_value = pandas.NaT
+    kind = "O"
     names = None
 
     @classmethod

db_dtypes/pandas_backports.py

@@ -42,7 +42,7 @@
 numpy_validate_max = pandas.compat.numpy.function.validate_max
 numpy_validate_min = pandas.compat.numpy.function.validate_min
 
-if pandas_release >= (1, 2):
+if pandas_release >= (1, 3):
     nanmedian = pandas.core.nanops.nanmedian
     numpy_validate_median = pandas.compat.numpy.function.validate_median
 

testing/constraints-3.9.txt

@@ -1 +1,2 @@
-sqlalchemy>=1.4.13
+# Make sure we test with pandas 1.3.0. The Python version isn't that relevant.
+pandas==1.3.0

tests/unit/test_date.py

@@ -19,6 +19,7 @@
 
 # To register the types.
 import db_dtypes  # noqa
+from db_dtypes import pandas_backports
 
 
 @pytest.mark.parametrize(
@@ -65,3 +66,29 @@ def test_date_parsing(value, expected):
 def test_date_parsing_errors(value, error):
     with pytest.raises(ValueError, match=error):
         pandas.Series([value], dtype="dbdate")
+
+
+@pytest.mark.skipif(
+    not hasattr(pandas_backports, "numpy_validate_median"),
+    reason="median not available with this version of pandas",
+)
+@pytest.mark.parametrize(
+    "values, expected",
+    [
+        (["1970-01-01", "1900-01-01", "2000-01-01"], datetime.date(1970, 1, 1)),
+        (
+            [
+                None,
+                "1900-01-01",
+                pandas.NA if hasattr(pandas, "NA") else None,
+                pandas.NaT,
+                float("nan"),
+            ],
+            datetime.date(1900, 1, 1),
+        ),
+        (["2222-02-01", "2222-02-03"], datetime.date(2222, 2, 2)),
+    ],
+)
+def test_date_median(values, expected):
+    series = pandas.Series(values, dtype="dbdate")
+    assert series.median() == expected