Warn when unit keyword is ignored (#62841)

Author: jbrockmendel

pandas/_libs/tslibs/timedeltas.pyx

@@ -2026,6 +2026,19 @@ class Timedelta(_Timedelta):
                 "milliseconds, microseconds, nanoseconds]"
             )
 
+        if (
+            unit is not None
+            and not (is_float_object(value) or is_integer_object(value))
+        ):
+            # GH#53198
+            warnings.warn(
+                "The 'unit' keyword is only used when the Timedelta input is "
+                f"an integer or float, not {type(value).__name__}. "
+                "To specify the storage unit of the output use `td.as_unit(unit)`",
+                UserWarning,
+                stacklevel=find_stack_level(),
+            )
+
         if value is _no_input:
             if not len(kwargs):
                 raise ValueError("cannot construct a Timedelta without a "

pandas/_libs/tslibs/timestamps.pyx

@@ -67,6 +67,7 @@ from pandas._libs.tslibs.dtypes cimport (
 )
 from pandas._libs.tslibs.util cimport (
     is_array,
+    is_float_object,
     is_integer_object,
 )
 
@@ -2654,6 +2655,19 @@ class Timestamp(_Timestamp):
             if hasattr(ts_input, "fold"):
                 ts_input = ts_input.replace(fold=fold)
 
+        if (
+            unit is not None
+            and not (is_float_object(ts_input) or is_integer_object(ts_input))
+        ):
+            # GH#53198
+            warnings.warn(
+                "The 'unit' keyword is only used when the Timestamp input is "
+                f"an integer or float, not {type(ts_input).__name__}. "
+                "To specify the storage unit of the output use `ts.as_unit(unit)`",
+                UserWarning,
+                stacklevel=find_stack_level(),
+            )
+
         # GH 30543 if pd.Timestamp already passed, return it
         # check that only ts_input is passed
         # checking verbosely, because cython doesn't optimize

pandas/core/computation/pytables.py

@@ -17,6 +17,7 @@
 
 import numpy as np
 
+from pandas._libs import lib
 from pandas._libs.tslibs import (
     Timedelta,
     Timestamp,
@@ -227,8 +228,10 @@ def stringify(value):
         elif kind in ("timedelta64", "timedelta"):
             if isinstance(conv_val, str):
                 conv_val = Timedelta(conv_val)
-            else:
+            elif lib.is_integer(conv_val) or lib.is_float(conv_val):
                 conv_val = Timedelta(conv_val, unit="s")
+            else:
+                conv_val = Timedelta(conv_val)
             conv_val = conv_val.as_unit("ns")._value
             return TermValue(int(conv_val), conv_val, kind)
         elif meta == "category":

pandas/core/resample.py

@@ -2642,7 +2642,7 @@ def _adjust_bin_edges(
                 edges_dti = binner.tz_localize(None)
                 edges_dti = (
                     edges_dti
-                    + Timedelta(days=1, unit=edges_dti.unit).as_unit(edges_dti.unit)
+                    + Timedelta(days=1).as_unit(edges_dti.unit)
                     - Timedelta(1, unit=edges_dti.unit).as_unit(edges_dti.unit)
                 )
                 bin_edges = edges_dti.tz_localize(binner.tz).asi8

pandas/core/tools/datetimes.py

@@ -599,7 +599,10 @@ def _adjust_to_origin(arg, origin, unit):
 
         # we are going to offset back to unix / epoch time
         try:
-            offset = Timestamp(origin, unit=unit)
+            if lib.is_integer(origin) or lib.is_float(origin):
+                offset = Timestamp(origin, unit=unit)
+            else:
+                offset = Timestamp(origin)
         except OutOfBoundsDatetime as err:
             raise OutOfBoundsDatetime(f"origin {origin} is Out of Bounds") from err
         except ValueError as err:

pandas/tests/scalar/timedelta/test_constructors.py

@@ -193,11 +193,15 @@ def test_construct_from_td64_with_unit():
     #  results, and in non-overflow cases is irrelevant GH#46827
     obj = np.timedelta64(123456789000000000, "h")
 
+    msg = "The 'unit' keyword is only used when the Timedelta input is"
+
     with pytest.raises(OutOfBoundsTimedelta, match="123456789000000000 hours"):
-        Timedelta(obj, unit="ps")
+        with tm.assert_produces_warning(UserWarning, match=msg):
+            Timedelta(obj, unit="ps")
 
     with pytest.raises(OutOfBoundsTimedelta, match="123456789000000000 hours"):
-        Timedelta(obj, unit="ns")
+        with tm.assert_produces_warning(UserWarning, match=msg):
+            Timedelta(obj, unit="ns")
 
     with pytest.raises(OutOfBoundsTimedelta, match="123456789000000000 hours"):
         Timedelta(obj)
@@ -633,6 +637,9 @@ def test_timedelta_pass_td_and_kwargs_raises():
         Timedelta(td, days=2)
 
 
+@pytest.mark.filterwarnings(
+    "ignore:The 'unit' keyword is only used when the Timedelta input:UserWarning"
+)
 @pytest.mark.parametrize(
     "constructor, value, unit",
     [

pandas/tests/scalar/timedelta/test_timedelta.py

@@ -83,7 +83,7 @@ def test_mul_preserves_reso(self, td, unit):
 
     def test_cmp_cross_reso(self, td):
         # numpy gets this wrong because of silent overflow
-        other = Timedelta(days=106751, unit="ns")
+        other = Timedelta(days=106751)
         assert other < td
         assert td > other
         assert not other == td

pandas/tests/scalar/timestamp/test_constructors.py

@@ -1076,7 +1076,9 @@ def test_timestamp_nano_range(nano):
 
 def test_non_nano_value():
     # https://github.com/pandas-dev/pandas/issues/49076
-    result = Timestamp("1800-01-01", unit="s").value
+    msg = "The 'unit' keyword is only used when"
+    with tm.assert_produces_warning(UserWarning, match=msg):
+        result = Timestamp("1800-01-01", unit="s").value
     # `.value` shows nanoseconds, even though unit is 's'
     assert result == -5364662400000000000