fixup time zone handling

Author: MarcoGorelli

packages/python/plotly/_plotly_utils/basevalidators.py

@@ -95,44 +95,21 @@ def copy_to_readonly_numpy_array(v, kind=None, force_numeric=False):
     }
 
     if isinstance(v, nw.Series):
-        if nw.dependencies.is_pandas_like_series(v_native := v.to_native()):
-            v = v_native
+        if v.dtype == nw.Datetime and v.dtype.time_zone is not None:
+            # Remove time zone so that local time is displayed
+            v = v.dt.replace_time_zone(None).to_numpy()
         else:
             v = v.to_numpy()
     elif isinstance(v, nw.DataFrame):
-        if nw.dependencies.is_pandas_like_dataframe(v_native := v.to_native()):
-            v = v_native
-        else:
-            v = v.to_numpy()
-
-    if pd and isinstance(v, (pd.Series, pd.Index)):
-        # Handle pandas Series and Index objects
-        if v.dtype.kind in numeric_kinds:
-            # Get the numeric numpy array so we use fast path below
-            v = v.values
-        elif v.dtype.kind == "M":
-            # Convert datetime Series/Index to numpy array of datetimes
-            if isinstance(v, pd.Series):
-                with warnings.catch_warnings():
-                    warnings.simplefilter("ignore", FutureWarning)
-                    # Series.dt.to_pydatetime will return Index[object]
-                    # https://github.com/pandas-dev/pandas/pull/52459
-                    v = np.array(v.dt.to_pydatetime())
-            else:
-                # DatetimeIndex
-                v = v.to_pydatetime()
-    elif pd and isinstance(v, pd.DataFrame) and len(set(v.dtypes)) == 1:
-        dtype = v.dtypes.tolist()[0]
-        if dtype.kind in numeric_kinds:
-            v = v.values
-        elif dtype.kind == "M":
-            with warnings.catch_warnings():
-                warnings.simplefilter("ignore", FutureWarning)
-                # Series.dt.to_pydatetime will return Index[object]
-                # https://github.com/pandas-dev/pandas/pull/52459
-                v = [
-                    np.array(row.dt.to_pydatetime()).tolist() for i, row in v.iterrows()
-                ]
+        schema = v.schema
+        overrides = {}
+        for key, val in schema.items():
+            if val == nw.Datetime and val.time_zone is not None:
+                # Remove time zone so that local time is displayed
+                overrides[key] = nw.col(key).dt.replace_time_zone(None)
+        if overrides:
+            v = v.with_columns(**overrides)
+        v = v.to_numpy()
 
     if not isinstance(v, np.ndarray):
         # v has its own logic on how to convert itself into a numpy array

packages/python/plotly/optional-requirements.txt

@@ -39,7 +39,7 @@ ipython
 
 ## pandas deps for some matplotlib functionality ##
 pandas
-narwhals>=1.12.0
+narwhals>=1.13.1
 
 ## scipy deps for some FigureFactory functions ##
 scipy

packages/python/plotly/plotly/express/_core.py

@@ -413,12 +413,20 @@ def make_trace_kwargs(args, trace_spec, trace_data, mapping_labels, sizeref):
                     # i.e. we can't do resampling, because then the X values might not line up!
                     non_missing = ~(x.is_null() | y.is_null())
                     trace_patch["x"] = (
-                        sorted_trace_data.filter(non_missing)
-                        .get_column(args["x"])
-                        .to_numpy()
+                        sorted_trace_data.filter(non_missing).get_column(args["x"])
                         # FIXME: Converting to numpy is needed to pass `test_trendline_on_timeseries`
                         # test, but I wonder if it is the right way to do it in the first place.
                     )
+                    if (
+                        trace_patch["x"].dtype == nw.Datetime
+                        and trace_patch["x"].dtype.time_zone is not None
+                    ):
+                        # Remove time zone so that local time is displayed
+                        trace_patch["x"] = (
+                            trace_patch["x"].dt.replace_time_zone(None).to_numpy()
+                        )
+                    else:
+                        trace_patch["x"] = trace_patch["x"].to_numpy()
 
