Fix wrong save of datetime64[s] in HDFStore

Author: chaoyihu

pandas/io/pytables.py

@@ -3036,7 +3036,10 @@ def read_array(self, key: str, start: int | None = None, stop: int | None = None
             if dtype and dtype.startswith("datetime64"):
                 # reconstruct a timezone if indicated
                 tz = getattr(attrs, "tz", None)
-                ret = _set_tz(ret, tz)
+                if dtype == "datetime64[s]":
+                    ret = _set_tz(ret, tz, unit="s")
+                else:
+                    ret = _set_tz(ret, tz)  # default unit is ns
 
             elif dtype == "timedelta64":
                 ret = np.asarray(ret, dtype="m8[ns]")
@@ -4964,19 +4967,22 @@ def _get_tz(tz: tzinfo) -> str | tzinfo:
     return zone
 
 
-def _set_tz(values: npt.NDArray[np.int64], tz: str | tzinfo | None) -> DatetimeArray:
+def _set_tz(
+    values: npt.NDArray[np.int64], tz: str | tzinfo | None, unit: str = "ns"
+) -> DatetimeArray:
     """
     Coerce the values to a DatetimeArray with appropriate tz.
 
     Parameters
     ----------
     values : ndarray[int64]
     tz : str, tzinfo, or None
+    unit : str. The default unit is ns. Needs to be specified otherwise.
     """
     assert values.dtype == "i8", values.dtype
     # Argument "tz" to "tz_to_dtype" has incompatible type "str | tzinfo | None";
     # expected "tzinfo"
-    dtype = tz_to_dtype(tz=tz, unit="ns")  # type: ignore[arg-type]
+    dtype = tz_to_dtype(tz=tz, unit=unit)  # type: ignore[arg-type]
     dta = DatetimeArray._from_sequence(values, dtype=dtype)
     return dta
 

pandas/tests/io/pytables/test_read.py

@@ -317,3 +317,14 @@ def test_read_infer_string(tmp_path, setup_path):
         columns=Index(["a"], dtype="string[pyarrow_numpy]"),
     )
     tm.assert_frame_equal(result, expected)
+
+
+def test_hdfstore_read_datetime64_unit_s():
+    # Fix issue 59004: HDFStore doesn't save datetime64[s] right
+    df_s = DataFrame(["2001-01-01", "2002-02-02"], dtype="datetime64[s]")
+    with HDFStore("deleteme.h5", mode="w") as store:
+        store.put("df_s", df_s)
+    with HDFStore("deleteme.h5", mode="r") as store:
+        df_fromstore = store.get("df_s")
+    tm.assert_frame_equal(df_s, df_fromstore)
+    Path("deleteme.h5").unlink()  # Delete created file