Opt out of floor division for float dtype time encoding

doc/whats-new.rst

@@ -46,7 +46,10 @@ Bug fixes
 - Make illegal path-like variable names when constructing a DataTree from a Dataset
   (:issue:`9339`, :pull:`9378`)
   By `Etienne Schalk <https://github.com/etienneschalk>`_.
-
+- Fix bug when encoding times with missing values as floats in the case when
+  the non-missing times could in theory be encoded with integers
+  (:issue:`9488`, :pull:`9497`). By `Spencer Clark
+  <https://github.com/spencerkclark>`_.
 
 
 Documentation

xarray/coding/times.py

@@ -771,7 +771,7 @@ def _eagerly_encode_cf_datetime(
         # needed time delta to encode faithfully to int64
         needed_time_delta = _time_units_to_timedelta64(needed_units)
 
-        floor_division = True
+        floor_division = np.issubdtype(dtype, np.integer) or dtype is None
         if time_delta > needed_time_delta:
             floor_division = False
             if dtype is None:
@@ -892,7 +892,7 @@ def _eagerly_encode_cf_timedelta(
     # needed time delta to encode faithfully to int64
     needed_time_delta = _time_units_to_timedelta64(needed_units)
 
-    floor_division = True
+    floor_division = np.issubdtype(dtype, np.integer) or dtype is None
     if time_delta > needed_time_delta:
         floor_division = False
         if dtype is None:

xarray/tests/test_coding_times.py

@@ -1383,24 +1383,46 @@ def test_roundtrip_timedelta64_nanosecond_precision_warning() -> None:
     assert decoded_var.encoding["dtype"] == np.int64
 
 
-def test_roundtrip_float_times() -> None:
-    # Regression test for GitHub issue #8271
-    fill_value = 20.0
-    times = [
-        np.datetime64("1970-01-01 00:00:00", "ns"),
-        np.datetime64("1970-01-01 06:00:00", "ns"),
-        np.datetime64("NaT", "ns"),
-    ]
+_TEST_ROUNDTRIP_FLOAT_TIMES_TESTS = {
+    "GH-8271": (
+        20.0,
+        np.array(
+            ["1970-01-01 00:00:00", "1970-01-01 06:00:00", "NaT"],
+            dtype="datetime64[ns]",
+        ),
+        "days since 1960-01-01",
+        np.array([3653, 3653.25, 20.0]),
+    ),
+    "GH-9488-datetime64[ns]": (
+        1.0e20,
+        np.array(["2010-01-01 12:00:00", "NaT"], dtype="datetime64[ns]"),
+        "seconds since 2010-01-01",
+        np.array([43200, 1.0e20]),
+    ),
+    "GH-9488-timedelta64[ns]": (
+        1.0e20,
+        np.array([1_000_000_000, "NaT"], dtype="timedelta64[ns]"),
+        "seconds",
+        np.array([1.0, 1.0e20]),
+    ),
+}
+
 
-    units = "days since 1960-01-01"
+@pytest.mark.parametrize(
+    ("fill_value", "times", "units", "encoded_values"),
+    _TEST_ROUNDTRIP_FLOAT_TIMES_TESTS.values(),
+    ids=_TEST_ROUNDTRIP_FLOAT_TIMES_TESTS.keys(),
+)
+def test_roundtrip_float_times(fill_value, times, units, encoded_values) -> None:
+    # Regression test for GitHub issues #8271 and #9488
     var = Variable(
         ["time"],
         times,
         encoding=dict(dtype=np.float64, _FillValue=fill_value, units=units),
     )
 
     encoded_var = conventions.encode_cf_variable(var)
-    np.testing.assert_array_equal(encoded_var, np.array([3653, 3653.25, 20.0]))
+    np.testing.assert_array_equal(encoded_var, encoded_values)
     assert encoded_var.attrs["units"] == units
     assert encoded_var.attrs["_FillValue"] == fill_value