                     trendline_function = trendline_functions[attr_value]
                     y_out, hover_header, fit_results = trendline_function(

packages/python/plotly/plotly/tests/test_optional/test_px/test_px_hover.py

@@ -190,14 +190,12 @@ def test_sunburst_hoverdict_color(constructor):
 
 
 def test_date_in_hover(request, constructor):
-    if "pyarrow_table" in str(constructor) or "polars_eager" in str(constructor):
-        # fig.data[0].customdata[0][0] is a numpy.datetime64 for non pandas
-        # input, and it does not keep the timezone when converting to py scalar
-        request.applymarker(pytest.mark.xfail)
-
     df = nw.from_native(
         constructor({"date": ["2015-04-04 19:31:30+01:00"], "value": [3]})
     ).with_columns(date=nw.col("date").str.to_datetime(format="%Y-%m-%d %H:%M:%S%z"))
     fig = px.scatter(df.to_native(), x="value", y="value", hover_data=["date"])
 
-    assert fig.data[0].customdata[0][0] == df.item(row=0, column="date")
+    # Check that what gets displayed is the local datetime
+    assert nw.to_py_scalar(fig.data[0].customdata[0][0]) == nw.to_py_scalar(
+        df.item(row=0, column="date")
+    ).replace(tzinfo=None)

packages/python/plotly/plotly/tests/test_optional/test_utils/test_utils.py

@@ -270,7 +270,7 @@ def test_encode_customdata_datetime_series(self):
         )
         self.assertTrue(
             fig_json.startswith(
-                '{"data":[{"customdata":["2010-01-01T00:00:00","2010-01-02T00:00:00"]'
+                '{"data":[{"customdata":["2010-01-01T00:00:00.000000000","2010-01-02T00:00:00.000000000"]'
             )
         )
 
@@ -292,8 +292,8 @@ def test_encode_customdata_datetime_homogeneous_dataframe(self):
         self.assertTrue(
             fig_json.startswith(
                 '{"data":[{"customdata":'
-                '[["2010-01-01T00:00:00","2011-01-01T00:00:00"],'
-                '["2010-01-02T00:00:00","2011-01-02T00:00:00"]'
+                '[["2010-01-01T00:00:00.000000000","2011-01-01T00:00:00.000000000"],'
+                '["2010-01-02T00:00:00.000000000","2011-01-02T00:00:00.000000000"]'
             )
         )
 

packages/python/plotly/requirements.txt

@@ -6,4 +6,4 @@
 ###################################################
 
 ## dataframe agnostic layer ##
-narwhals>=1.12.0
+narwhals>=1.13.1

packages/python/plotly/setup.py

@@ -603,7 +603,7 @@ def run(self):
     data_files=[
         ("etc/jupyter/nbconfig/notebook.d", ["jupyterlab-plotly.json"]),
     ],
-    install_requires=["narwhals>=1.12.0", "packaging"],
+    install_requires=["narwhals>=1.13.1", "packaging"],
     zip_safe=False,
     cmdclass=dict(
         build_py=js_prerelease(versioneer_cmds["build_py"]),

packages/python/plotly/test_requirements/requirements_310_core.txt

@@ -1,3 +1,3 @@
 requests==2.25.1
 pytest==7.4.4
-narwhals>=1.12.0
+narwhals>=1.13.1

packages/python/plotly/test_requirements/requirements_310_optional.txt

@@ -21,4 +21,4 @@ kaleido
 orjson==3.8.12
 polars[timezone]
 pyarrow
-narwhals>=1.12.0
+narwhals>=1.13.1

packages/python/plotly/test_requirements/requirements_311_core.txt

@@ -1,3 +1,3 @@
 requests==2.25.1
 pytest==7.4.4
-narwhals>=1.12.0
+narwhals>=1.13.1